dvc: bump to 2.6.3

Replacing DVC_ANALYTICS. This is following our other env conventions and
will not have effect if set to false/no/etc, and will only work if set
to true/yes/y/etc. Will disable analytics even if it is explicitly
enabled in our configs.

Kudos @skshetry

.codecov.yml

@@ -3,7 +3,7 @@ coverage:
   status:
     project:
       default:
-        target: 85
+        target: 90
         threshold: 10%
     patch:
       default:

.github/CODEOWNERS

@@ -0,0 +1 @@
+*       @iterative/dvc

.github/workflows/packages.yaml

@@ -7,7 +7,7 @@ on:
     types: [released, prereleased]
 jobs:
   deb:
-    runs-on: ubuntu-16.04
+    runs-on: ubuntu-18.04
     steps:
     - uses: actions/[email protected]
     - name: Set up Python 3.8
@@ -36,7 +36,7 @@ jobs:
         asset_name: dvc_${{ github.event.release.tag_name }}_amd64.deb
         asset_content_type: binary/octet-stream
   rpm:
-    runs-on: ubuntu-16.04
+    runs-on: ubuntu-18.04
     steps:
     - uses: actions/[email protected]
     - name: Set up Python 3.8
@@ -125,7 +125,7 @@ jobs:
         asset_name: dvc-${{ github.event.release.tag_name }}.exe
         asset_content_type: binary/octet-stream
   pip:
-    runs-on: ubuntu-16.04
+    runs-on: ubuntu-18.04
     steps:
     - uses: actions/[email protected]
     - name: Set up Python 3.7

.github/workflows/tests.yaml

@@ -42,13 +42,14 @@ jobs:
       matrix:
         os: [ubuntu-18.04, windows-2019, macos-10.15]
         pyv: ["3.6", "3.7", "3.8", "3.9"]
-# macos 11.0 workers take too long to start up
-#        include:
-#          - os: "macos-11.0"
-#            pyv: "3.9"
+        include:
+          - os: "macos-10.15"
+            pyv: "3.10.0-rc.1"
         exclude:
           - os: macos-10.15
             pyv: "3.9"
+          - os: macos-10.15
+            pyv: "3.7"
 
     steps:
     - uses: actions/[email protected]
@@ -68,10 +69,14 @@ jobs:
         key: ${{ runner.os }}-pip-${{ hashFiles('setup.py') }}
         restore-keys: |
           ${{ runner.os }}-pip-
+    - name: install libgit2 on Python3.10 in macOS
+      if: matrix.pyv == '3.10.0-rc.1' && matrix.os == 'macos-10.15'
+      run: |
+        brew install libgit2
     - name: install
       if: steps.cache.pip-cache-dir.cache-hit != 'true'
       run: |
-        pip install wheel
+        pip install --upgrade pip setuptools wheel
         pip install -e ".[all,tests]"
     - name: install temporary dependencies
       run: |
@@ -100,12 +105,15 @@ jobs:
         GDRIVE_CREDENTIALS_DATA: ${{ secrets.GDRIVE_CREDENTIALS_DATA }}
         AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
         AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
+        OSS_ACCESS_KEY_ID: ${{ secrets.OSS_ACCESS_KEY_ID}}
+        OSS_ACCESS_KEY_SECRET: ${{ secrets.OSS_ACCESS_KEY_SECRET}}
+        OSS_ENDPOINT: ${{ secrets.OSS_ENDPOINT}}
       run: >-
         python -m tests -n=4
         --cov-report=xml --cov-report=term
         --all ${{ env.extra_test_args }}
     - name: upload coverage report
-      uses: codecov/codecov-action@v1.5.0
+      uses: codecov/codecov-action@v2.0.2
       with:
         file: ./coverage.xml
         fail_ci_if_error: false
@@ -117,7 +125,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
     - name: Slack Notification
-      uses: rtCamp/action-slack-notify@v2.1.3
+      uses: rtCamp/action-slack-notify@v2.2.0
       env:
         SLACK_WEBHOOK: ${{ secrets.SLACK_WEBHOOK }}
         SLACK_COLOR: ${{ job.status }}

.pre-commit-config.yaml

@@ -1,14 +1,24 @@
+ci:
+  skip: [pylint, dvc-pre-commit]
+
 repos:
   - hooks:
       - id: black
         language_version: python3
     repo: https://github.com/ambv/black
-    rev: 21.5b1
+    rev: 21.7b0
+  - hooks:
+      - id: codespell
+        args: 
+          - --ignore-words-list
+          - ba,datas,fo,uptodate
+    repo: https://github.com/codespell-project/codespell
+    rev: v2.1.0
   - hooks:
       - id: isort
         language_version: python3
     repo: https://github.com/timothycrosley/isort
-    rev: 5.8.0
+    rev: 5.9.3
   - hooks:
       - id: flake8
         language_version: python3
@@ -23,8 +33,13 @@ repos:
       -  id: mypy
          language_version: python3
          files: ^dvc/
+         additional_dependencies:
+           - types-requests
+           - types-paramiko
+           - types-tabulate
+           - types-toml
     repo: https://github.com/pre-commit/mirrors-mypy
-    rev: 'v0.812'
+    rev: 'v0.910'
   - repo: local
     hooks:
     - id: pylint

README.rst

@@ -29,9 +29,9 @@ learning projects. Key features:
 #. Helps manage experiments with Git tags/branches and **metrics** tracking.
 
 **DVC** aims to replace spreadsheet and document sharing tools (such as Excel or Google Docs)
-which are being used frequently as both knowledge repositories and team ledgers.
-DVC also replaces both ad-hoc scripts to track, move, and deploy different model versions;
-as well as ad-hoc data file suffixes and prefixes.
+frequently used as both knowledge repositories and team ledgers.
+DVC also replaces both ad-hoc scripts to track, move, and deploy different model versions
+and ad-hoc data file suffixes and prefixes.
 
 .. contents:: **Contents**
   :backlinks: none
@@ -184,7 +184,7 @@ Comparison to related technologies
 
 #. *Data Engineering* tools such as `AirFlow <https://airflow.apache.org/>`_, `Luigi <https://github.com/spotify/luigi>`_ and
    others - in DVC data, model and ML pipelines represent a single ML project and focuses on data scientist' experience while
-   data engineering pipeline orchestrates multiple data projects and focuses on efficent execution. DVC project can be used
+   data engineering pipeline orchestrates multiple data projects and focuses on efficient execution. DVC project can be used
    from data pipeline as a single execution step. `AirFlow DVC <https://github.com/covid-genomics/airflow-dvc>`_ is an example
    of such integration.
 
@@ -204,6 +204,10 @@ Comparison to related technologies
 #. `DAGsHub <https://dagshub.com/>`_ - online service of Git+DVC repositories with pipeline and metrics visualization and
    DVC-specific cloud storage.
 
+#. `DVC Studio <https://studio.iterative.ai/>`_ - online service of DVC repositories visualisation from DVC team. Also,
+   integrated with `CML (CI/CD for ML) <https://cml.dev/>`_ for training models in clouds and Kubernetes.
+
+
 Contributing
 ============
 

dvc/analytics.py

@@ -2,6 +2,8 @@ import json
 import logging
 import os
 
+from .env import DVC_NO_ANALYTICS
+
 logger = logging.getLogger(__name__)
 
 
@@ -32,7 +34,7 @@ def collect_and_send_report(args=None, return_code=None):
 
     with tempfile.NamedTemporaryFile(delete=False, mode="w") as fobj:
         json.dump(report, fobj)
-        daemon(["analytics", fobj.name])
+    daemon(["analytics", fobj.name])
 
 
 def is_enabled():
@@ -42,9 +44,12 @@ def is_enabled():
     if env2bool("DVC_TEST"):
         return False
 
-    enabled = to_bool(
-        Config(validate=False).get("core", {}).get("analytics", "true")
-    )
+    enabled = not os.getenv(DVC_NO_ANALYTICS)
+    if enabled:
+        enabled = to_bool(
+            Config(validate=False).get("core", {}).get("analytics", "true")
+        )
+
     logger.debug("Analytics is {}abled.".format("en" if enabled else "dis"))
 
     return enabled

dvc/checkout.py

@@ -11,17 +11,17 @@ from dvc.exceptions import (
 )
 from dvc.ignore import DvcIgnoreFilter
 from dvc.objects import check, load
-from dvc.objects.errors import ObjectFormatError
-from dvc.objects.stage import stage
-from dvc.remote.slow_link_detection import (  # type: ignore[attr-defined]
+from dvc.objects.db.slow_link_detection import (  # type: ignore[attr-defined]
     slow_link_guard,
 )
+from dvc.objects.errors import ObjectFormatError
+from dvc.objects.stage import stage
 from dvc.types import Optional
 
 logger = logging.getLogger(__name__)
 
 
-def _changed(path_info, fs, obj, cache):
+def _changed(path_info, fs, obj, cache, state=None):
     logger.trace("checking if '%s'('%s') has changed.", path_info, obj)
 
     try:
@@ -32,11 +32,15 @@ def _changed(path_info, fs, obj, cache):
         )
         return True
 
-    try:
-        actual = stage(cache, path_info, fs, obj.hash_info.name).hash_info
-    except FileNotFoundError:
-        logger.debug("'%s' doesn't exist.", path_info)
-        return True
+    actual = state.get(path_info, fs) if state else None
+    if actual is None:
+        try:
+            actual = stage(
+                cache, path_info, fs, obj.hash_info.name, dry_run=True
+            )[1].hash_info
+        except FileNotFoundError:
+            logger.debug("'%s' doesn't exist.", path_info)
+            return True
 
     if obj.hash_info != actual:
         logger.debug(
@@ -59,7 +63,9 @@ def _remove(path_info, fs, cache, force=False):
         fs.remove(path_info)
         return
 
-    current = stage(cache, path_info, fs, fs.PARAM_CHECKSUM).hash_info
+    current = stage(cache, path_info, fs, fs.PARAM_CHECKSUM, dry_run=True)[
+        1
+    ].hash_info
     try:
         obj = load(cache, current)
         check(cache, obj)
@@ -120,9 +126,11 @@ def _try_links(cache, from_info, to_info, link_types):
 
 
 def _link(cache, from_info, to_info):
-    assert cache.fs.isfile(from_info)
     cache.makedirs(to_info.parent)
-    _try_links(cache, from_info, to_info, cache.cache_types)
+    try:
+        _try_links(cache, from_info, to_info, cache.cache_types)
+    except FileNotFoundError as exc:
+        raise CheckoutError([to_info]) from exc
 
 
 def _cache_is_copy(cache, path_info):
@@ -134,7 +142,7 @@ def _cache_is_copy(cache, path_info):
         return True
 
     workspace_file = path_info.with_name("." + uuid())
-    test_cache_file = cache.fs.path_info / ".cache_type_test_file"
+    test_cache_file = cache.path_info / ".cache_type_test_file"
     if not cache.fs.exists(test_cache_file):
         cache.makedirs(test_cache_file.parent)
         with cache.fs.open(test_cache_file, "wb") as fobj:
@@ -163,7 +171,7 @@ def _checkout_file(
     modified = False
     cache_info = cache.hash_to_path_info(obj.hash_info.value)
     if fs.exists(path_info):
-        if not relink and _changed(path_info, fs, obj, cache):
+        if not relink and _changed(path_info, fs, obj, cache, state=state):
             modified = True
             _remove(path_info, fs, cache, force=force)
             _link(cache, cache_info, path_info)
@@ -222,19 +230,23 @@ def _checkout_dir(
 
     logger.debug("Linking directory '%s'.", path_info)
 
+    failed = []
     for entry_key, entry_obj in obj:
-        entry_modified = _checkout_file(
-            path_info.joinpath(*entry_key),
-            fs,
-            entry_obj,
-            cache,
-            force,
-            progress_callback,
-            relink,
-            state=None,
-        )
-        if entry_modified:
-            modified = True
+        try:
+            entry_modified = _checkout_file(
+                path_info.joinpath(*entry_key),
+                fs,
+                entry_obj,
+                cache,
+                force,
+                progress_callback,
+                relink,
+                state=None,
+            )
+            if entry_modified:
+                modified = True
+        except CheckoutError as exc:
+            failed.extend(exc.target_infos)
 
     modified = (
         _remove_redundant_files(
@@ -243,6 +255,9 @@ def _checkout_dir(
         or modified
     )
 
+    if failed:
+        raise CheckoutError(failed)
+
     if state:
         state.save(path_info, fs, obj.hash_info)
 
@@ -310,21 +325,9 @@ def checkout(
         _remove(path_info, fs, cache, force=force)
         failed = path_info
 
-    elif not relink and not _changed(path_info, fs, obj, cache):
+    elif not relink and not _changed(path_info, fs, obj, cache, state=state):
         logger.trace("Data '%s' didn't change.", path_info)  # type: ignore
         skip = True
-    else:
-        try:
-            check(cache, obj)
-        except (FileNotFoundError, ObjectFormatError):
-            if not quiet:
-                logger.warning(
-                    "Cache '%s' not found. File '%s' won't be created.",
-                    obj.hash_info,
-                    path_info,
-                )
-            _remove(path_info, fs, cache, force=force)
-            failed = path_info
 
     if failed or skip:
         if progress_callback and obj:

dvc/command/base.py

@@ -33,17 +33,12 @@ class CmdBase(ABC):
 
     def __init__(self, args):
         from dvc.repo import Repo
-        from dvc.updater import Updater
 
         os.chdir(args.cd)
 
         self.repo = Repo(uninitialized=self.UNINITIALIZED)
         self.config = self.repo.config
         self.args = args
-        if self.repo.tmp_dir:
-            hardlink_lock = self.config["core"].get("hardlink_lock", False)
-            updater = Updater(self.repo.tmp_dir, hardlink_lock=hardlink_lock)
-            updater.check()
 
     def do_run(self):
         with self.repo:

dvc/command/config.py

@@ -3,6 +3,7 @@ import logging
 import os
 
 from dvc.command.base import CmdBaseNoRepo, append_doc_link
+from dvc.ui import ui
 
 logger = logging.getLogger(__name__)
 
@@ -71,7 +72,9 @@ class CmdConfig(CmdBaseNoRepo):
             prefix = self._config_file_prefix(
                 self.args.show_origin, self.config, level
             )
-            logger.info("\n".join(self._format_config(conf, prefix)))
+            configs = list(self._format_config(conf, prefix))
+            if configs:
+                ui.write("\n".join(configs))
 
         return 0
 
@@ -96,7 +99,7 @@ class CmdConfig(CmdBaseNoRepo):
                 prefix = self._config_file_prefix(
                     self.args.show_origin, self.config, level
                 )
-                logger.info("{}{}".format(prefix, conf[section][opt]))
+                ui.write(prefix, conf[section][opt], sep="")
                 break
 
         return 0

dvc/command/dag.py

@@ -36,18 +36,20 @@ def _collect_targets(repo, target, outs):
         return [stage.addressing for stage, _ in pairs]
 
     targets = []
+
+    outs_trie = repo.index.outs_trie
     for stage, info in pairs:
         if not info:
             targets.extend([str(out) for out in stage.outs])
             continue
 
-        for out in repo.outs_trie.itervalues(prefix=info.parts):  # noqa: B301
+        for out in outs_trie.itervalues(prefix=info.parts):  # noqa: B301
             targets.extend(str(out))
 
     return targets
 
 
-def _transform(repo, outs):
+def _transform(index, outs):
     import networkx as nx
 
     from dvc.stage import Stage
@@ -55,7 +57,7 @@ def _transform(repo, outs):
     def _relabel(node) -> str:
         return node.addressing if isinstance(node, Stage) else str(node)
 
-    G = repo.outs_graph if outs else repo.graph
+    G = index.outs_graph if outs else index.graph
     return nx.relabel_nodes(G, _relabel, copy=True)
 
 
@@ -85,7 +87,7 @@ def _filter(G, targets, full):
 
 def _build(repo, target=None, full=False, outs=False):
     targets = _collect_targets(repo, target, outs)
-    G = _transform(repo, outs)
+    G = _transform(repo.index, outs)
     return _filter(G, targets, full)
 
 

dvc/command/data_sync.py

@@ -9,10 +9,11 @@ logger = logging.getLogger(__name__)
 
 class CmdDataBase(CmdBase):
     def log_summary(self, stats):
+        from dvc.ui import ui
         from dvc.utils.humanize import get_summary
 
         default_msg = "Everything is up to date."
-        logger.info(get_summary(stats.items()) or default_msg)
+        ui.write(get_summary(stats.items()) or default_msg)
 
 
 class CmdDataPull(CmdDataBase):

dvc/command/destroy.py

@@ -1,7 +1,6 @@
 import argparse
 import logging
 
-from dvc import prompt
 from dvc.command.base import CmdBase, append_doc_link
 
 logger = logging.getLogger(__name__)
@@ -10,6 +9,7 @@ logger = logging.getLogger(__name__)
 class CmdDestroy(CmdBase):
     def run(self):
         from dvc.exceptions import DvcException
+        from dvc.ui import ui
 
         try:
             statement = (
@@ -19,7 +19,7 @@ class CmdDestroy(CmdBase):
                 "Are you sure you want to continue?"
             )
 
-            if not self.args.force and not prompt.confirm(statement):
+            if not self.args.force and not ui.confirm(statement):
                 raise DvcException(
                     "cannot destroy without a confirmation from the user."
                     " Use `-f` to force."

dvc/command/diff.py

@@ -3,10 +3,9 @@ import json
 import logging
 import os
 
-import colorama
-
 from dvc.command import completion
 from dvc.command.base import CmdBase, append_doc_link
+from dvc.ui import ui
 
 logger = logging.getLogger(__name__)
 
@@ -18,13 +17,12 @@ def _digest(checksum):
 
 
 def _show_md(diff, show_hash=False, hide_missing=False):
-    from dvc.ui import ui
-
     headers = ["Status", "Hash", "Path"] if show_hash else ["Status", "Path"]
     rows = []
     statuses = ["added", "deleted", "renamed", "modified"]
     if not hide_missing:
         statuses.append("not in cache")
+
     for status in statuses:
         entries = diff.get(status, [])
         if not entries:
@@ -44,7 +42,7 @@ def _show_md(diff, show_hash=False, hide_missing=False):
 
 class CmdDiff(CmdBase):
     @staticmethod
-    def _format(diff, hide_missing=False):
+    def _show_diff(diff, hide_missing=False):
         """
         Given a diff structure, generate a string of paths separated
         by new lines and grouped together by their state.
@@ -71,15 +69,14 @@ class CmdDiff(CmdBase):
         """
 
         colors = {
-            "added": colorama.Fore.GREEN,
-            "modified": colorama.Fore.YELLOW,
-            "deleted": colorama.Fore.RED,
-            "renamed": colorama.Fore.GREEN,
-            "not in cache": colorama.Fore.YELLOW,
+            "added": "green",
+            "modified": "yellow",
+            "deleted": "red",
+            "renamed": "green",
+            "not in cache": "yellow",
         }
 
         summary = {}
-        groups = []
 
         states = ["added", "deleted", "renamed", "modified"]
         if not hide_missing:
@@ -91,7 +88,8 @@ class CmdDiff(CmdBase):
             if not entries:
                 continue
 
-            content = []
+            header = state.capitalize()
+            ui.write(f"[{colors[state]}]{header}[/]:", styled=True)
 
             for entry in entries:
                 path = entry["path"]
@@ -99,7 +97,7 @@ class CmdDiff(CmdBase):
                     path = f"{path['old']} -> {path['new']}"
                 checksum = entry.get("hash")
                 summary[state] += 1 if not path.endswith(os.sep) else 0
-                content.append(
+                ui.write(
                     "{space}{checksum}{separator}{path}".format(
                         space="    ",
                         checksum=_digest(checksum) if checksum else "",
@@ -108,27 +106,17 @@ class CmdDiff(CmdBase):
                     )
                 )
 
-            groups.append(
-                "{color}{header}{nc}:\n{content}".format(
-                    color=colors[state],
-                    header=state.capitalize(),
-                    nc=colorama.Fore.RESET,
-                    content="\n".join(content),
-                )
-            )
+            ui.write()
 
         if not sum(summary.values()):
             return None
 
-        fmt = (
-            "files summary: {added} added, {deleted} deleted,"
-            " {renamed} renamed, {modified} modified"
+        states_summary = ", ".join(
+            f"{summary[state]} {state}"
+            for state in states
+            if summary[state] > 0
         )
-        if not hide_missing:
-            fmt += ", {not in cache} not in cache"
-        groups.append(fmt.format_map(summary))
-
-        return "\n\n".join(groups)
+        ui.write("files summary:", states_summary)
 
     def run(self):
         from dvc.exceptions import DvcException
@@ -155,13 +143,11 @@ class CmdDiff(CmdBase):
                 diff[key] = entries
 
             if self.args.show_json:
-                logger.info(json.dumps(diff))
+                ui.write(json.dumps(diff))
             elif self.args.show_md:
-                logger.info(_show_md(diff, show_hash, hide_missing))
+                _show_md(diff, show_hash, hide_missing)
             elif diff:
-                output = self._format(diff, hide_missing)
-                if output:
-                    logger.info(output)
+                self._show_diff(diff, hide_missing)
 
         except DvcException:
             logger.exception("failed to get diff")

dvc/command/experiments.py

@@ -14,7 +14,9 @@ from dvc.command.metrics import DEFAULT_PRECISION
 from dvc.command.repro import CmdRepro
 from dvc.command.repro import add_arguments as add_repro_arguments
 from dvc.exceptions import DvcException, InvalidArgumentError
+from dvc.ui import ui
 from dvc.utils.flatten import flatten
+from dvc.utils.serialize import encode_exception
 
 if TYPE_CHECKING:
     from rich.text import Text
@@ -27,6 +29,7 @@ logger = logging.getLogger(__name__)
 
 SHOW_MAX_WIDTH = 1024
 FILL_VALUE = "-"
+FILL_VALUE_ERRORED = "!"
 
 
 def _filter_name(names, label, filter_strs):
@@ -89,6 +92,7 @@ def _filter_names(
 
 def _update_names(names, items):
     for name, item in items:
+        item = item.get("data", {})
         if isinstance(item, dict):
             item = flatten(item)
             names[name].update({key: None for key in item})
@@ -99,7 +103,8 @@ def _collect_names(all_experiments, **kwargs):
     param_names = defaultdict(dict)
 
     for _, experiments in all_experiments.items():
-        for exp in experiments.values():
+        for exp_data in experiments.values():
+            exp = exp_data.get("data", {})
             _update_names(metric_names, exp.get("metrics", {}).items())
             _update_names(param_names, exp.get("params", {}).items())
     metric_names = _filter_names(
@@ -149,8 +154,15 @@ def _collect_rows(
         )
 
     new_checkpoint = True
-    for i, (rev, exp) in enumerate(experiments.items()):
-        queued = str(exp.get("queued") or "")
+    for i, (rev, results) in enumerate(experiments.items()):
+        exp = results.get("data", {})
+        if exp.get("running"):
+            state = "Running"
+        elif exp.get("queued"):
+            state = "Queued"
+        else:
+            state = FILL_VALUE
+        executor = exp.get("executor", FILL_VALUE)
         is_baseline = rev == "baseline"
 
         if is_baseline:
@@ -162,7 +174,7 @@ def _collect_rows(
         tip = exp.get("checkpoint_tip")
 
         parent_rev = exp.get("checkpoint_parent", "")
-        parent_exp = experiments.get(parent_rev, {})
+        parent_exp = experiments.get(parent_rev, {}).get("data", {})
         parent_tip = parent_exp.get("checkpoint_tip")
 
         parent = ""
@@ -189,15 +201,27 @@ def _collect_rows(
         row = [
             exp_name,
             name_rev,
-            queued,
             typ,
             _format_time(exp.get("timestamp")),
             parent,
+            state,
+            executor,
         ]
+        fill_value = FILL_VALUE_ERRORED if results.get("error") else FILL_VALUE
         _extend_row(
-            row, metric_names, exp.get("metrics", {}).items(), precision
+            row,
+            metric_names,
+            exp.get("metrics", {}).items(),
+            precision,
+            fill_value=fill_value,
+        )
+        _extend_row(
+            row,
+            param_names,
+            exp.get("params", {}).items(),
+            precision,
+            fill_value=fill_value,
         )
-        _extend_row(row, param_names, exp.get("params", {}).items(), precision)
 
         yield row
 
@@ -260,19 +284,23 @@ def _format_time(timestamp):
     return timestamp.strftime(fmt)
 
 
-def _extend_row(row, names, items, precision):
+def _extend_row(row, names, items, precision, fill_value=FILL_VALUE):
     from rich.text import Text
 
     from dvc.compare import _format_field, with_value
 
     if not items:
-        row.extend(FILL_VALUE for keys in names.values() for _ in keys)
+        row.extend(fill_value for keys in names.values() for _ in keys)
         return
 
-    for fname, item in items:
+    for fname, data in items:
+        item = data.get("data", {})
         item = flatten(item) if isinstance(item, dict) else {fname: item}
         for name in names[fname]:
-            value = with_value(item.get(name), FILL_VALUE)
+            value = with_value(
+                item.get(name),
+                FILL_VALUE_ERRORED if data.get("error", None) else fill_value,
+            )
             # wrap field data in rich.Text, otherwise rich may
             # interpret unescaped braces from list/dict types as rich
             # markup tags
@@ -304,7 +332,15 @@ def experiments_table(
 
     from dvc.compare import TabularData
 
-    headers = ["Experiment", "rev", "queued", "typ", "Created", "parent"]
+    headers = [
+        "Experiment",
+        "rev",
+        "typ",
+        "Created",
+        "parent",
+        "State",
+        "Executor",
+    ]
     td = TabularData(
         lconcat(headers, metric_headers, param_headers), fill_value=FILL_VALUE
     )
@@ -340,8 +376,7 @@ def prepare_exp_id(kwargs) -> "Text":
     text.append(suff)
 
     tree = experiment_types[typ]
-    queued = "*" if kwargs.get("queued") else ""
-    pref = (f"{tree} " if tree else "") + queued
+    pref = f"{tree} " if tree else ""
     return Text(pref) + text
 
 
@@ -381,9 +416,13 @@ def show_experiments(
     if no_timestamp:
         td.drop("Created")
 
+    for col in ("State", "Executor"):
+        if td.is_empty(col):
+            td.drop(col)
+
     row_styles = lmap(baseline_styler, td.column("typ"))
 
-    merge_headers = ["Experiment", "rev", "queued", "typ", "parent"]
+    merge_headers = ["Experiment", "rev", "typ", "parent"]
     td.column("Experiment")[:] = map(prepare_exp_id, td.as_dict(merge_headers))
     td.drop(*merge_headers[1:])
 
@@ -391,12 +430,14 @@ def show_experiments(
     styles = {
         "Experiment": {"no_wrap": True, "header_style": "black on grey93"},
         "Created": {"header_style": "black on grey93"},
+        "State": {"header_style": "black on grey93"},
+        "Executor": {"header_style": "black on grey93"},
     }
     header_bg_colors = {"metrics": "cornsilk1", "params": "light_cyan1"}
     styles.update(
         {
             header: {
-                "justify": "left" if typ == "metrics" else "params",
+                "justify": "right" if typ == "metrics" else "left",
                 "header_style": f"black on {header_bg_colors[typ]}",
                 "collapse": idx != 0,
                 "no_wrap": typ == "metrics",
@@ -429,7 +470,7 @@ def _normalize_headers(names):
 def _format_json(item):
     if isinstance(item, (date, datetime)):
         return item.isoformat()
-    raise TypeError
+    return encode_exception(item)
 
 
 class CmdExperimentsShow(CmdBase):
@@ -450,7 +491,7 @@ class CmdExperimentsShow(CmdBase):
         if self.args.show_json:
             import json
 
-            logger.info(json.dumps(all_experiments, default=_format_json))
+            ui.write(json.dumps(all_experiments, default=_format_json))
         else:
             show_experiments(
                 all_experiments,
@@ -494,7 +535,7 @@ class CmdExperimentsDiff(CmdBase):
         if self.args.show_json:
             import json
 
-            logger.info(json.dumps(diff))
+            ui.write(json.dumps(diff))
         else:
             from dvc.compare import show_diff
 
@@ -502,8 +543,6 @@ class CmdExperimentsDiff(CmdBase):
             diffs = [("metrics", "Metric"), ("params", "Param")]
             for idx, (key, title) in enumerate(diffs):
                 if idx:
-                    from dvc.ui import ui
-
                     # we are printing tables even in `--quiet` mode
                     # so we should also be printing the "table" separator
                     ui.write(force=True)
@@ -537,7 +576,7 @@ class CmdExperimentsRun(CmdRepro):
                 )
 
         if self.args.reset:
-            logger.info("Any existing checkpoints will be reset and re-run.")
+            ui.write("Any existing checkpoints will be reset and re-run.")
 
         results = self.repo.experiments.run(
             name=self.args.name,
@@ -554,7 +593,7 @@ class CmdExperimentsRun(CmdRepro):
         if self.args.metrics and results:
             metrics = self.repo.metrics.show(revs=list(results))
             metrics.pop("workspace", None)
-            logger.info(show_metrics(metrics))
+            show_metrics(metrics)
 
         return 0
 
@@ -590,7 +629,7 @@ class CmdExperimentsGC(CmdRepro):
         msg += " of the current repo."
         if self.args.queued:
             msg += " Run queued experiments will be preserved."
-        if self.args.queued:
+        else:
             msg += " Run queued experiments will be removed."
 
         logger.warning(msg)
@@ -608,12 +647,13 @@ class CmdExperimentsGC(CmdRepro):
         )
 
         if removed:
-            logger.info(
-                f"Removed {removed} experiments. To remove unused cache files "
-                "use 'dvc gc'."
+            ui.write(
+                f"Removed {removed} experiments.",
+                "To remove unused cache files",
+                "use 'dvc gc'.",
             )
         else:
-            logger.info("No experiments to remove.")
+            ui.write("No experiments to remove.")
         return 0
 
 
@@ -662,15 +702,15 @@ class CmdExperimentsPush(CmdBase):
             run_cache=self.args.run_cache,
         )
 
-        logger.info(
-            "Pushed experiment '%s' to Git remote '%s'.",
-            self.args.experiment,
-            self.args.git_remote,
+        ui.write(
+            f"Pushed experiment '{self.args.experiment}'"
+            f"to Git remote '{self.args.git_remote}'."
         )
         if not self.args.push_cache:
-            logger.info(
-                "To push cached outputs for this experiment to DVC remote "
-                "storage, re-run this command without '--no-cache'."
+            ui.write(
+                "To push cached outputs",
+                "for this experiment to DVC remote storage,"
+                "re-run this command without '--no-cache'.",
             )
 
         return 0
@@ -689,15 +729,15 @@ class CmdExperimentsPull(CmdBase):
             run_cache=self.args.run_cache,
         )
 
-        logger.info(
-            "Pulled experiment '%s' from Git remote '%s'. ",
-            self.args.experiment,
-            self.args.git_remote,
+        ui.write(
+            f"Pulled experiment '{self.args.experiment}'",
+            f"from Git remote '{self.args.git_remote}'.",
         )
         if not self.args.pull_cache:
-            logger.info(
-                "To pull cached outputs for this experiment from DVC remote "
-                "storage, re-run this command without '--no-cache'."
+            ui.write(
+                "To pull cached outputs for this experiment"
+                "from DVC remote storage,"
+                "re-run this command without '--no-cache'."
             )
 
         return 0
@@ -937,9 +977,7 @@ def add_parser(subparsers, parent_parser):
     )
     experiments_diff_parser.set_defaults(func=CmdExperimentsDiff)
 
-    EXPERIMENTS_RUN_HELP = (
-        "Reproduce complete or partial experiment pipelines."
-    )
+    EXPERIMENTS_RUN_HELP = "Run or resume an experiment."
     experiments_run_parser = experiments_subparsers.add_parser(
         "run",
         parents=[parent_parser],

dvc/command/gc.py

@@ -2,8 +2,8 @@ import argparse
 import logging
 import os
 
-from dvc import prompt
 from dvc.command.base import CmdBase, append_doc_link
+from dvc.ui import ui
 
 logger = logging.getLogger(__name__)
 
@@ -44,7 +44,7 @@ class CmdGC(CmdBase):
         logger.warning(msg)
 
         msg = "Are you sure you want to proceed?"
-        if not self.args.force and not prompt.confirm(msg):
+        if not self.args.force and not ui.confirm(msg):
             return 1
 
         self.repo.gc(

dvc/command/get.py

@@ -12,12 +12,13 @@ logger = logging.getLogger(__name__)
 class CmdGet(CmdBaseNoRepo):
     def _show_url(self):
         from dvc.api import get_url
+        from dvc.ui import ui
 
         try:
             url = get_url(
                 self.args.path, repo=self.args.url, rev=self.args.rev
             )
-            logger.info(url)
+            ui.write(url, force=True)
         except DvcException:
             logger.exception("failed to show URL")
             return 1

dvc/command/init.py

@@ -12,8 +12,10 @@ logger = logging.getLogger(__name__)
 
 
 def _welcome_message():
+    from dvc.ui import ui
+
     if analytics.is_enabled():
-        logger.info(
+        ui.write(
             boxify(
                 "DVC has enabled anonymous aggregate usage analytics.\n"
                 "Read the analytics documentation (and how to opt-out) here:\n"
@@ -30,7 +32,7 @@ def _welcome_message():
         f"- Star us on GitHub: {fmt_link('https://github.com/iterative/dvc')}"
     ).format(yellow=colorama.Fore.YELLOW, nc=colorama.Fore.RESET)
 
-    logger.info(msg)
+    ui.write(msg)
 
 
 class CmdInit(CmdBaseNoRepo):

dvc/command/live.py

@@ -1,11 +1,9 @@
 import argparse
-import logging
 import os
 
 from dvc.command import completion
 from dvc.command.base import CmdBase, fix_subparsers
-
-logger = logging.getLogger(__name__)
+from dvc.ui import ui
 
 
 class CmdLive(CmdBase):
@@ -14,12 +12,13 @@ class CmdLive(CmdBase):
     def _run(self, target, revs=None):
         metrics, plots = self.repo.live.show(target=target, revs=revs)
 
-        html_path = self.args.target + ".html"
-        self.repo.plots.write_html(html_path, plots, metrics)
+        if plots:
+            html_path = self.args.target + ".html"
+            self.repo.plots.write_html(html_path, plots, metrics)
 
-        logger.info(f"\nfile://{os.path.abspath(html_path)}")
-
-        return 0
+            ui.write("\nfile://", os.path.abspath(html_path), sep="")
+            return 0
+        return 1
 
 
 class CmdLiveShow(CmdLive):

dvc/command/metrics.py

@@ -3,7 +3,9 @@ import logging
 
 from dvc.command import completion
 from dvc.command.base import CmdBase, append_doc_link, fix_subparsers
-from dvc.exceptions import BadMetricError, DvcException
+from dvc.exceptions import DvcException
+from dvc.ui import ui
+from dvc.utils.serialize import encode_exception
 
 logger = logging.getLogger(__name__)
 
@@ -32,17 +34,10 @@ class CmdMetricsShow(CmdMetricsBase):
         if self.args.show_json:
             import json
 
-            logger.info(json.dumps(metrics))
+            ui.write(json.dumps(metrics, default=encode_exception))
         else:
             from dvc.compare import show_metrics
 
-            # When `metrics` contains a `None` key, it means that some files
-            # specified as `targets` in `repo.metrics.show` didn't contain any
-            # metrics.
-            missing = metrics.pop(None, None)
-            if missing:
-                raise BadMetricError(missing)
-
             show_metrics(
                 metrics,
                 markdown=self.args.show_md,
@@ -73,7 +68,7 @@ class CmdMetricsDiff(CmdMetricsBase):
         if self.args.show_json:
             import json
 
-            logger.info(json.dumps(diff))
+            ui.write(json.dumps(diff))
         else:
             from dvc.compare import show_diff
 

dvc/command/params.py

@@ -4,6 +4,7 @@ import logging
 from dvc.command import completion
 from dvc.command.base import CmdBase, append_doc_link, fix_subparsers
 from dvc.exceptions import DvcException
+from dvc.ui import ui
 
 logger = logging.getLogger(__name__)
 
@@ -27,7 +28,7 @@ class CmdParamsDiff(CmdBase):
         if self.args.show_json:
             import json
 
-            logger.info(json.dumps(diff))
+            ui.write(json.dumps(diff))
         else:
             from dvc.compare import show_diff
 

dvc/command/plots.py

@@ -4,6 +4,7 @@ import logging
 from dvc.command import completion
 from dvc.command.base import CmdBase, append_doc_link, fix_subparsers
 from dvc.exceptions import DvcException
+from dvc.ui import ui
 from dvc.utils import format_link
 
 logger = logging.getLogger(__name__)
@@ -38,30 +39,38 @@ class CmdPlots(CmdBase):
 
             if self.args.show_vega:
                 target = self.args.targets[0]
-                logger.info(plots[target])
+                ui.write(plots[target])
                 return 0
 
-            rel: str = self.args.out or "plots.html"
-            path = (Path.cwd() / rel).resolve()
-            self.repo.plots.write_html(
-                path, plots=plots, html_template_path=self.args.html_template
-            )
-
         except DvcException:
             logger.exception("")
             return 1
 
-        assert path.is_absolute()  # as_uri throws ValueError if not absolute
-        url = path.as_uri()
-        logger.info(url)
-        if self.args.open:
-            import webbrowser
+        if plots:
+            rel: str = self.args.out or "plots.html"
+            path: Path = (Path.cwd() / rel).resolve()
+            self.repo.plots.write_html(
+                path, plots=plots, html_template_path=self.args.html_template
+            )
 
-            opened = webbrowser.open(rel)
-            if not opened:
-                logger.error("Failed to open. Please try opening it manually.")
-                return 1
+            assert (
+                path.is_absolute()
+            )  # as_uri throws ValueError if not absolute
+            url = path.as_uri()
+            ui.write(url)
+            if self.args.open:
+                import webbrowser
 
+                opened = webbrowser.open(rel)
+                if not opened:
+                    ui.error_write(
+                        "Failed to open. Please try opening it manually."
+                    )
+                    return 1
+        else:
+            ui.error_write(
+                "No plots were loaded, visualization file will not be created."
+            )
         return 0
 
 

dvc/command/remote.py

@@ -1,12 +1,10 @@
 import argparse
-import logging
 
 from dvc.command.base import append_doc_link, fix_subparsers
 from dvc.command.config import CmdConfig
+from dvc.ui import ui
 from dvc.utils import format_link
 
-logger = logging.getLogger(__name__)
-
 
 class CmdRemote(CmdConfig):
     def __init__(self, args):
@@ -27,7 +25,7 @@ class CmdRemoteAdd(CmdRemote):
         from dvc.config import ConfigError
 
         if self.args.default:
-            logger.info(f"Setting '{self.args.name}' as a default remote.")
+            ui.write(f"Setting '{self.args.name}' as a default remote.")
 
         with self.config.edit(self.args.level) as conf:
             if self.args.name in conf["remote"] and not self.args.force:
@@ -90,7 +88,7 @@ class CmdRemoteDefault(CmdRemote):
             try:
                 print(conf["core"]["remote"])
             except KeyError:
-                logger.info("No default remote set")
+                ui.write("No default remote set")
                 return 1
         else:
             with self.config.edit(self.args.level) as conf:
@@ -116,7 +114,7 @@ class CmdRemoteList(CmdRemote):
     def run(self):
         conf = self.config.read(self.args.level)
         for name, conf in conf["remote"].items():
-            logger.info("{}\t{}".format(name, conf["url"]))
+            ui.write(name, conf["url"], sep="\t")
         return 0
 
 

dvc/command/repro.py

@@ -1,20 +1,19 @@
 import argparse
-import logging
 
 from dvc.command import completion
 from dvc.command.base import CmdBase, append_doc_link
 from dvc.command.status import CmdDataStatus
 
-logger = logging.getLogger(__name__)
-
 
 class CmdRepro(CmdBase):
     def run(self):
+        from dvc.ui import ui
+
         stages = self.repo.reproduce(**self._repro_kwargs)
         if len(stages) == 0:
-            logger.info(CmdDataStatus.UP_TO_DATE_MSG)
+            ui.write(CmdDataStatus.UP_TO_DATE_MSG)
         else:
-            logger.info(
+            ui.write(
                 "Use `dvc push` to send your updates to " "remote storage."
             )
 

dvc/command/root.py

@@ -10,8 +10,9 @@ logger = logging.getLogger(__name__)
 class CmdRoot(CmdBaseNoRepo):
     def run(self):
         from dvc.repo import Repo
+        from dvc.ui import ui
 
-        logger.info(relpath(Repo.find_root()))
+        ui.write(relpath(Repo.find_root()))
         return 0
 
 

dvc/command/stage.py

@@ -66,7 +66,7 @@ def prepare_stages_data(
 class CmdStageList(CmdBase):
     def _get_stages(self) -> Iterable["Stage"]:
         if self.args.all:
-            stages: List["Stage"] = self.repo.stages  # type: ignore
+            stages: List["Stage"] = self.repo.index.stages  # type: ignore
             logger.trace(  # type: ignore[attr-defined]
                 "%d no. of stages found", len(stages)
             )

dvc/command/status.py

@@ -2,6 +2,7 @@ import logging
 
 from dvc.command.data_sync import CmdDataBase
 from dvc.exceptions import DvcException
+from dvc.ui import ui
 from dvc.utils import format_link
 
 logger = logging.getLogger(__name__)
@@ -26,7 +27,7 @@ class CmdDataStatus(CmdDataBase):
         ind = indent * self.STATUS_INDENT
 
         if isinstance(status, str):
-            logger.info(f"{ind}{status}")
+            ui.write(f"{ind}{status}")
             return
 
         if isinstance(status, list):
@@ -38,14 +39,13 @@ class CmdDataStatus(CmdDataBase):
 
         for key, value in status.items():
             if isinstance(value, str):
-                logger.info("{}{}{}".format(ind, self._normalize(value), key))
+                ui.write(f"{ind}{self._normalize(value)}{key}")
             elif value:
-                logger.info(f"{ind}{key}:")
+                ui.write(f"{ind}{key}:")
                 self._show(value, indent + 1)
 
     def run(self):
         from dvc.repo import lock_repo
-        from dvc.ui import ui
 
         indent = 1 if self.args.cloud else 0
 
@@ -72,7 +72,7 @@ class CmdDataStatus(CmdDataBase):
             if self.args.show_json:
                 import json
 
-                logger.info(json.dumps(st))
+                ui.write(json.dumps(st))
                 return 0
 
             if st:
@@ -80,7 +80,7 @@ class CmdDataStatus(CmdDataBase):
                 return 0
 
             # additional hints for the user
-            if not self.repo.stages:
+            if not self.repo.index.stages:
                 ui.write(self.EMPTY_PROJECT_MSG)
             elif self.args.cloud or self.args.remote:
                 remote = self.args.remote or self.repo.config["core"].get(

dvc/command/version.py

@@ -2,6 +2,7 @@ import argparse
 import logging
 
 from dvc.command.base import CmdBaseNoRepo, append_doc_link
+from dvc.ui import ui
 
 logger = logging.getLogger(__name__)
 
@@ -9,9 +10,12 @@ logger = logging.getLogger(__name__)
 class CmdVersion(CmdBaseNoRepo):
     def run(self):
         from dvc.info import get_dvc_info
+        from dvc.updater import notify_updates
 
         dvc_info = get_dvc_info()
-        logger.info(dvc_info)
+        ui.write(dvc_info, force=True)
+
+        notify_updates()
         return 0
 
 

dvc/compare.py

@@ -134,6 +134,10 @@ class TabularData(MutableSequence[Sequence["CellT"]]):
         self.drop(*(set(self._keys) - set(col_names)))
         self._keys = list(col_names)
 
+    def is_empty(self, col_name: str) -> bool:
+        col = self.column(col_name)
+        return not any(item != self._fill_value for item in col)
+
     def to_csv(self) -> str:
         import csv
         from io import StringIO
@@ -277,8 +281,9 @@ def metrics_table(
     td = TabularData(["Revision", "Path"], fill_value="-")
 
     for branch, val in metrics.items():
-        for fname, metric in val.items():
+        for fname, metric in val.get("data", {}).items():
             row_data: Dict[str, str] = {"Revision": branch, "Path": fname}
+            metric = metric.get("data", {})
             flattened = (
                 flatten(format_dict(metric))
                 if isinstance(metric, dict)

dvc/config.py

@@ -20,7 +20,15 @@ class ConfigError(DvcException):
         super().__init__(f"config file error: {msg}")
 
 
-class NoRemoteError(ConfigError):
+class RemoteConfigError(ConfigError):
+    pass
+
+
+class NoRemoteError(RemoteConfigError):
+    pass
+
+
+class RemoteNotFoundError(RemoteConfigError):
     pass
 
 

dvc/config_schema.py

@@ -93,7 +93,7 @@ HTTP_COMMON = {
     "user": str,
     "password": str,
     "ask_password": Bool,
-    "ssl_verify": Bool,
+    "ssl_verify": Any(Bool, str),
     "method": str,
 }
 WEBDAV_COMMON = {
@@ -104,6 +104,7 @@ WEBDAV_COMMON = {
     "cert_path": str,
     "key_path": str,
     "timeout": Coerce(int),
+    "ssl_verify": Any(Bool, str),
 }
 
 SCHEMA = {
@@ -145,7 +146,7 @@ SCHEMA = {
                     "session_token": str,
                     Optional("listobjects", default=False): Bool,  # obsoleted
                     Optional("use_ssl", default=True): Bool,
-                    Optional("ssl_verify", default=True): Bool,
+                    "ssl_verify": Any(Bool, str),
                     "sse": str,
                     "sse_kms_key_id": str,
                     "acl": str,
@@ -188,6 +189,11 @@ SCHEMA = {
                     "tenant_id": str,
                     "client_id": str,
                     "client_secret": str,
+                    "allow_anonymous_login": Bool,
+                    "exclude_environment_credential": Bool,
+                    "exclude_visual_studio_code_credential": Bool,
+                    "exclude_shared_token_cache_credential": Bool,
+                    "exclude_managed_identity_credential": Bool,
                     **REMOTE_COMMON,
                 },
                 "oss": {

dvc/dagascii.py

@@ -163,7 +163,7 @@ class AsciiCanvas:
         self.point(x0 + width, y0 + height, "+")
 
 
-def _build_sugiyama_layout(vertexes, edges):
+def _build_sugiyama_layout(vertices, edges):
     #
     # Just a reminder about naming conventions:
     # +------------X
@@ -174,17 +174,17 @@ def _build_sugiyama_layout(vertexes, edges):
     # Y
     #
 
-    vertexes = {v: Vertex(f" {v} ") for v in vertexes}
+    vertices = {v: Vertex(f" {v} ") for v in vertices}
     # NOTE: reverting edges to correctly orientate the graph
-    edges = [Edge(vertexes[e], vertexes[s]) for s, e in edges]
-    vertexes = vertexes.values()
-    graph = Graph(vertexes, edges)
+    edges = [Edge(vertices[e], vertices[s]) for s, e in edges]
+    vertices = vertices.values()
+    graph = Graph(vertices, edges)
 
-    for vertex in vertexes:
+    for vertex in vertices:
         vertex.view = VertexViewer(vertex.data)
 
     # NOTE: determine min box length to create the best layout
-    minw = min(v.view.w for v in vertexes)
+    minw = min(v.view.w for v in vertices)
 
     for edge in edges:
         edge.view = EdgeViewer()
@@ -204,11 +204,11 @@ def _build_sugiyama_layout(vertexes, edges):
     return sug
 
 
-def draw(vertexes, edges):
+def draw(vertices, edges):
     """Build a DAG and draw it in ASCII.
 
     Args:
-        vertexes (list): list of graph vertexes.
+        vertices (list): list of graph vertices.
         edges (list): list of graph edges.
     """
     # pylint: disable=too-many-locals
@@ -217,7 +217,7 @@ def draw(vertexes, edges):
     Xs = []  # pylint: disable=invalid-name
     Ys = []  # pylint: disable=invalid-name
 
-    sug = _build_sugiyama_layout(vertexes, edges)
+    sug = _build_sugiyama_layout(vertices, edges)
 
     for vertex in sug.g.sV:
         # NOTE: moving boxes w/2 to the left

dvc/data_cloud.py

@@ -1,6 +1,13 @@
 """Manages dvc remotes that user can use with push/pull/status commands."""
 
 import logging
+from typing import TYPE_CHECKING, Iterable, Optional
+
+from dvc.objects.db import get_index
+
+if TYPE_CHECKING:
+    from dvc.hash_info import HashInfo
+    from dvc.objects.db.base import ObjectDB
 
 logger = logging.getLogger(__name__)
 
@@ -19,14 +26,18 @@ class DataCloud:
     def __init__(self, repo):
         self.repo = repo
 
-    def get_remote(self, name=None, command="<command>"):
+    def get_remote_odb(
+        self,
+        name: Optional[str] = None,
+        command: str = "<command>",
+    ) -> "ObjectDB":
         from dvc.config import NoRemoteError
 
         if not name:
             name = self.repo.config["core"].get("remote")
 
         if name:
-            return self._init_remote(name)
+            return self._init_odb(name)
 
         if bool(self.repo.config["remote"]):
             error_msg = (
@@ -43,81 +54,107 @@ class DataCloud:
 
         raise NoRemoteError(error_msg)
 
-    def _init_remote(self, name):
-        from dvc.remote import get_remote
+    def _init_odb(self, name):
+        from dvc.fs import get_cloud_fs
+        from dvc.objects.db import get_odb
 
-        return get_remote(self.repo, name=name)
+        cls, config, path_info = get_cloud_fs(self.repo, name=name)
+        config["tmp_dir"] = self.repo.tmp_dir
+        return get_odb(cls(**config), path_info, **config)
 
-    def push(self, cache, jobs=None, remote=None, show_checksums=False):
+    def push(
+        self,
+        objs: Iterable["HashInfo"],
+        jobs: Optional[int] = None,
+        remote: Optional[str] = None,
+        odb: Optional["ObjectDB"] = None,
+    ):
         """Push data items in a cloud-agnostic way.
 
         Args:
-            cache (NamedCache): named checksums to push to the cloud.
-            jobs (int): number of jobs that can be running simultaneously.
-            remote (dvc.remote.base.BaseRemote): optional remote to push to.
+            objs: objects to push to the cloud.
+            jobs: number of jobs that can be running simultaneously.
+            remote: optional name of remote to push to.
                 By default remote from core.remote config option is used.
-            show_checksums (bool): show checksums instead of file names in
-                information messages.
+            odb: optional ODB to push to. Overrides remote.
         """
-        remote = self.get_remote(remote, "push")
+        from dvc.objects.transfer import transfer
 
-        return remote.push(
+        if not odb:
+            odb = self.get_remote_odb(remote, "push")
+        return transfer(
             self.repo.odb.local,
-            cache,
+            odb,
+            objs,
             jobs=jobs,
-            show_checksums=show_checksums,
+            dest_index=get_index(odb),
+            cache_odb=self.repo.odb.local,
         )
 
-    def pull(self, cache, jobs=None, remote=None, show_checksums=False):
+    def pull(
+        self,
+        objs: Iterable["HashInfo"],
+        jobs: Optional[int] = None,
+        remote: Optional[str] = None,
+        odb: Optional["ObjectDB"] = None,
+    ):
         """Pull data items in a cloud-agnostic way.
 
         Args:
-            cache (NamedCache): named checksums to pull from the cloud.
-            jobs (int): number of jobs that can be running simultaneously.
-            remote (dvc.remote.base.BaseRemote): optional remote to pull from.
+            objs: objects to pull from the cloud.
+            jobs: number of jobs that can be running simultaneously.
+            remote: optional name of remote to pull from.
                 By default remote from core.remote config option is used.
-            show_checksums (bool): show checksums instead of file names in
-                information messages.
+            odb: optional ODB to pull from. Overrides remote.
         """
-        remote = self.get_remote(remote, "pull")
+        from dvc.objects.transfer import transfer
 
-        return remote.pull(
+        if not odb:
+            odb = self.get_remote_odb(remote, "pull")
+        return transfer(
+            odb,
             self.repo.odb.local,
-            cache,
+            objs,
             jobs=jobs,
-            show_checksums=show_checksums,
+            src_index=get_index(odb),
+            cache_odb=self.repo.odb.local,
+            verify=odb.verify,
         )
 
     def status(
         self,
-        cache,
-        jobs=None,
-        remote=None,
-        show_checksums=False,
-        log_missing=True,
+        objs: Iterable["HashInfo"],
+        jobs: Optional[int] = None,
+        remote: Optional[str] = None,
+        odb: Optional["ObjectDB"] = None,
+        log_missing: bool = True,
     ):
         """Check status of data items in a cloud-agnostic way.
 
         Args:
-            cache (NamedCache): named checksums to check status for.
-            jobs (int): number of jobs that can be running simultaneously.
-            remote (dvc.remote.base.BaseRemote): optional remote to compare
+            objs: objects to check status for.
+            jobs: number of jobs that can be running simultaneously.
+            remote: optional remote to compare
                 cache to. By default remote from core.remote config option
                 is used.
-            show_checksums (bool): show checksums instead of file names in
-                information messages.
-            log_missing (bool): log warning messages if file doesn't exist
+            odb: optional ODB to check status from. Overrides remote.
+            log_missing: log warning messages if file doesn't exist
                 neither in cache, neither in cloud.
         """
-        remote = self.get_remote(remote, "status")
-        return remote.status(
+        from dvc.objects.status import compare_status
+
+        if not odb:
+            odb = self.get_remote_odb(remote, "status")
+        return compare_status(
             self.repo.odb.local,
-            cache,
+            odb,
+            objs,
             jobs=jobs,
-            show_checksums=show_checksums,
             log_missing=log_missing,
+            dest_index=get_index(odb),
+            cache_odb=self.repo.odb.local,
         )
 
     def get_url_for(self, remote, checksum):
-        remote = self.get_remote(remote)
-        return str(remote.odb.hash_to_path_info(checksum))
+        remote_odb = self.get_remote_odb(remote)
+        return str(remote_odb.hash_to_path_info(checksum))

dvc/dependency/param.py

@@ -98,7 +98,7 @@ class ParamsDependency(Dependency):
                 f"Unable to read parameters from '{self}'"
             ) from exc
 
-    def read_params_d(self):
+    def read_params_d(self, **kwargs):
         config = self._read()
 
         ret = {}

dvc/dependency/repo.py

@@ -1,9 +1,19 @@
+import os
+from collections import defaultdict
+from copy import copy
+from typing import TYPE_CHECKING, Dict, Optional, Set, Tuple
+
 from voluptuous import Required
 
 from dvc.path_info import PathInfo
 
 from .base import Dependency
 
+if TYPE_CHECKING:
+    from dvc.hash_info import HashInfo
+    from dvc.objects.db.base import ObjectDB
+    from dvc.objects.file import HashFile
+
 
 class RepoDependency(Dependency):
     PARAM_REPO = "repo"
@@ -21,6 +31,7 @@ class RepoDependency(Dependency):
 
     def __init__(self, def_repo, stage, *args, **kwargs):
         self.def_repo = def_repo
+        self._staged_objs: Dict[str, "HashFile"] = {}
         super().__init__(stage, *args, **kwargs)
 
     def _parse_path(self, fs, path_info):
@@ -30,37 +41,12 @@ class RepoDependency(Dependency):
     def is_in_repo(self):
         return False
 
-    @property
-    def repo_pair(self):
-        d = self.def_repo
-        rev = d.get(self.PARAM_REV_LOCK) or d.get(self.PARAM_REV)
-        return d[self.PARAM_URL], rev
-
     def __str__(self):
-        return "{} ({})".format(self.def_path, self.def_repo[self.PARAM_URL])
-
-    def _make_repo(self, *, locked=True, **kwargs):
-        from dvc.external_repo import external_repo
-
-        d = self.def_repo
-        rev = (d.get("rev_lock") if locked else None) or d.get("rev")
-        return external_repo(d["url"], rev=rev, **kwargs)
-
-    def _get_hash(self, locked=True):
-        from dvc.objects.stage import stage
-
-        with self._make_repo(locked=locked) as repo:
-            path_info = PathInfo(repo.root_dir) / self.def_path
-            return stage(
-                self.repo.odb.local,
-                path_info,
-                repo.repo_fs,
-                self.repo.odb.local.fs.PARAM_CHECKSUM,
-            ).hash_info
+        return f"{self.def_path} ({self.def_repo[self.PARAM_URL]})"
 
     def workspace_status(self):
-        current = self._get_hash(locked=True)
-        updated = self._get_hash(locked=False)
+        current = self.get_obj(locked=True).hash_info
+        updated = self.get_obj(locked=False).hash_info
 
         if current != updated:
             return {str(self): "update available"}
@@ -78,31 +64,16 @@ class RepoDependency(Dependency):
 
     def download(self, to, jobs=None):
         from dvc.checkout import checkout
-        from dvc.config import NoRemoteError
-        from dvc.exceptions import NoOutputOrStageError
-        from dvc.objects import save
-        from dvc.objects.stage import stage
 
-        odb = self.repo.odb.local
-
-        with self._make_repo(cache_dir=odb.cache_dir) as repo:
-            if self.def_repo.get(self.PARAM_REV_LOCK) is None:
-                self.def_repo[self.PARAM_REV_LOCK] = repo.get_rev()
-            path_info = PathInfo(repo.root_dir) / self.def_path
-            try:
-                repo.fetch([path_info.fspath], jobs=jobs, recursive=True)
-            except (NoOutputOrStageError, NoRemoteError):
-                pass
-            obj = stage(
-                odb, path_info, repo.repo_fs, odb.fs.PARAM_CHECKSUM, jobs=jobs
-            )
-            save(odb, obj, jobs=jobs)
+        for odb, objs in self.get_used_objs().items():
+            self.repo.cloud.pull(objs, jobs=jobs, odb=odb)
 
+        obj = self.get_obj()
         checkout(
             to.path_info,
             to.fs,
             obj,
-            odb,
+            self.repo.odb.local,
             dvcignore=None,
             state=self.repo.state,
         )
@@ -110,7 +81,6 @@ class RepoDependency(Dependency):
     def update(self, rev=None):
         if rev:
             self.def_repo[self.PARAM_REV] = rev
-
         with self._make_repo(locked=False) as repo:
             self.def_repo[self.PARAM_REV_LOCK] = repo.get_rev()
 
@@ -119,3 +89,124 @@ class RepoDependency(Dependency):
         # origin project url and rev_lock, and it makes RepoDependency
         # immutable, hence its impossible for checksum to change.
         return False
+
+    def get_used_objs(
+        self, **kwargs
+    ) -> Dict[Optional["ObjectDB"], Set["HashInfo"]]:
+        used, _ = self._get_used_and_obj(**kwargs)
+        return used
+
+    def _get_used_and_obj(
+        self, obj_only=False, **kwargs
+    ) -> Tuple[Dict[Optional["ObjectDB"], Set["HashInfo"]], "HashFile"]:
+        from dvc.config import NoRemoteError
+        from dvc.exceptions import NoOutputOrStageError, PathMissingError
+        from dvc.objects.stage import stage
+        from dvc.objects.tree import Tree
+
+        local_odb = self.repo.odb.local
+        locked = kwargs.pop("locked", True)
+        with self._make_repo(
+            locked=locked, cache_dir=local_odb.cache_dir
+        ) as repo:
+            used_obj_ids = defaultdict(set)
+            rev = repo.get_rev()
+            if locked and self.def_repo.get(self.PARAM_REV_LOCK) is None:
+                self.def_repo[self.PARAM_REV_LOCK] = rev
+
+            path_info = PathInfo(repo.root_dir) / str(self.def_path)
+            if not obj_only:
+                try:
+                    for odb, obj_ids in repo.used_objs(
+                        [os.fspath(path_info)],
+                        force=True,
+                        jobs=kwargs.get("jobs"),
+                        recursive=True,
+                    ).items():
+                        if odb is None:
+                            odb = repo.cloud.get_remote_odb()
+                            odb.read_only = True
+                        self._check_circular_import(odb, obj_ids)
+                        used_obj_ids[odb].update(obj_ids)
+                except (NoRemoteError, NoOutputOrStageError):
+                    pass
+
+            try:
+                staging, staged_obj = stage(
+                    local_odb,
+                    path_info,
+                    repo.repo_fs,
+                    local_odb.fs.PARAM_CHECKSUM,
+                )
+            except FileNotFoundError as exc:
+                raise PathMissingError(
+                    self.def_path, self.def_repo[self.PARAM_URL]
+                ) from exc
+            staging = copy(staging)
+            staging.read_only = True
+
+            self._staged_objs[rev] = staged_obj
+            used_obj_ids[staging].add(staged_obj.hash_info)
+            if isinstance(staged_obj, Tree):
+                used_obj_ids[staging].update(
+                    entry.hash_info for _, entry in staged_obj
+                )
+            return used_obj_ids, staged_obj
+
+    def _check_circular_import(self, odb, obj_ids):
+        from dvc.exceptions import CircularImportError
+        from dvc.fs.repo import RepoFileSystem
+        from dvc.objects.db.reference import ReferenceObjectDB
+        from dvc.objects.tree import Tree
+
+        if not isinstance(odb, ReferenceObjectDB):
+            return
+
+        def iter_objs():
+            for hash_info in obj_ids:
+                if hash_info.isdir:
+                    tree = Tree.load(odb, hash_info)
+                    yield from (odb.get(entry.hash_info) for _, entry in tree)
+                else:
+                    yield odb.get(hash_info)
+
+        checked_urls = set()
+        for obj in iter_objs():
+            if not isinstance(obj.fs, RepoFileSystem):
+                continue
+            if (
+                obj.fs.repo_url in checked_urls
+                or obj.fs.root_dir in checked_urls
+            ):
+                continue
+            self_url = self.repo.url or self.repo.root_dir
+            if (
+                obj.fs.repo_url is not None
+                and obj.fs.repo_url == self_url
+                or obj.fs.root_dir == self.repo.root_dir
+            ):
+                raise CircularImportError(self, obj.fs.repo_url, self_url)
+            checked_urls.update([obj.fs.repo_url, obj.fs.root_dir])
+
+    def get_obj(self, filter_info=None, **kwargs):
+        locked = kwargs.get("locked", True)
+        rev = self._get_rev(locked=locked)
+        if rev in self._staged_objs:
+            return self._staged_objs[rev]
+        _, obj = self._get_used_and_obj(
+            obj_only=True, filter_info=filter_info, **kwargs
+        )
+        return obj
+
+    def _make_repo(self, locked=True, **kwargs):
+        from dvc.external_repo import external_repo
+
+        d = self.def_repo
+        rev = self._get_rev(locked=locked)
+        return external_repo(d[self.PARAM_URL], rev=rev, **kwargs)
+
+    def _get_rev(self, locked=True):
+        d = self.def_repo
+        return (d.get(self.PARAM_REV_LOCK) if locked else None) or d.get(
+            self.PARAM_REV
+        )

dvc/dvcfile.py

@@ -200,9 +200,7 @@ class SingleStageFile(FileMixin):
         assert not isinstance(stage, PipelineStage)
         if self.verify:
             check_dvcfile_path(self.repo, self.path)
-        logger.debug(
-            "Saving information to '{file}'.".format(file=relpath(self.path))
-        )
+        logger.debug(f"Saving information to '{relpath(self.path)}'.")
         dump_yaml(self.path, serialize.to_single_stage_file(stage))
         self.repo.scm.track_file(self.relpath)
 
@@ -300,7 +298,7 @@ class PipelineFile(FileMixin):
         if not self.exists():
             return
 
-        with open(self.path, "r", encoding="utf-8") as f:
+        with open(self.path, encoding="utf-8") as f:
             d = parse_yaml_for_update(f.read(), self.path)
 
         self.validate(d, self.path)

dvc/env.py

@@ -6,3 +6,7 @@ DVCLIVE_PATH = "DVCLIVE_PATH"
 DVCLIVE_SUMMARY = "DVCLIVE_SUMMARY"
 DVCLIVE_HTML = "DVCLIVE_HTML"
 DVCLIVE_RESUME = "DVCLIVE_RESUME"
+DVC_IGNORE_ISATTY = "DVC_IGNORE_ISATTY"
+DVC_EXP_GIT_REMOTE = "DVC_EXP_GIT_REMOTE"
+DVC_EXP_AUTO_PUSH = "DVC_EXP_AUTO_PUSH"
+DVC_NO_ANALYTICS = "DVC_NO_ANALYTICS"

dvc/exceptions.py

@@ -1,5 +1,4 @@
 """Exceptions raised by the dvc."""
-from typing import List
 
 
 class DvcException(Exception):
@@ -173,36 +172,6 @@ class BadMetricError(DvcException):
         )
 
 
-class MetricsError(DvcException):
-    pass
-
-
-class NoMetricsParsedError(MetricsError):
-    def __init__(self, command):
-        super().__init__(
-            f"Could not parse {command} files. Use `-v` option to see more "
-            "details."
-        )
-
-
-class NoMetricsFoundError(MetricsError):
-    def __init__(self, command, run_options):
-        super().__init__(
-            f"No {command} files in this repository. "
-            f"Use `{run_options}` options for "
-            f"`dvc run` to mark stage outputs as {command}."
-        )
-
-
-class MetricDoesNotExistError(MetricsError):
-    def __init__(self, targets: List[str]):
-        if len(targets) == 1:
-            msg = "'{}' does not exist."
-        else:
-            msg = "'{}' do not exist."
-        super().__init__(msg.format(", ".join(targets)))
-
-
 class RecursiveAddingWhileUsingFilename(DvcException):
     def __init__(self):
         super().__init__(
@@ -259,18 +228,21 @@ class GitHookAlreadyExistsError(DvcException):
         )
 
 
-class DownloadError(DvcException):
+class FileTransferError(DvcException):
+    _METHOD = "transfer"
+
     def __init__(self, amount):
         self.amount = amount
 
-        super().__init__(f"{amount} files failed to download")
+        super().__init__(f"{amount} files failed to {self._METHOD}")
 
 
-class UploadError(DvcException):
-    def __init__(self, amount):
-        self.amount = amount
+class DownloadError(FileTransferError):
+    _METHOD = "download"
 
-        super().__init__(f"{amount} files failed to upload")
+
+class UploadError(FileTransferError):
+    _METHOD = "upload"
 
 
 class CheckoutError(DvcException):
@@ -384,3 +356,11 @@ class CacheLinkError(DvcException):
         )
         super().__init__(msg)
         self.path_infos = path_infos
+
+
+class CircularImportError(DvcException):
+    def __init__(self, dep, a, b):
+        super().__init__(
+            f"'{dep}' contains invalid circular import. "
+            f"DVC repo '{a}' already imports from '{b}'."
+        )

dvc/external_repo.py

@@ -42,14 +42,6 @@ def external_repo(
     config = _get_remote_config(url) if os.path.isdir(url) else {}
     config.update(cache_config)
 
-    def make_repo(path, **_kwargs):
-        _config = cache_config.copy()
-        if os.path.isdir(url):
-            rel = os.path.relpath(path, _kwargs["scm"].root_dir)
-            repo_path = os.path.join(url, rel)
-            _config.update(_get_remote_config(repo_path))
-        return Repo(path, config=_config, **_kwargs)
-
     root_dir = path if for_write else os.path.realpath(path)
     repo_kwargs = dict(
         root_dir=root_dir,
@@ -57,7 +49,7 @@ def external_repo(
         scm=None if for_write else Git(root_dir),
         rev=None if for_write else rev,
         config=config,
-        repo_factory=make_repo,
+        repo_factory=erepo_factory(url, cache_config),
         **kwargs,
     )
 
@@ -87,6 +79,18 @@ def external_repo(
             _remove(path)
 
 
+def erepo_factory(url, cache_config, *args, **kwargs):
+    def make_repo(path, **_kwargs):
+        _config = cache_config.copy()
+        if os.path.isdir(url):
+            rel = os.path.relpath(path, _kwargs["scm"].root_dir)
+            repo_path = os.path.join(url, rel)
+            _config.update(_get_remote_config(repo_path))
+        return Repo(path, config=_config, **_kwargs)
+
+    return make_repo
+
+
 CLONES: Dict[str, str] = {}
 CACHE_DIRS: Dict[str, str] = {}
 

dvc/fs/__init__.py

@@ -12,8 +12,7 @@ from .local import LocalFileSystem
 from .oss import OSSFileSystem
 from .s3 import S3FileSystem
 from .ssh import SSHFileSystem
-from .webdav import WebDAVFileSystem
-from .webdavs import WebDAVSFileSystem
+from .webdav import WebDAVFileSystem, WebDAVSFileSystem
 from .webhdfs import WebHDFSFileSystem
 
 FS_MAP = {
@@ -33,15 +32,21 @@ FS_MAP = {
 }
 
 
-def get_fs_cls(remote_conf):
-    scheme = urlparse(remote_conf["url"]).scheme
+def get_fs_cls(remote_conf, scheme=None):
+    if not scheme:
+        scheme = urlparse(remote_conf["url"]).scheme
     return FS_MAP.get(scheme, LocalFileSystem)
 
 
 def get_fs_config(config, **kwargs):
     name = kwargs.get("name")
     if name:
-        remote_conf = config["remote"][name.lower()]
+        try:
+            remote_conf = config["remote"][name.lower()]
+        except KeyError:
+            from dvc.config import RemoteNotFoundError
+
+            raise RemoteNotFoundError(f"remote '{name}' doesn't exist")
     else:
         remote_conf = kwargs
     return _resolve_remote_refs(config, remote_conf)
@@ -106,4 +111,11 @@ def get_cloud_fs(repo, **kwargs):
     if cls == GDriveFileSystem and repo:
         remote_conf["gdrive_credentials_tmp_dir"] = repo.tmp_dir
 
-    return cls, remote_conf
+    url = remote_conf.pop("url")
+    path_info = cls.PATH_CLS(
+        cls._strip_protocol(url)  # pylint:disable=protected-access
+    )
+
+    extras = cls._get_kwargs_from_urls(url)  # pylint:disable=protected-access
+    conf = {**extras, **remote_conf}  # remote config takes priority
+    return cls, conf, path_info

dvc/fs/azure.py

@@ -1,11 +1,10 @@
-import asyncio
 import logging
 import os
-import sys
 import threading
-from contextlib import contextmanager
 
-from funcy import cached_property, wrap_prop
+from fsspec.asyn import fsspec_loop
+from fsspec.utils import infer_storage_options
+from funcy import cached_property, memoize, wrap_prop
 
 from dvc.exceptions import DvcException
 from dvc.path_info import CloudURLInfo
@@ -22,51 +21,23 @@ _DEFAULT_CREDS_STEPS = (
 )
 
 
-@contextmanager
-def _temp_event_loop():
-    """When trying to initalize azure filesystem instances
-    with DefaultCredentials, the authentication process requires
-    to have an access to a separate event loop. The normal calls
-    are run in a separate loop managed by the fsspec, but the
-    DefaultCredentials assumes that the callee is managing their
-    own event loop. This function checks whether is there any
-    event loop set, and if not it creates a temporary event loop
-    and set it. After the context is finalized, it restores the
-    original event loop back (if there is any)."""
-
-    try:
-        original_loop = asyncio.get_event_loop()
-        original_policy = asyncio.get_event_loop_policy()
-    except RuntimeError:
-        original_loop = None
-        original_policy = None
-
-    # From 3.8>= and onwards, asyncio changed the default
-    # loop policy for windows to use proactor loops instead
-    # of selector based ones. Due to that, proxied connections
-    # doesn't work with the aiohttp and this is most likely an
-    # upstream bug that needs to be solved outside of DVC. Until
-    # such issue is resolved, we need to manage this;
-    # https://github.com/aio-libs/aiohttp/issues/4536
-    if sys.version_info >= (3, 8) and os.name == "nt":
-        asyncio.set_event_loop_policy(asyncio.WindowsSelectorEventLoopPolicy())
-
-    loop = original_loop or asyncio.new_event_loop()
-
-    try:
-        asyncio.set_event_loop(loop)
-        yield
-    finally:
-        if original_loop is None:
-            loop.close()
-        asyncio.set_event_loop(original_loop)
-        asyncio.set_event_loop_policy(original_policy)
-
-
 class AzureAuthError(DvcException):
     pass
 
 
+@memoize
+def _az_config():
+    # NOTE: ideally we would've used get_default_cli().config from
+    # azure.cli.core, but azure-cli-core has a lot of conflicts with other
+    # dependencies. So instead we are just use knack directly
+    from knack.config import CLIConfig
+
+    config_dir = os.getenv(
+        "AZURE_CONFIG_DIR", os.path.expanduser(os.path.join("~", ".azure"))
+    )
+    return CLIConfig(config_dir=config_dir, config_env_var_prefix="AZURE")
+
+
 # pylint:disable=abstract-method
 class AzureFileSystem(ObjectFSWrapper):
     scheme = Schemes.AZURE
@@ -79,19 +50,21 @@ class AzureFileSystem(ObjectFSWrapper):
         "azure-identity": "azure.identity",
     }
 
-    def __init__(self, **config):
-        super().__init__(**config)
+    @classmethod
+    def _strip_protocol(cls, path: str):
+        bucket = infer_storage_options(path).get("host")
+        if bucket:
+            return path
 
-        url = config.get("url")
-        self.path_info = self.PATH_CLS(url)
-        self.bucket = self.path_info.bucket
+        bucket = _az_config().get("storage", "container_name", None)
+        return f"azure://{bucket}"
 
-        if not self.bucket:
-            container = self._az_config.get("storage", "container_name", None)
-            url = f"azure://{container}"
-
-        self.path_info = self.PATH_CLS(url)
-        self.bucket = self.path_info.bucket
+    @staticmethod
+    def _get_kwargs_from_urls(urlpath):
+        ops = infer_storage_options(urlpath)
+        if "host" in ops:
+            return {"bucket": ops["host"]}
+        return {}
 
     def _prepare_credentials(self, **config):
         from azure.identity.aio import DefaultAzureCredential
@@ -102,28 +75,54 @@ class AzureFileSystem(ObjectFSWrapper):
         login_info = {}
         login_info["connection_string"] = config.get(
             "connection_string",
-            self._az_config.get("storage", "connection_string", None),
+            _az_config().get("storage", "connection_string", None),
         )
         login_info["account_name"] = config.get(
-            "account_name", self._az_config.get("storage", "account", None)
+            "account_name", _az_config().get("storage", "account", None)
         )
         login_info["account_key"] = config.get(
-            "account_key", self._az_config.get("storage", "key", None)
+            "account_key", _az_config().get("storage", "key", None)
         )
         login_info["sas_token"] = config.get(
-            "sas_token", self._az_config.get("storage", "sas_token", None)
+            "sas_token", _az_config().get("storage", "sas_token", None)
         )
         login_info["tenant_id"] = config.get("tenant_id")
         login_info["client_id"] = config.get("client_id")
         login_info["client_secret"] = config.get("client_secret")
 
-        if not any(
-            value for key, value in login_info.items() if key != "account_name"
-        ):
-            login_info["credential"] = DefaultAzureCredential(
-                exclude_interactive_browser_credential=False
+        if not (login_info["account_name"] or login_info["connection_string"]):
+            raise AzureAuthError(
+                "Authentication to Azure Blob Storage requires either "
+                "account_name or connection_string.\nLearn more about "
+                "configuration settings at "
+                + format_link("https://man.dvc.org/remote/modify")
             )
 
+        any_secondary = any(
+            value for key, value in login_info.items() if key != "account_name"
+        )
+        if (
+            login_info["account_name"]
+            and not any_secondary
+            and not config.get("allow_anonymous_login", False)
+        ):
+            with fsspec_loop():
+                login_info["credential"] = DefaultAzureCredential(
+                    exclude_interactive_browser_credential=False,
+                    exclude_environment_credential=config.get(
+                        "exclude_environment_credential", False
+                    ),
+                    exclude_visual_studio_code_credential=config.get(
+                        "exclude_visual_studio_code_credential", False
+                    ),
+                    exclude_shared_token_cache_credential=config.get(
+                        "exclude_shared_token_cache_credential", False
+                    ),
+                    exclude_managed_identity_credential=config.get(
+                        "exclude_managed_identity_credential", False
+                    ),
+                )
+
         for login_method, required_keys in [  # noqa
             ("connection string", ["connection_string"]),
             (
@@ -146,18 +145,6 @@ class AzureFileSystem(ObjectFSWrapper):
         self.login_method = login_method
         return login_info
 
-    @cached_property
-    def _az_config(self):
-        # NOTE: ideally we would've used get_default_cli().config from
-        # azure.cli.core, but azure-cli-core has a lot of conflicts with other
-        # dependencies. So instead we are just use knack directly
-        from knack.config import CLIConfig
-
-        config_dir = os.getenv(
-            "AZURE_CONFIG_DIR", os.path.expanduser(os.path.join("~", ".azure"))
-        )
-        return CLIConfig(config_dir=config_dir, config_env_var_prefix="AZURE")
-
     @wrap_prop(threading.Lock())
     @cached_property
     def fs(self):
@@ -165,23 +152,10 @@ class AzureFileSystem(ObjectFSWrapper):
         from azure.core.exceptions import AzureError
 
         try:
-            with _temp_event_loop():
-                file_system = AzureBlobFileSystem(**self.fs_args)
-                if self.bucket not in [
-                    container.rstrip("/") for container in file_system.ls("/")
-                ]:
-                    file_system.mkdir(self.bucket)
+            return AzureBlobFileSystem(**self.fs_args)
         except (ValueError, AzureError) as e:
             raise AzureAuthError(
                 f"Authentication to Azure Blob Storage via {self.login_method}"
                 " failed.\nLearn more about configuration settings at"
                 f" {format_link('https://man.dvc.org/remote/modify')}"
             ) from e
-
-        return file_system
-
-    def open(
-        self, path_info, mode="r", **kwargs
-    ):  # pylint: disable=arguments-differ
-        with _temp_event_loop():
-            return self.fs.open(self._with_bucket(path_info), mode=mode)

dvc/fs/base.py

@@ -33,6 +33,8 @@ class RemoteMissingDepsError(DvcException):
 
 
 class BaseFileSystem:
+    sep = "/"
+
     scheme = "base"
     REQUIRES: ClassVar[Dict[str, str]] = {}
     PATH_CLS = URLInfo  # type: Any
@@ -55,10 +57,21 @@ class BaseFileSystem:
     def __init__(self, **kwargs):
         self._check_requires(**kwargs)
 
-        self.path_info = None
-
         self.jobs = kwargs.get("jobs") or self._JOBS
         self.hash_jobs = kwargs.get("checksum_jobs") or self.HASH_JOBS
+        self._config = kwargs
+
+    @property
+    def config(self):
+        return self._config
+
+    @classmethod
+    def _strip_protocol(cls, path: str):
+        return path
+
+    @staticmethod
+    def _get_kwargs_from_urls(urlpath):  # pylint:disable=unused-argument
+        return {}
 
     @classmethod
     def get_missing_deps(cls):
@@ -164,7 +177,7 @@ class BaseFileSystem:
     def ls(self, path_info, detail=False):
         raise RemoteActionNotImplemented("ls", self.scheme)
 
-    def find(self, path_info, detail=False):
+    def find(self, path_info, detail=False, prefix=None):
         raise RemoteActionNotImplemented("find", self.scheme)
 
     def is_empty(self, path_info):
@@ -179,7 +192,7 @@ class BaseFileSystem:
     def remove(self, path_info):
         raise RemoteActionNotImplemented("remove", self.scheme)
 
-    def makedirs(self, path_info):
+    def makedirs(self, path_info, **kwargs):
         """Optional: Implement only if the remote needs to create
         directories before copying/linking/moving data
         """
@@ -229,14 +242,23 @@ class BaseFileSystem:
             no_progress_bar=no_progress_bar,
         )
 
-    def upload_fobj(self, fobj, to_info, no_progress_bar=False, **pbar_args):
+    def upload_fobj(
+        self, fobj, to_info, no_progress_bar=False, size=None, **pbar_args
+    ):
         if not hasattr(self, "_upload_fobj"):
             raise RemoteActionNotImplemented("upload_fobj", self.scheme)
 
         with Tqdm.wrapattr(
-            fobj, "read", disable=no_progress_bar, bytes=True, **pbar_args
+            fobj,
+            "read",
+            disable=no_progress_bar,
+            bytes=True,
+            total=size,
+            **pbar_args,
         ) as wrapped:
-            self._upload_fobj(wrapped, to_info)  # pylint: disable=no-member
+            self._upload_fobj(  # pylint: disable=no-member
+                wrapped, to_info, size=size
+            )
 
     def download(
         self,

dvc/fs/dvc.py

@@ -23,6 +23,8 @@ class DvcFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
         repo: DVC repo.
     """
 
+    sep = os.sep
+
     scheme = "local"
     PARAM_CHECKSUM = "md5"
 
@@ -30,6 +32,10 @@ class DvcFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
         super().__init__(**kwargs)
         self.repo = kwargs["repo"]
 
+    @property
+    def config(self):
+        raise NotImplementedError
+
     def _find_outs(self, path, *args, **kwargs):
         outs = self.repo.find_outs_by_path(path, *args, **kwargs)
 
@@ -49,14 +55,14 @@ class DvcFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
         # NOTE: use string paths here for performance reasons
         key = tuple(relpath(path_info, out.path_info).split(os.sep))
         out.get_dir_cache(remote=remote)
+        if out.obj is None:
+            raise FileNotFoundError
         obj = out.obj.trie.get(key)
         if obj:
             return obj.hash_info
         raise FileNotFoundError
 
-    def open(  # type: ignore
-        self, path: PathInfo, mode="r", encoding=None, remote=None, **kwargs
-    ):  # pylint: disable=arguments-differ
+    def _get_fs_path(self, path: PathInfo, remote=None):
         try:
             outs = self._find_outs(path, strict=False)
         except OutputNotFoundError as exc:
@@ -76,33 +82,37 @@ class DvcFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
             from dvc.config import NoRemoteError
 
             try:
-                remote_obj = self.repo.cloud.get_remote(remote)
+                remote_odb = self.repo.cloud.get_remote_odb(remote)
             except NoRemoteError as exc:
                 raise FileNotFoundError from exc
             if out.is_dir_checksum:
                 checksum = self._get_granular_hash(path, out).value
             else:
                 checksum = out.hash_info.value
-            remote_info = remote_obj.odb.hash_to_path_info(checksum)
-            return remote_obj.fs.open(
-                remote_info, mode=mode, encoding=encoding
-            )
+            remote_info = remote_odb.hash_to_path_info(checksum)
+            return remote_odb.fs, remote_info
 
         if out.is_dir_checksum:
             checksum = self._get_granular_hash(path, out).value
             cache_path = out.odb.hash_to_path_info(checksum).url
         else:
             cache_path = out.cache_path
-        return open(cache_path, mode=mode, encoding=encoding)
+        return out.odb.fs, cache_path
 
-    def exists(self, path):  # pylint: disable=arguments-differ
+    def open(  # type: ignore
+        self, path: PathInfo, mode="r", encoding=None, **kwargs
+    ):  # pylint: disable=arguments-renamed
+        fs, fspath = self._get_fs_path(path, **kwargs)
+        return fs.open(fspath, mode=mode, encoding=encoding)
+
+    def exists(self, path):  # pylint: disable=arguments-renamed
         try:
             self.metadata(path)
             return True
         except FileNotFoundError:
             return False
 
-    def isdir(self, path):  # pylint: disable=arguments-differ
+    def isdir(self, path):  # pylint: disable=arguments-renamed
         try:
             meta = self.metadata(path)
             return meta.isdir
@@ -125,7 +135,7 @@ class DvcFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
         except FileNotFoundError:
             return True
 
-    def isfile(self, path):  # pylint: disable=arguments-differ
+    def isfile(self, path):  # pylint: disable=arguments-renamed
         try:
             meta = self.metadata(path)
             return meta.isfile
@@ -231,7 +241,7 @@ class DvcFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
 
     def info(self, path_info):
         meta = self.metadata(path_info)
-        ret = {"type": "dir" if meta.isdir else "file"}
+        ret = {"type": "directory" if meta.isdir else "file"}
         if meta.is_output and len(meta.outs) == 1 and meta.outs[0].hash_info:
             hash_info = meta.outs[0].hash_info
             ret["size"] = hash_info.size
@@ -245,3 +255,9 @@ class DvcFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
                 ret[obj.hash_info.name] = obj.hash_info.value
 
         return ret
+
+    def _download(self, from_info, to_file, **kwargs):
+        fs, path = self._get_fs_path(from_info)
+        fs._download(  # pylint: disable=protected-access
+            path, to_file, **kwargs
+        )

dvc/fs/fsspec_wrapper.py

@@ -7,10 +7,13 @@ from funcy import cached_property
 from dvc.progress import Tqdm
 
 from .base import BaseFileSystem
+from .local import LocalFileSystem
 
 
 # pylint: disable=no-member
 class FSSpecWrapper(BaseFileSystem):
+    TRAVERSE_PREFIX_LEN = 2
+
     def __init__(self, **kwargs):
         super().__init__(**kwargs)
         self.fs_args = {"skip_instance_cache": True}
@@ -20,21 +23,20 @@ class FSSpecWrapper(BaseFileSystem):
     def fs(self):
         raise NotImplementedError
 
-    @lru_cache(512)
     def _with_bucket(self, path):
-        if isinstance(path, self.PATH_CLS):
-            return f"{path.bucket}/{path.path}"
-        return path
+        return str(path)
 
     def _strip_bucket(self, entry):
-        try:
-            bucket, path = entry.split("/", 1)
-        except ValueError:
-            # If there is no path attached, only returns
-            # the bucket (top-level).
-            bucket, path = entry, None
+        return entry
 
-        return path or bucket
+    @staticmethod
+    def _get_kwargs_from_urls(urlpath):
+        from fsspec.utils import infer_storage_options
+
+        options = infer_storage_options(urlpath)
+        options.pop("path", None)
+        options.pop("protocol", None)
+        return options
 
     def _strip_buckets(self, entries, detail=False):
         for entry in entries:
@@ -46,7 +48,7 @@ class FSSpecWrapper(BaseFileSystem):
             yield entry
 
     def _entry_hook(self, entry):
-        """Simple hook method to be overriden when wanted to process
+        """Simple hook method to be overridden when wanted to process
         entries within info() and ls(detail=True) calls"""
         return entry
 
@@ -84,6 +86,7 @@ class FSSpecWrapper(BaseFileSystem):
         return self.fs.open(self._with_bucket(path_info), mode=mode)
 
     def copy(self, from_info, to_info):
+        self.makedirs(to_info.parent)
         self.fs.copy(self._with_bucket(from_info), self._with_bucket(to_info))
 
     def exists(self, path_info) -> bool:
@@ -94,7 +97,8 @@ class FSSpecWrapper(BaseFileSystem):
         files = self.fs.ls(path, detail=detail)
         yield from self._strip_buckets(files, detail=detail)
 
-    def find(self, path_info, detail=False):
+    # pylint: disable=unused-argument
+    def find(self, path_info, detail=False, prefix=None):
         path = self._with_bucket(path_info)
         files = self.fs.find(path, detail=detail)
         if detail:
@@ -103,11 +107,14 @@ class FSSpecWrapper(BaseFileSystem):
         yield from self._strip_buckets(files, detail=detail)
 
     def walk_files(self, path_info, **kwargs):
-        for file in self.find(path_info):
+        for file in self.find(path_info, **kwargs):
             yield path_info.replace(path=file)
 
+    def move(self, from_info, to_info):
+        self.fs.move(self._with_bucket(from_info), self._with_bucket(to_info))
+
     def remove(self, path_info):
-        self.fs.rm(self._with_bucket(path_info))
+        self.fs.rm_file(self._with_bucket(path_info))
 
     def info(self, path_info):
         info = self.fs.info(self._with_bucket(path_info)).copy()
@@ -115,20 +122,27 @@ class FSSpecWrapper(BaseFileSystem):
         info["name"] = self._strip_bucket(info["name"])
         return info
 
-    def _upload_fobj(self, fobj, to_info):
+    def makedirs(self, path_info, **kwargs):
+        self.fs.makedirs(
+            self._with_bucket(path_info), exist_ok=kwargs.pop("exist_ok", True)
+        )
+
+    def _upload_fobj(self, fobj, to_info, size=None):
+        self.makedirs(to_info.parent)
         with self.open(to_info, "wb") as fdest:
             shutil.copyfileobj(fobj, fdest, length=fdest.blocksize)
 
     def _upload(
         self, from_file, to_info, name=None, no_progress_bar=False, **kwargs
     ):
-        total = os.path.getsize(from_file)
+        self.makedirs(to_info.parent)
+        size = os.path.getsize(from_file)
         with open(from_file, "rb") as fobj:
             self.upload_fobj(
                 fobj,
-                self._with_bucket(to_info),
+                to_info,
+                size=size,
                 desc=name,
-                total=total,
                 no_progress_bar=no_progress_bar,
             )
         self.fs.invalidate_cache(self._with_bucket(to_info.parent))
@@ -153,6 +167,31 @@ class FSSpecWrapper(BaseFileSystem):
 
 # pylint: disable=abstract-method
 class ObjectFSWrapper(FSSpecWrapper):
+    TRAVERSE_PREFIX_LEN = 3
+
+    @lru_cache(512)
+    def _with_bucket(self, path):
+        if isinstance(path, self.PATH_CLS):
+            return f"{path.bucket}/{path.path}"
+        return path
+
+    def _strip_bucket(self, entry):
+        try:
+            bucket, path = entry.split("/", 1)
+        except ValueError:
+            # If there is no path attached, only returns
+            # the bucket (top-level).
+            bucket, path = entry, None
+        return path or bucket
+
+    def makedirs(self, path_info, **kwargs):
+        # For object storages make this method a no-op. The original
+        # fs.makedirs() method will only check if the bucket exists
+        # and create if it doesn't though we don't want to support
+        # that behavior, and the check will cost some time so we'll
+        # simply ignore all mkdir()/makedirs() calls.
+        return None
+
     def _isdir(self, path_info):
         # Directory in object storages are interpreted differently
         # among different fsspec providers, so this logic is a temporary
@@ -167,11 +206,18 @@ class ObjectFSWrapper(FSSpecWrapper):
             and entry["name"].endswith("/")
         )
 
-    def find(self, path_info, detail=False):
-        path = self._with_bucket(path_info)
-        files = self.fs.find(path, detail=detail)
+    def find(self, path_info, detail=False, prefix=None):
+        if prefix:
+            path = self._with_bucket(path_info.parent)
+            files = self.fs.find(
+                path, detail=detail, prefix=path_info.parts[-1]
+            )
+        else:
+            path = self._with_bucket(path_info)
+            files = self.fs.find(path, detail=detail)
+
         if detail:
-            files = files.values()
+            files = list(files.values())
 
         # When calling find() on a file, it returns the same file in a list.
         # For object-based storages, the same behavior applies to empty
@@ -182,3 +228,44 @@ class ObjectFSWrapper(FSSpecWrapper):
             return None
 
         yield from self._strip_buckets(files, detail=detail)
+
+
+_LOCAL_FS = LocalFileSystem()
+
+
+class CallbackMixin:
+    """Use the native ``get_file()``/``put_file()`` APIs
+    if the target filesystem supports callbacks."""
+
+    def _upload(
+        self, from_file, to_info, name=None, no_progress_bar=False, **pbar_args
+    ):
+        with Tqdm(
+            desc=name,
+            disable=no_progress_bar,
+            bytes=True,
+            total=-1,
+            **pbar_args,
+        ) as pbar:
+            self.fs.put_file(
+                os.fspath(from_file),
+                self._with_bucket(to_info),
+                callback=pbar.as_callback(_LOCAL_FS, from_file),
+            )
+        self.fs.invalidate_cache(self._with_bucket(to_info.parent))
+
+    def _download(
+        self, from_info, to_file, name=None, no_progress_bar=False, **pbar_args
+    ):
+        with Tqdm(
+            desc=name,
+            disable=no_progress_bar,
+            bytes=True,
+            total=-1,
+            **pbar_args,
+        ) as pbar:
+            self.fs.get_file(
+                self._with_bucket(from_info),
+                os.fspath(to_file),
+                callback=pbar.as_callback(self, from_info),
+            )

dvc/fs/gdrive.py

@@ -1,24 +1,18 @@
-import io
 import logging
 import os
 import posixpath
 import re
 import threading
-from collections import defaultdict
-from contextlib import contextmanager
 from urllib.parse import urlparse
 
-from funcy import cached_property, retry, wrap_prop, wrap_with
-from funcy.py3 import cat
+from funcy import cached_property, retry, wrap_prop
 
-from dvc.exceptions import DvcException, FileMissingError
+from dvc.exceptions import DvcException
 from dvc.path_info import CloudURLInfo
-from dvc.progress import Tqdm
 from dvc.scheme import Schemes
 from dvc.utils import format_link, tmp_fname
-from dvc.utils.stream import IterStream
 
-from .base import BaseFileSystem
+from .fsspec_wrapper import CallbackMixin, FSSpecWrapper
 
 logger = logging.getLogger(__name__)
 FOLDER_MIME_TYPE = "application/vnd.google-apps.folder"
@@ -85,7 +79,9 @@ class GDriveURLInfo(CloudURLInfo):
         self._spath = re.sub("/{2,}", "/", self._spath.rstrip("/"))
 
 
-class GDriveFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
+class GDriveFileSystem(
+    CallbackMixin, FSSpecWrapper
+):  # pylint:disable=abstract-method
     scheme = Schemes.GDRIVE
     PATH_CLS = GDriveURLInfo
     PARAM_CHECKSUM = "checksum"
@@ -142,6 +138,10 @@ class GDriveFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
             )
         )
 
+    @staticmethod
+    def _get_kwargs_from_urls(urlpath):
+        return {"url": urlpath}
+
     def _validate_config(self):
         # Validate Service Account configuration
         if (
@@ -209,9 +209,9 @@ class GDriveFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
 
     @wrap_prop(threading.RLock())
     @cached_property
-    def _drive(self):
+    def fs(self):
         from pydrive2.auth import GoogleAuth
-        from pydrive2.drive import GoogleDrive
+        from pydrive2.fs import GDriveFileSystem as _GDriveFileSystem
 
         temporary_save_path = self._gdrive_user_credentials_path
         is_credentials_temp = os.getenv(
@@ -286,352 +286,19 @@ class GDriveFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
             if is_credentials_temp:
                 os.remove(temporary_save_path)
 
-        return GoogleDrive(gauth)
-
-    @wrap_prop(threading.RLock())
-    @cached_property
-    def _ids_cache(self):
-        cache = {
-            "dirs": defaultdict(list),
-            "ids": {},
-            "root_id": self._get_item_id(
-                self.path_info,
-                use_cache=False,
-                hint="Confirm the directory exists and you can access it.",
-            ),
-        }
-
-        self._cache_path_id(self._path, cache["root_id"], cache=cache)
-
-        for item in self._gdrive_list(
-            "'{}' in parents and trashed=false".format(cache["root_id"])
-        ):
-            item_path = (self.path_info / item["title"]).path
-            self._cache_path_id(item_path, item["id"], cache=cache)
-
-        return cache
-
-    def _cache_path_id(self, path, *item_ids, cache=None):
-        cache = cache or self._ids_cache
-        for item_id in item_ids:
-            cache["dirs"][path].append(item_id)
-            cache["ids"][item_id] = path
-
-    @cached_property
-    def _list_params(self):
-        params = {"corpora": "default"}
-        if self._bucket != "root" and self._bucket != "appDataFolder":
-            drive_id = self._gdrive_shared_drive_id(self._bucket)
-            if drive_id:
-                logger.debug(
-                    "GDrive remote '{}' is using shared drive id '{}'.".format(
-                        self.path_info, drive_id
-                    )
-                )
-                params["driveId"] = drive_id
-                params["corpora"] = "drive"
-        return params
-
-    @_gdrive_retry
-    def _gdrive_shared_drive_id(self, item_id):
-        from pydrive2.files import ApiRequestError
-
-        param = {"id": item_id}
-        # it does not create a file on the remote
-        item = self._drive.CreateFile(param)
-        # ID of the shared drive the item resides in.
-        # Only populated for items in shared drives.
-        try:
-            item.FetchMetadata("driveId")
-        except ApiRequestError as exc:
-            error_code = exc.error.get("code", 0)
-            if error_code == 404:
-                raise DvcException(
-                    "'{}' for '{}':\n\n"
-                    "1. Confirm the directory exists and you can access it.\n"
-                    "2. Make sure that credentials in '{}'\n"
-                    "   are correct for this remote e.g. "
-                    "use the `gdrive_user_credentials_file` config\n"
-                    "   option if you use multiple GDrive remotes with "
-                    "different email accounts.\n\nDetails".format(
-                        item_id, self.path_info, self.credentials_location
-                    )
-                ) from exc
-            raise
-
-        return item.get("driveId", None)
-
-    @_gdrive_retry
-    def _gdrive_upload_fobj(self, fobj, parent_id, title):
-        item = self._drive.CreateFile(
-            {"title": title, "parents": [{"id": parent_id}]}
-        )
-        item.content = fobj
-        item.Upload()
-        return item
-
-    @_gdrive_retry
-    def _gdrive_download_file(
-        self, item_id, to_file, progress_desc, no_progress_bar
-    ):
-        param = {"id": item_id}
-        # it does not create a file on the remote
-        gdrive_file = self._drive.CreateFile(param)
-
-        with Tqdm(
-            desc=progress_desc,
-            disable=no_progress_bar,
-            bytes=True,
-            # explicit `bar_format` as `total` will be set by `update_to`
-            bar_format=Tqdm.BAR_FMT_DEFAULT,
-        ) as pbar:
-            gdrive_file.GetContentFile(to_file, callback=pbar.update_to)
-
-    @contextmanager
-    @_gdrive_retry
-    def open(self, path_info, mode="r", encoding=None, **kwargs):
-        assert mode in {"r", "rt", "rb"}
-
-        item_id = self._get_item_id(path_info)
-        param = {"id": item_id}
-        # it does not create a file on the remote
-        gdrive_file = self._drive.CreateFile(param)
-        fd = gdrive_file.GetContentIOBuffer()
-        stream = IterStream(iter(fd))
-
-        if mode != "rb":
-            stream = io.TextIOWrapper(stream, encoding=encoding)
-
-        yield stream
-
-    @_gdrive_retry
-    def gdrive_delete_file(self, item_id):
-        from pydrive2.files import ApiRequestError
-
-        param = {"id": item_id}
-        # it does not create a file on the remote
-        item = self._drive.CreateFile(param)
-
-        try:
-            item.Trash() if self._trash_only else item.Delete()
-        except ApiRequestError as exc:
-            http_error_code = exc.error.get("code", 0)
-            if (
-                http_error_code == 403
-                and self._list_params["corpora"] == "drive"
-                and exc.GetField("location") == "file.permissions"
-            ):
-                raise DvcException(
-                    "Insufficient permissions to {}. You should have {} "
-                    "access level for the used shared drive. More details "
-                    "at {}.".format(
-                        "move the file into Trash"
-                        if self._trash_only
-                        else "permanently delete the file",
-                        "Manager or Content Manager"
-                        if self._trash_only
-                        else "Manager",
-                        "https://support.google.com/a/answer/7337554",
-                    )
-                ) from exc
-            raise
-
-    def _gdrive_list(self, query):
-        param = {"q": query, "maxResults": 1000}
-        param.update(self._list_params)
-        file_list = self._drive.ListFile(param)
-
-        # Isolate and decorate fetching of remote drive items in pages.
-        get_list = _gdrive_retry(lambda: next(file_list, None))
-
-        # Fetch pages until None is received, lazily flatten the thing.
-        return cat(iter(get_list, None))
-
-    @_gdrive_retry
-    def _gdrive_create_dir(self, parent_id, title):
-        parent = {"id": parent_id}
-        item = self._drive.CreateFile(
-            {"title": title, "parents": [parent], "mimeType": FOLDER_MIME_TYPE}
-        )
-        item.Upload()
-        return item
-
-    @wrap_with(threading.RLock())
-    def _create_dir(self, parent_id, title, remote_path):
-        cached = self._ids_cache["dirs"].get(remote_path)
-        if cached:
-            return cached[0]
-
-        item = self._gdrive_create_dir(parent_id, title)
-
-        if parent_id == self._ids_cache["root_id"]:
-            self._cache_path_id(remote_path, item["id"])
-
-        return item["id"]
-
-    def _get_remote_item_ids(self, parent_ids, title):
-        if not parent_ids:
-            return None
-        query = "trashed=false and ({})".format(
-            " or ".join(
-                f"'{parent_id}' in parents" for parent_id in parent_ids
-            )
-        )
-        query += " and title='{}'".format(title.replace("'", "\\'"))
-
-        # GDrive list API is case insensitive, we need to compare
-        # all results and pick the ones with the right title
-        return [
-            item["id"]
-            for item in self._gdrive_list(query)
-            if item["title"] == title
-        ]
-
-    def _get_cached_item_ids(self, path, use_cache):
-        if not path:
-            return [self._bucket]
-        if use_cache:
-            return self._ids_cache["dirs"].get(path, [])
-        return []
-
-    def _path_to_item_ids(self, path, create=False, use_cache=True):
-        item_ids = self._get_cached_item_ids(path, use_cache)
-        if item_ids:
-            return item_ids
-
-        parent_path, title = posixpath.split(path)
-        parent_ids = self._path_to_item_ids(parent_path, create, use_cache)
-        item_ids = self._get_remote_item_ids(parent_ids, title)
-        if item_ids:
-            return item_ids
-
-        return (
-            [self._create_dir(min(parent_ids), title, path)] if create else []
+        return _GDriveFileSystem(
+            self._with_bucket(self.path_info),
+            gauth,
+            trash_only=self._trash_only,
         )
 
-    def _get_item_id(self, path_info, create=False, use_cache=True, hint=None):
-        assert path_info.bucket == self._bucket
+    def _with_bucket(self, path):
+        if isinstance(path, self.PATH_CLS):
+            return posixpath.join(path.bucket, path.path)
 
-        item_ids = self._path_to_item_ids(path_info.path, create, use_cache)
-        if item_ids:
-            return min(item_ids)
+        return super()._with_bucket(path)
 
-        assert not create
-        raise FileMissingError(path_info, hint)
-
-    def exists(self, path_info) -> bool:
-        try:
-            self._get_item_id(path_info)
-        except FileMissingError:
-            return False
-        else:
-            return True
-
-    def _gdrive_list_ids(self, query_ids):
-        query = " or ".join(
-            f"'{query_id}' in parents" for query_id in query_ids
-        )
-        query = f"({query}) and trashed=false"
-        return self._gdrive_list(query)
-
-    def find(self, path_info, detail=False):
-        root_path = path_info.path
-        seen_paths = set()
-
-        dir_ids = [self._ids_cache["ids"].copy()]
-        while dir_ids:
-            query_ids = {
-                dir_id: dir_name
-                for dir_id, dir_name in dir_ids.pop().items()
-                if posixpath.commonpath([root_path, dir_name]) == root_path
-                if dir_id not in seen_paths
-            }
-            if not query_ids:
-                continue
-
-            seen_paths |= query_ids.keys()
-
-            new_query_ids = {}
-            dir_ids.append(new_query_ids)
-            for item in self._gdrive_list_ids(query_ids):
-                parent_id = item["parents"][0]["id"]
-                item_path = posixpath.join(query_ids[parent_id], item["title"])
-                if item["mimeType"] == FOLDER_MIME_TYPE:
-                    new_query_ids[item["id"]] = item_path
-                    self._cache_path_id(item_path, item["id"])
-                    continue
-
-                if detail:
-                    yield {
-                        "type": "file",
-                        "name": item_path,
-                        "size": item["fileSize"],
-                        "checksum": item["md5Checksum"],
-                    }
-                else:
-                    yield item_path
-
-    def ls(self, path_info, detail=False):
-        cached = path_info.path in self._ids_cache["dirs"]
-        if cached:
-            dir_ids = self._ids_cache["dirs"][path_info.path]
-        else:
-            dir_ids = self._path_to_item_ids(path_info.path)
-
-        if not dir_ids:
-            return None
-
-        root_path = path_info.path
-        for item in self._gdrive_list_ids(dir_ids):
-            item_path = posixpath.join(root_path, item["title"])
-            if detail:
-                if item["mimeType"] == FOLDER_MIME_TYPE:
-                    yield {"type": "directory", "name": item_path}
-                else:
-                    yield {
-                        "type": "file",
-                        "name": item_path,
-                        "size": item["fileSize"],
-                        "checksum": item["md5Checksum"],
-                    }
-            else:
-                yield item_path
-
-        if not cached:
-            self._cache_path_id(root_path, *dir_ids)
-
-    def walk_files(self, path_info, **kwargs):
-        for file in self.find(path_info):
-            yield path_info.replace(path=file)
-
-    def remove(self, path_info):
-        item_id = self._get_item_id(path_info)
-        self.gdrive_delete_file(item_id)
-
-    def info(self, path_info):
-        item_id = self._get_item_id(path_info)
-        gdrive_file = self._drive.CreateFile({"id": item_id})
-        gdrive_file.FetchMetadata(fields="fileSize")
-        return {"size": gdrive_file.get("fileSize")}
-
-    def _upload_fobj(self, fobj, to_info):
-        dirname = to_info.parent
-        assert dirname
-        parent_id = self._get_item_id(dirname, create=True)
-        self._gdrive_upload_fobj(fobj, parent_id, to_info.name)
-
-    def _upload(
-        self, from_file, to_info, name=None, no_progress_bar=False, **_kwargs
-    ):
-        with open(from_file, "rb") as fobj:
-            self.upload_fobj(
-                fobj,
-                to_info,
-                no_progress_bar=no_progress_bar,
-                desc=name or to_info.name,
-                total=os.path.getsize(from_file),
-            )
-
-    def _download(self, from_info, to_file, name=None, no_progress_bar=False):
-        item_id = self._get_item_id(from_info)
-        self._gdrive_download_file(item_id, to_file, name, no_progress_bar)
+    def _upload_fobj(self, fobj, to_info, size: int = None):
+        rpath = self._with_bucket(to_info)
+        self.makedirs(os.path.dirname(rpath))
+        return self.fs.upload_fobj(fobj, rpath, size=size)

dvc/fs/git.py

@@ -1,5 +1,6 @@
 import errno
 import os
+import stat
 
 from dvc.utils import is_exec, relpath
 
@@ -9,6 +10,8 @@ from .base import BaseFileSystem
 class GitFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
     """Proxies the repo file access methods to Git objects"""
 
+    sep = os.sep
+
     scheme = "local"
 
     def __init__(self, root_dir, trie):
@@ -33,8 +36,8 @@ class GitFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
         return tuple(relparts)
 
     def open(
-        self, path, mode="r", encoding=None
-    ):  # pylint: disable=arguments-differ
+        self, path, mode="r", encoding=None, **kwargs
+    ):  # pylint: disable=arguments-renamed
         # NOTE: this is incorrect, we need to use locale to determine default
         # encoding.
         encoding = encoding or "utf-8"
@@ -104,7 +107,10 @@ class GitFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
             raise FileNotFoundError(
                 errno.ENOENT, os.strerror(errno.ENOENT), path_info
             )
-        return {"size": st.st_size}
+        return {
+            "size": st.st_size,
+            "type": "directory" if stat.S_ISDIR(st.st_mode) else "file",
+        }
 
     def stat(self, path):
         key = self._get_key(path)
@@ -120,3 +126,17 @@ class GitFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
             for file in files:
                 # NOTE: os.path.join is ~5.5 times slower
                 yield f"{root}{os.sep}{file}"
+
+    def _download(
+        self, from_info, to_file, name=None, no_progress_bar=False, **kwargs
+    ):
+        import shutil
+
+        from dvc.progress import Tqdm
+
+        with open(to_file, "wb+") as to_fobj:
+            with Tqdm.wrapattr(
+                to_fobj, "write", desc=name, disable=no_progress_bar
+            ) as wrapped:
+                with self.open(from_info, "rb", **kwargs) as from_fobj:
+                    shutil.copyfileobj(from_fobj, wrapped)

dvc/fs/gs.py

@@ -16,12 +16,7 @@ class GSFileSystem(ObjectFSWrapper):
     REQUIRES = {"gcsfs": "gcsfs"}
     PARAM_CHECKSUM = "etag"
     DETAIL_FIELDS = frozenset(("etag", "size"))
-
-    def __init__(self, **config):
-        super().__init__(**config)
-
-        url = config.get("url", "gs://")
-        self.path_info = self.PATH_CLS(url)
+    TRAVERSE_PREFIX_LEN = 2
 
     def _prepare_credentials(self, **config):
         login_info = {"consistency": None}

dvc/fs/hdfs.py

@@ -1,3 +1,4 @@
+import errno
 import io
 import logging
 import os
@@ -6,7 +7,6 @@ import shutil
 import subprocess
 from collections import deque
 from contextlib import closing, contextmanager
-from urllib.parse import urlparse
 
 from dvc.hash_info import HashInfo
 from dvc.progress import Tqdm
@@ -68,24 +68,19 @@ class HDFSFileSystem(BaseFileSystem):
     def __init__(self, **config):
         super().__init__(**config)
 
-        self.path_info = None
-        url = config.get("url")
-        if not url:
-            return
-
-        parsed = urlparse(url)
-        user = parsed.username or config.get("user")
-
-        self.path_info = self.PATH_CLS.from_parts(
-            scheme=self.scheme,
-            host=parsed.hostname,
-            user=user,
-            port=parsed.port,
-            path=parsed.path,
-        )
+        self.host = config["host"]
+        self.user = config.get("user")
+        self.port = config.get("port")
 
         self.kerb_ticket = config.get("kerb_ticket")
 
+    @staticmethod
+    def _get_kwargs_from_urls(urlpath):
+        from fsspec.implementations.hdfs import PyArrowHDFS
+
+        # pylint:disable=protected-access
+        return PyArrowHDFS._get_kwargs_from_urls(urlpath)
+
     def hdfs(self, path_info):
         import pyarrow.fs
 
@@ -188,7 +183,7 @@ class HDFSFileSystem(BaseFileSystem):
                 else:
                     hdfs.delete_file(path_info.path)
 
-    def makedirs(self, path_info):
+    def makedirs(self, path_info, **kwargs):
         with self.hdfs(path_info) as hdfs:
             # NOTE: fs.create_dir creates parents by default
             hdfs.create_dir(path_info.path)
@@ -226,8 +221,19 @@ class HDFSFileSystem(BaseFileSystem):
 
     def info(self, path_info):
         with self.hdfs(path_info) as hdfs:
-            finfo = hdfs.get_file_info(path_info.path)
-            return {"size": finfo.size}
+            from pyarrow.fs import FileType
+
+            file_info = hdfs.get_file_info(path_info.path)
+            if file_info.type is FileType.Directory:
+                kind = "directory"
+            elif file_info.type is FileType.File:
+                kind = "file"
+            else:
+                raise FileNotFoundError(
+                    errno.ENOENT, os.strerror(errno.ENOENT), path_info
+                )
+
+            return {"size": file_info.size, "type": kind}
 
     def checksum(self, path_info):
         return HashInfo(
@@ -236,7 +242,7 @@ class HDFSFileSystem(BaseFileSystem):
             size=self.getsize(path_info),
         )
 
-    def _upload_fobj(self, fobj, to_info):
+    def _upload_fobj(self, fobj, to_info, **kwargs):
         with self.hdfs(to_info) as hdfs:
             with hdfs.open_output_stream(to_info.path) as fdest:
                 shutil.copyfileobj(fobj, fdest)

dvc/fs/http.py

@@ -2,6 +2,7 @@ import logging
 import os.path
 import threading
 from typing import Optional
+from urllib.parse import urlparse
 
 from funcy import cached_property, memoize, wrap_prop, wrap_with
 
@@ -30,6 +31,7 @@ class HTTPFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
     PATH_CLS = HTTPURLInfo
     PARAM_CHECKSUM = "etag"
     CAN_TRAVERSE = False
+    REQUIRES = {"requests": "requests"}
 
     SESSION_RETRIES = 5
     SESSION_BACKOFF_FACTOR = 0.1
@@ -39,13 +41,7 @@ class HTTPFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
     def __init__(self, **config):
         super().__init__(**config)
 
-        url = config.get("url")
-        if url:
-            self.path_info = self.PATH_CLS(url)
-            self.user = config.get("user", None)
-            self.host = self.path_info.host
-        else:
-            self.path_info = None
+        self.user = config.get("user", None)
 
         self.auth = config.get("auth", None)
         self.custom_auth_header = config.get("custom_auth_header", None)
@@ -55,12 +51,12 @@ class HTTPFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
         self.ssl_verify = config.get("ssl_verify", True)
         self.method = config.get("method", "POST")
 
-    def _auth_method(self):
+    def _auth_method(self, url):
         from requests.auth import HTTPBasicAuth, HTTPDigestAuth
 
         if self.auth:
             if self.ask_password and self.password is None:
-                self.password = ask_password(self.host, self.user)
+                self.password = ask_password(urlparse(url).hostname, self.user)
             if self.auth == "basic":
                 return HTTPBasicAuth(self.user, self.password)
             if self.auth == "digest":
@@ -103,7 +99,7 @@ class HTTPFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
             res = self._session.request(
                 method,
                 url,
-                auth=self._auth_method(),
+                auth=self._auth_method(url),
                 headers=self.headers,
                 **kwargs,
             )
@@ -150,9 +146,9 @@ class HTTPFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
         resp = self._head(path_info.url)
         etag = resp.headers.get("ETag") or resp.headers.get("Content-MD5")
         size = self._content_length(resp)
-        return {"etag": etag, "size": size}
+        return {"etag": etag, "size": size, "type": "file"}
 
-    def _upload_fobj(self, fobj, to_info):
+    def _upload_fobj(self, fobj, to_info, **kwargs):
         def chunks(fobj):
             while True:
                 chunk = fobj.read(self.CHUNK_SIZE)
@@ -189,9 +185,9 @@ class HTTPFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
             self.upload_fobj(
                 fobj,
                 to_info,
+                size=None if no_progress_bar else os.path.getsize(from_file),
                 no_progress_bar=no_progress_bar,
                 desc=name or to_info.url,
-                total=None if no_progress_bar else os.path.getsize(from_file),
             )
 
     @staticmethod

dvc/fs/local.py

@@ -14,6 +14,8 @@ logger = logging.getLogger(__name__)
 
 
 class LocalFileSystem(BaseFileSystem):
+    sep = os.sep
+
     scheme = Schemes.LOCAL
     PATH_CLS = PathInfo
     PARAM_CHECKSUM = "md5"
@@ -25,8 +27,6 @@ class LocalFileSystem(BaseFileSystem):
 
         super().__init__(**config)
         self.fs = LocalFS()
-        url = config.get("url")
-        self.path_info = self.PATH_CLS(url) if url else None
 
     @staticmethod
     def open(path_info, mode="r", encoding=None, **kwargs):
@@ -79,8 +79,8 @@ class LocalFileSystem(BaseFileSystem):
                 raise NotImplementedError
         remove(path_info)
 
-    def makedirs(self, path_info):
-        makedirs(path_info, exist_ok=True)
+    def makedirs(self, path_info, **kwargs):
+        makedirs(path_info, exist_ok=kwargs.pop("exist_ok", True))
 
     def isexec(self, path_info):
         mode = self.stat(path_info).st_mode
@@ -90,7 +90,9 @@ class LocalFileSystem(BaseFileSystem):
         return os.stat(path)
 
     def move(self, from_info, to_info):
-        if from_info.scheme != "local" or to_info.scheme != "local":
+        if (
+            isinstance(from_info, PathInfo) and from_info.scheme != "local"
+        ) or (isinstance(to_info, PathInfo) and to_info.scheme != "local"):
             raise NotImplementedError
 
         self.makedirs(to_info.parent)
@@ -105,7 +107,7 @@ class LocalFileSystem(BaseFileSystem):
             self.remove(tmp_info)
             raise
 
-    def _upload_fobj(self, fobj, to_info):
+    def _upload_fobj(self, fobj, to_info, **kwargs):
         self.makedirs(to_info.parent)
         tmp_info = to_info.parent / tmp_fname("")
         try:
@@ -157,7 +159,7 @@ class LocalFileSystem(BaseFileSystem):
         st = os.stat(path_info)
         return {
             "size": st.st_size,
-            "type": "dir" if stat.S_ISDIR(st.st_mode) else "file",
+            "type": "directory" if stat.S_ISDIR(st.st_mode) else "file",
         }
 
     def _upload(

dvc/fs/memory.py

@@ -1,34 +1,35 @@
-from .base import BaseFileSystem
+import threading
+
+from funcy import cached_property, wrap_prop
+
+from dvc.path_info import CloudURLInfo
+from dvc.scheme import Schemes
+
+from .fsspec_wrapper import FSSpecWrapper
 
 
-class MemoryFileSystem(BaseFileSystem):
-    scheme = "local"
+class MemoryFileSystem(FSSpecWrapper):  # pylint:disable=abstract-method
+    scheme = Schemes.MEMORY
     PARAM_CHECKSUM = "md5"
+    PATH_CLS = CloudURLInfo
+    TRAVERSE_PREFIX_LEN = 2
+    DEFAULT_BLOCKSIZE = 4096
 
-    def __init__(self, **kwargs):
+    def __eq__(self, other):
+        # NOTE: all fsspec MemoryFileSystem instances are equivalent and use a
+        # single global store
+        return isinstance(other, type(self))
+
+    __hash__ = FSSpecWrapper.__hash__
+
+    @wrap_prop(threading.Lock())
+    @cached_property
+    def fs(self):
         from fsspec.implementations.memory import MemoryFileSystem as MemFS
 
-        super().__init__(**kwargs)
+        return MemFS(**self.fs_args)
 
-        self.fs = MemFS()
-
-    def exists(self, path_info) -> bool:
-        return self.fs.exists(path_info.fspath)
-
-    def open(self, path_info, mode="r", encoding=None, **kwargs):
-        return self.fs.open(
-            path_info.fspath, mode=mode, encoding=encoding, **kwargs
-        )
-
-    def info(self, path_info):
-        return self.fs.info(path_info.fspath)
-
-    def stat(self, path_info):
-        import os
-
-        info = self.fs.info(path_info.fspath)
-
-        return os.stat_result((0, 0, 0, 0, 0, 0, info["size"], 0, 0, 0))
-
-    def walk_files(self, path_info, **kwargs):
-        raise NotImplementedError
+    def open(self, *args, **kwargs):
+        with super().open(*args, **kwargs) as fobj:
+            fobj.blocksize = self.DEFAULT_BLOCKSIZE
+            return fobj

dvc/fs/oss.py

@@ -5,137 +5,37 @@ import threading
 from funcy import cached_property, wrap_prop
 
 from dvc.path_info import CloudURLInfo
-from dvc.progress import Tqdm
 from dvc.scheme import Schemes
 
-from .base import BaseFileSystem
+from .fsspec_wrapper import CallbackMixin, ObjectFSWrapper
 
 logger = logging.getLogger(__name__)
 
 
-class OSSFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
-    """
-    oss2 document:
-    https://www.alibabacloud.com/help/doc-detail/32026.htm
-
-
-    Examples
-    ----------
-    $ dvc remote add myremote oss://my-bucket/path
-    Set key id, key secret and endpoint using modify command
-    $ dvc remote modify myremote oss_key_id my-key-id
-    $ dvc remote modify myremote oss_key_secret my-key-secret
-    $ dvc remote modify myremote oss_endpoint endpoint
-    or environment variables
-    $ export OSS_ACCESS_KEY_ID="my-key-id"
-    $ export OSS_ACCESS_KEY_SECRET="my-key-secret"
-    $ export OSS_ENDPOINT="endpoint"
-    """
-
+# pylint:disable=abstract-method
+class OSSFileSystem(CallbackMixin, ObjectFSWrapper):
     scheme = Schemes.OSS
     PATH_CLS = CloudURLInfo
-    REQUIRES = {"oss2": "oss2"}
+    REQUIRES = {"ossfs": "ossfs"}
     PARAM_CHECKSUM = "etag"
     COPY_POLL_SECONDS = 5
     LIST_OBJECT_PAGE_SIZE = 100
+    DETAIL_FIELDS = frozenset(("etag", "size"))
 
-    def __init__(self, **config):
-        super().__init__(**config)
-
-        url = config.get("url")
-        self.path_info = self.PATH_CLS(url) if url else None
-
-        self.bucket = self.path_info.bucket
-        self.endpoint = config.get("oss_endpoint") or os.getenv("OSS_ENDPOINT")
-
-        self.key_id = (
-            config.get("oss_key_id")
-            or os.getenv("OSS_ACCESS_KEY_ID")
-            or "defaultId"
+    def _prepare_credentials(self, **config):
+        login_info = {}
+        login_info["key"] = config.get("oss_key_id") or os.getenv(
+            "OSS_ACCESS_KEY_ID"
         )
-
-        self.key_secret = (
-            config.get("oss_key_secret")
-            or os.getenv("OSS_ACCESS_KEY_SECRET")
-            or "defaultSecret"
+        login_info["secret"] = config.get("oss_key_secret") or os.getenv(
+            "OSS_ACCESS_KEY_SECRET"
         )
+        login_info["endpoint"] = config.get("oss_endpoint")
+        return login_info
 
     @wrap_prop(threading.Lock())
     @cached_property
-    def oss_service(self):
-        import oss2
+    def fs(self):
+        from ossfs import OSSFileSystem as _OSSFileSystem
 
-        logger.debug(f"key id: {self.key_id}")
-        logger.debug(f"key secret: {self.key_secret}")
-
-        auth = oss2.Auth(self.key_id, self.key_secret)
-        bucket = oss2.Bucket(auth, self.endpoint, self.bucket)
-
-        # Ensure bucket exists
-        try:
-            bucket.get_bucket_info()
-        except oss2.exceptions.NoSuchBucket:
-            bucket.create_bucket(
-                oss2.BUCKET_ACL_PUBLIC_READ,
-                oss2.models.BucketCreateConfig(
-                    oss2.BUCKET_STORAGE_CLASS_STANDARD
-                ),
-            )
-        return bucket
-
-    def _generate_download_url(self, path_info, expires=3600):
-        assert path_info.bucket == self.bucket
-
-        return self.oss_service.sign_url("GET", path_info.path, expires)
-
-    def exists(self, path_info) -> bool:
-        paths = self._list_paths(path_info)
-        return any(path_info.path == path for path in paths)
-
-    def _list_paths(self, path_info):
-        import oss2
-
-        for blob in oss2.ObjectIterator(
-            self.oss_service, prefix=path_info.path
-        ):
-            yield blob.key
-
-    def walk_files(self, path_info, **kwargs):
-        if not kwargs.pop("prefix", False):
-            path_info = path_info / ""
-        for fname in self._list_paths(path_info):
-            if fname.endswith("/"):
-                continue
-
-            yield path_info.replace(path=fname)
-
-    def remove(self, path_info):
-        if path_info.scheme != self.scheme:
-            raise NotImplementedError
-
-        logger.debug(f"Removing oss://{path_info}")
-        self.oss_service.delete_object(path_info.path)
-
-    def _upload_fobj(self, fobj, to_info):
-        self.oss_service.put_object(to_info.path, fobj)
-
-    def _upload(
-        self, from_file, to_info, name=None, no_progress_bar=False, **_kwargs
-    ):
-        with Tqdm(desc=name, disable=no_progress_bar, bytes=True) as pbar:
-            self.oss_service.put_object_from_file(
-                to_info.path, from_file, progress_callback=pbar.update_to
-            )
-
-    def _download(
-        self, from_info, to_file, name=None, no_progress_bar=False, **_kwargs
-    ):
-        with Tqdm(desc=name, disable=no_progress_bar, bytes=True) as pbar:
-            import oss2
-
-            oss2.resumable_download(
-                self.oss_service,
-                from_info.path,
-                to_file,
-                progress_callback=pbar.update_to,
-            )
+        return _OSSFileSystem(**self.fs_args)

dvc/fs/repo.py

@@ -31,16 +31,33 @@ class RepoFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
         kwargs: Additional keyword arguments passed to the `DvcFileSystem()`.
     """
 
+    sep = os.sep
+
     scheme = "local"
     PARAM_CHECKSUM = "md5"
+    PARAM_REPO_URL = "repo_url"
+    PARAM_REPO_ROOT = "repo_root"
+    PARAM_REV = "rev"
+    PARAM_CACHE_DIR = "cache_dir"
+    PARAM_CACHE_TYPES = "cache_types"
+    PARAM_SUBREPOS = "subrepos"
 
     def __init__(
-        self, repo=None, subrepos=False, repo_factory: RepoFactory = None
+        self,
+        repo: Optional["Repo"] = None,
+        subrepos=False,
+        repo_factory: RepoFactory = None,
+        **kwargs,
     ):
         super().__init__()
 
         from dvc.utils.collections import PathStringTrie
 
+        if repo is None:
+            repo, repo_factory = self._repo_from_fs_config(
+                subrepos=subrepos, **kwargs
+            )
+
         if not repo_factory:
             from dvc.repo import Repo
 
@@ -64,6 +81,73 @@ class RepoFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
         if hasattr(repo, "dvc_dir"):
             self._dvcfss[repo.root_dir] = DvcFileSystem(repo=repo)
 
+    @property
+    def repo_url(self):
+        if self._main_repo is None:
+            return None
+        return self._main_repo.url
+
+    @property
+    def config(self):
+        return {
+            self.PARAM_REPO_URL: self.repo_url,
+            self.PARAM_REPO_ROOT: self.root_dir,
+            self.PARAM_REV: getattr(self._main_repo.fs, "rev", None),
+            self.PARAM_CACHE_DIR: os.path.abspath(
+                self._main_repo.odb.local.cache_dir
+            ),
+            self.PARAM_CACHE_TYPES: self._main_repo.odb.local.cache_types,
+            self.PARAM_SUBREPOS: self._traverse_subrepos,
+        }
+
+    @classmethod
+    def _repo_from_fs_config(
+        cls, **config
+    ) -> Tuple["Repo", Optional["RepoFactory"]]:
+        from dvc.external_repo import erepo_factory, external_repo
+        from dvc.repo import Repo
+
+        url = config.get(cls.PARAM_REPO_URL)
+        root = config.get(cls.PARAM_REPO_ROOT)
+        assert url or root
+
+        def _open(*args, **kwargs):
+            # NOTE: if original repo was an erepo (and has a URL),
+            # we cannot use Repo.open() since it will skip erepo
+            # cache/remote setup for local URLs
+            if url is None:
+                return Repo.open(*args, **kwargs)
+            return external_repo(*args, **kwargs)
+
+        cache_dir = config.get(cls.PARAM_CACHE_DIR)
+        cache_config = (
+            {}
+            if not cache_dir
+            else {
+                "cache": {
+                    "dir": cache_dir,
+                    "type": config.get(cls.PARAM_CACHE_TYPES),
+                }
+            }
+        )
+        repo_kwargs: dict = {
+            "rev": config.get(cls.PARAM_REV),
+            "subrepos": config.get(cls.PARAM_SUBREPOS, False),
+            "uninitialized": True,
+        }
+        factory: Optional["RepoFactory"] = None
+        if url is None:
+            repo_kwargs["config"] = cache_config
+        else:
+            repo_kwargs["cache_dir"] = cache_dir
+            factory = erepo_factory(url, cache_config)
+
+        with _open(
+            url if url else root,
+            **repo_kwargs,
+        ) as repo:
+            return repo, factory
+
     def _get_repo(self, path: str) -> Optional["Repo"]:
         """Returns repo that the path falls in, using prefix.
 
@@ -128,7 +212,7 @@ class RepoFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
 
     def open(
         self, path, mode="r", encoding="utf-8", **kwargs
-    ):  # pylint: disable=arguments-differ
+    ):  # pylint: disable=arguments-renamed
         if "b" in mode:
             encoding = None
 
@@ -159,13 +243,13 @@ class RepoFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
         except FileNotFoundError:
             return False
 
-        (out,) = meta.outs
-        assert len(meta.outs) == 1
-        if fs.exists(out.path_info):
-            return False
+        for out in meta.outs:
+            if fs.exists(out.path_info):
+                return False
+
         return True
 
-    def isdir(self, path):  # pylint: disable=arguments-differ
+    def isdir(self, path):  # pylint: disable=arguments-renamed
         fs, dvc_fs = self._get_fs_pair(path)
 
         if dvc_fs and dvc_fs.repo.dvcignore.is_ignored_dir(path):
@@ -186,17 +270,17 @@ class RepoFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
         except FileNotFoundError:
             return False
 
-        (out,) = meta.outs
-        assert len(meta.outs) == 1
-        if fs.exists(out.path_info):
-            return False
+        for out in meta.outs:
+            if fs.exists(out.path_info):
+                return False
+
         return meta.isdir
 
     def isdvc(self, path, **kwargs):
         _, dvc_fs = self._get_fs_pair(path)
         return dvc_fs is not None and dvc_fs.isdvc(path, **kwargs)
 
-    def isfile(self, path):  # pylint: disable=arguments-differ
+    def isfile(self, path):  # pylint: disable=arguments-renamed
         fs, dvc_fs = self._get_fs_pair(path)
 
         if dvc_fs and dvc_fs.repo.dvcignore.is_ignored_file(path):
@@ -369,19 +453,20 @@ class RepoFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
             for fname in files:
                 yield PathInfo(root) / fname
 
-    def _download(
-        self, from_info, to_file, name=None, no_progress_bar=False, **kwargs
-    ):
-        import shutil
+    def _download(self, from_info, to_file, **kwargs):
+        fs, dvc_fs = self._get_fs_pair(from_info)
+        try:
+            fs._download(  # pylint: disable=protected-access
+                from_info, to_file, **kwargs
+            )
+            return
+        except FileNotFoundError:
+            if not dvc_fs:
+                raise
 
-        from dvc.progress import Tqdm
-
-        with open(to_file, "wb+") as to_fobj:
-            with Tqdm.wrapattr(
-                to_fobj, "write", desc=name, disable=no_progress_bar
-            ) as wrapped:
-                with self.open(from_info, "rb", **kwargs) as from_fobj:
-                    shutil.copyfileobj(from_fobj, wrapped)
+        dvc_fs._download(  # pylint: disable=protected-access
+            from_info, to_file, **kwargs
+        )
 
     def metadata(self, path):
         abspath = os.path.abspath(path)

dvc/fs/s3.py

@@ -29,12 +29,6 @@ class BaseS3FileSystem(ObjectFSWrapper):
         "grant_write_acp": "GrantWriteACP",
     }
 
-    def __init__(self, **config):
-        super().__init__(**config)
-
-        url = config.get("url", "s3://")
-        self.path_info = self.PATH_CLS(url)
-
     _TRANSFER_CONFIG_ALIASES = {
         "max_queue_size": "max_io_queue",
         "max_concurrent_requests": "max_concurrency",
@@ -103,7 +97,7 @@ class BaseS3FileSystem(ObjectFSWrapper):
         client = login_info["client_kwargs"]
         client["region_name"] = config.get("region")
         client["endpoint_url"] = config.get("endpointurl")
-        client["verify"] = config.get("ssl_verify", True)
+        client["verify"] = config.get("ssl_verify")
 
         # encryptions
         additional = login_info["s3_additional_kwargs"]
@@ -194,6 +188,7 @@ class S3FileSystem(BaseS3FileSystem):  # pylint:disable=abstract-method
             "s3",
             endpoint_url=client_kwargs.get("endpoint_url"),
             use_ssl=login_info["use_ssl"],
+            verify=client_kwargs.get("verify"),
         )
 
     def _get_obj(self, path_info):

dvc/fs/ssh.py

@@ -0,0 +1,125 @@
+import getpass
+import os.path
+import threading
+
+from funcy import cached_property, first, memoize, silent, wrap_prop, wrap_with
+
+from dvc import prompt
+from dvc.scheme import Schemes
+from dvc.utils.fs import as_atomic
+
+from .fsspec_wrapper import CallbackMixin, FSSpecWrapper
+
+_SSH_TIMEOUT = 60 * 30
+_SSH_CONFIG_FILE = os.path.expanduser(os.path.join("~", ".ssh", "config"))
+
+
+@wrap_with(threading.Lock())
+@memoize
+def ask_password(host, user, port):
+    return prompt.password(
+        "Enter a private key passphrase or a password for "
+        "host '{host}' port '{port}' user '{user}'".format(
+            host=host, port=port, user=user
+        )
+    )
+
+
+# pylint:disable=abstract-method
+class SSHFileSystem(CallbackMixin, FSSpecWrapper):
+    scheme = Schemes.SSH
+    REQUIRES = {"sshfs": "sshfs"}
+
+    DEFAULT_PORT = 22
+    PARAM_CHECKSUM = "md5"
+
+    def _with_bucket(self, path):
+        if isinstance(path, self.PATH_CLS):
+            path_str = path.path
+        else:
+            path_str = super()._with_bucket(path)
+        return os.path.normpath(path_str)
+
+    def _prepare_credentials(self, **config):
+        self.CAN_TRAVERSE = True
+        from sshfs.config import parse_config
+
+        login_info = {}
+
+        try:
+            user_ssh_config = parse_config(host=config["host"])
+        except FileNotFoundError:
+            user_ssh_config = {}
+
+        login_info["host"] = user_ssh_config.get("Hostname", config["host"])
+
+        login_info["username"] = (
+            config.get("user")
+            or config.get("username")
+            or user_ssh_config.get("User")
+            or getpass.getuser()
+        )
+
+        login_info["port"] = (
+            config.get("port")
+            or silent(int)(user_ssh_config.get("Port"))
+            or self.DEFAULT_PORT
+        )
+
+        login_info["password"] = config.get("password")
+
+        if user_ssh_config.get("IdentityFile"):
+            config.setdefault(
+                "keyfile", first(user_ssh_config.get("IdentityFile"))
+            )
+
+        keyfile = config.get("keyfile")
+        if keyfile:
+            login_info["client_keys"] = [os.path.expanduser(keyfile)]
+        login_info["timeout"] = config.get("timeout", _SSH_TIMEOUT)
+
+        # These two settings fine tune the asyncssh to use the
+        # fastest encryption algorithm and disable compression
+        # altogether (since it blocking, it is slowing down
+        # the transfers in a considerable rate, and even for
+        # compressible data it is making it extremely slow).
+        # See: https://github.com/ronf/asyncssh/issues/374
+        login_info["encryption_algs"] = [
+            "[email protected]",
+            "aes256-ctr",
+            "aes192-ctr",
+            "aes128-ctr",
+        ]
+        login_info["compression_algs"] = None
+
+        login_info["gss_auth"] = config.get("gss_auth", False)
+        login_info["agent_forwarding"] = config.get("agent_forwarding", True)
+        login_info["proxy_command"] = user_ssh_config.get("ProxyCommand")
+
+        if config.get("ask_password") and login_info["password"] is None:
+            login_info["password"] = ask_password(
+                login_info["host"], login_info["username"], login_info["port"]
+            )
+
+        # We are going to automatically add stuff to known_hosts
+        # something like paramiko's AutoAddPolicy()
+        login_info["known_hosts"] = None
+        return login_info
+
+    @wrap_prop(threading.Lock())
+    @cached_property
+    def fs(self):
+        from sshfs import SSHFileSystem as _SSHFileSystem
+
+        return _SSHFileSystem(**self.fs_args)
+
+    # Ensure that if an interrupt happens during the transfer, we don't
+    # pollute the cache.
+
+    def _upload_fobj(self, fobj, to_info, *args, **kwargs):
+        with as_atomic(self, to_info) as tmp_file:
+            super()._upload_fobj(fobj, tmp_file, *args, **kwargs)
+
+    def _upload(self, from_file, to_info, *args, **kwargs):
+        with as_atomic(self, to_info) as tmp_file:
+            super()._upload(from_file, tmp_file, *args, **kwargs)

dvc/fs/ssh/__init__.py

@@ -1,273 +0,0 @@
-import getpass
-import io
-import logging
-import os
-import shutil
-import threading
-from contextlib import closing, contextmanager
-from urllib.parse import urlparse
-
-from funcy import first, memoize, silent, wrap_with
-
-from dvc import prompt
-from dvc.hash_info import HashInfo
-from dvc.scheme import Schemes
-
-from ..base import BaseFileSystem
-from ..pool import get_connection
-
-logger = logging.getLogger(__name__)
-
-
-@wrap_with(threading.Lock())
-@memoize
-def ask_password(host, user, port):
-    return prompt.password(
-        "Enter a private key passphrase or a password for "
-        "host '{host}' port '{port}' user '{user}'".format(
-            host=host, port=port, user=user
-        )
-    )
-
-
-class SSHFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
-    scheme = Schemes.SSH
-    REQUIRES = {"paramiko": "paramiko"}
-    _JOBS = 4
-
-    PARAM_CHECKSUM = "md5"
-    # At any given time some of the connections will go over network and
-    # paramiko stuff, so we would ideally have it double of server processors.
-    # We use conservative setting of 4 instead to not exhaust max sessions.
-    CHECKSUM_JOBS = 4
-    DEFAULT_CACHE_TYPES = ["copy"]
-    TRAVERSE_PREFIX_LEN = 2
-
-    DEFAULT_PORT = 22
-    TIMEOUT = 1800
-
-    def __init__(self, **config):
-        super().__init__(**config)
-        url = config.get("url")
-        if url:
-            parsed = urlparse(url)
-            user_ssh_config = self._load_user_ssh_config(parsed.hostname)
-
-            host = user_ssh_config.get("hostname", parsed.hostname)
-            user = (
-                config.get("user")
-                or parsed.username
-                or user_ssh_config.get("user")
-                or getpass.getuser()
-            )
-            port = (
-                config.get("port")
-                or parsed.port
-                or self._try_get_ssh_config_port(user_ssh_config)
-                or self.DEFAULT_PORT
-            )
-            self.path_info = self.PATH_CLS.from_parts(
-                scheme=self.scheme,
-                host=host,
-                user=user,
-                port=port,
-                path=parsed.path,
-            )
-        else:
-            self.path_info = None
-            user_ssh_config = {}
-
-        self.keyfile = config.get(
-            "keyfile"
-        ) or self._try_get_ssh_config_keyfile(user_ssh_config)
-        self.timeout = config.get("timeout", self.TIMEOUT)
-        self.password = config.get("password", None)
-        self.ask_password = config.get("ask_password", False)
-        self.gss_auth = config.get("gss_auth", False)
-        proxy_command = user_ssh_config.get("proxycommand", False)
-        if proxy_command:
-            import paramiko
-
-            self.sock = paramiko.ProxyCommand(proxy_command)
-        else:
-            self.sock = None
-        self.allow_agent = config.get("allow_agent", True)
-
-    @staticmethod
-    def ssh_config_filename():
-        return os.path.expanduser(os.path.join("~", ".ssh", "config"))
-
-    @staticmethod
-    def _load_user_ssh_config(hostname):
-        import paramiko
-
-        user_config_file = SSHFileSystem.ssh_config_filename()
-        user_ssh_config = {}
-        if hostname and os.path.exists(user_config_file):
-            ssh_config = paramiko.SSHConfig()
-            with open(user_config_file) as f:
-                # For whatever reason parsing directly from f is unreliable
-                f_copy = io.StringIO(f.read())
-                ssh_config.parse(f_copy)
-            user_ssh_config = ssh_config.lookup(hostname)
-        return user_ssh_config
-
-    @staticmethod
-    def _try_get_ssh_config_port(user_ssh_config):
-        return silent(int)(user_ssh_config.get("port"))
-
-    @staticmethod
-    def _try_get_ssh_config_keyfile(user_ssh_config):
-        return first(user_ssh_config.get("identityfile") or ())
-
-    def ensure_credentials(self, path_info=None):
-        if path_info is None:
-            path_info = self.path_info
-
-        # NOTE: we use the same password regardless of the server :(
-        if self.ask_password and self.password is None:
-            host, user, port = path_info.host, path_info.user, path_info.port
-            self.password = ask_password(host, user, port)
-
-    def ssh(self, path_info):
-        self.ensure_credentials(path_info)
-
-        from .connection import SSHConnection
-
-        return get_connection(
-            SSHConnection,
-            path_info.host,
-            username=path_info.user,
-            port=path_info.port,
-            key_filename=self.keyfile,
-            timeout=self.timeout,
-            password=self.password,
-            gss_auth=self.gss_auth,
-            sock=self.sock,
-            allow_agent=self.allow_agent,
-        )
-
-    @contextmanager
-    def open(self, path_info, mode="r", encoding=None, **kwargs):
-        assert mode in {"r", "rt", "rb", "wb"}
-
-        with self.ssh(path_info) as ssh, closing(
-            ssh.sftp.open(path_info.path, mode)
-        ) as fd:
-            if "b" in mode:
-                yield fd
-            else:
-                yield io.TextIOWrapper(fd, encoding=encoding)
-
-    def exists(self, path_info) -> bool:
-        with self.ssh(path_info) as ssh:
-            return ssh.exists(path_info.path)
-
-    def isdir(self, path_info):
-        with self.ssh(path_info) as ssh:
-            return ssh.isdir(path_info.path)
-
-    def isfile(self, path_info):
-        with self.ssh(path_info) as ssh:
-            return ssh.isfile(path_info.path)
-
-    def walk_files(self, path_info, **kwargs):
-        with self.ssh(path_info) as ssh:
-            for fname in ssh.walk_files(path_info.path):
-                yield path_info.replace(path=fname)
-
-    def remove(self, path_info):
-        if path_info.scheme != self.scheme:
-            raise NotImplementedError
-
-        with self.ssh(path_info) as ssh:
-            ssh.remove(path_info.path)
-
-    def makedirs(self, path_info):
-        with self.ssh(path_info) as ssh:
-            ssh.makedirs(path_info.path)
-
-    def move(self, from_info, to_info):
-        if from_info.scheme != self.scheme or to_info.scheme != self.scheme:
-            raise NotImplementedError
-
-        with self.ssh(from_info) as ssh:
-            ssh.move(from_info.path, to_info.path)
-
-    def copy(self, from_info, to_info):
-        if not from_info.scheme == to_info.scheme == self.scheme:
-            raise NotImplementedError
-
-        with self.ssh(from_info) as ssh:
-            ssh.atomic_copy(from_info.path, to_info.path)
-
-    def symlink(self, from_info, to_info):
-        if not from_info.scheme == to_info.scheme == self.scheme:
-            raise NotImplementedError
-
-        with self.ssh(from_info) as ssh:
-            ssh.symlink(from_info.path, to_info.path)
-
-    def hardlink(self, from_info, to_info):
-        if not from_info.scheme == to_info.scheme == self.scheme:
-            raise NotImplementedError
-
-        # See dvc/remote/local/__init__.py - hardlink()
-        if self.getsize(from_info) == 0:
-
-            with self.ssh(to_info) as ssh:
-                ssh.sftp.open(to_info.path, "w").close()
-
-            logger.debug(
-                "Created empty file: {src} -> {dest}".format(
-                    src=str(from_info), dest=str(to_info)
-                )
-            )
-            return
-
-        with self.ssh(from_info) as ssh:
-            ssh.hardlink(from_info.path, to_info.path)
-
-    def reflink(self, from_info, to_info):
-        if from_info.scheme != self.scheme or to_info.scheme != self.scheme:
-            raise NotImplementedError
-
-        with self.ssh(from_info) as ssh:
-            ssh.reflink(from_info.path, to_info.path)
-
-    def md5(self, path_info):
-        with self.ssh(path_info) as ssh:
-            return HashInfo(
-                "md5",
-                ssh.md5(path_info.path),
-                size=ssh.getsize(path_info.path),
-            )
-
-    def info(self, path_info):
-        with self.ssh(path_info) as ssh:
-            return ssh.info(path_info.path)
-
-    def _upload_fobj(self, fobj, to_info):
-        self.makedirs(to_info.parent)
-        with self.open(to_info, mode="wb") as fdest:
-            shutil.copyfileobj(fobj, fdest)
-
-    def _download(self, from_info, to_file, name=None, no_progress_bar=False):
-        with self.ssh(from_info) as ssh:
-            ssh.download(
-                from_info.path,
-                to_file,
-                progress_title=name,
-                no_progress_bar=no_progress_bar,
-            )
-
-    def _upload(
-        self, from_file, to_info, name=None, no_progress_bar=False, **_kwargs
-    ):
-        with self.ssh(to_info) as ssh:
-            ssh.upload(
-                from_file,
-                to_info.path,
-                progress_title=name,
-                no_progress_bar=no_progress_bar,
-            )

dvc/fs/ssh/connection.py

@@ -1,356 +0,0 @@
-import errno
-import logging
-import os
-import posixpath
-import shlex
-import stat
-from contextlib import suppress
-
-from funcy import cached_property
-
-from dvc.exceptions import DvcException
-from dvc.progress import Tqdm
-from dvc.utils import tmp_fname
-
-from ..base import RemoteCmdError
-
-try:
-    import paramiko
-except ImportError:
-    paramiko = None  # type: ignore
-
-
-logger = logging.getLogger(__name__)
-
-
-def sizeof_fmt(num, suffix="B"):
-    """Convert number of bytes to human-readable string"""
-    for unit in ["", "K", "M", "G", "T", "P", "E", "Z"]:
-        if abs(num) < 1024.0:
-            return f"{num:3.1f}{unit}{suffix}"
-        num /= 1024.0
-    return "{:.1f}{}{}".format(num, "Y", suffix)
-
-
-class SSHConnection:
-    def __init__(self, host, *args, **kwargs):
-        logger.debug(
-            "Establishing ssh connection with '{host}' "
-            "through port '{port}' as user '{username}'".format(
-                host=host, **kwargs
-            )
-        )
-
-        kwargs.setdefault("timeout", 1800)
-        self.timeout = kwargs["timeout"]
-
-        self._ssh = paramiko.SSHClient()
-
-        # Explicitly disable paramiko logger. Due to how paramiko dynamically
-        # loads loggers, it is not disabled by DVC disable_other_loggers().
-        # See https://github.com/iterative/dvc/issues/3482
-        self._ssh.set_log_channel("dvc.paramiko")
-        logging.getLogger("dvc.paramiko").disabled = True
-        logging.getLogger("dvc.paramiko.sftp").disabled = True
-
-        self._ssh.load_system_host_keys()
-        self._ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())
-
-        self._ssh.connect(host, *args, **kwargs)
-        self._ssh.get_transport().set_keepalive(10)
-        self._sftp_channels = []
-
-    @property
-    def sftp(self):
-        if not self._sftp_channels:
-            self._sftp_channels = [self._ssh.open_sftp()]
-        return self._sftp_channels[0]
-
-    def close(self):
-        for sftp in self._sftp_channels:
-            sftp.close()
-        self._ssh.close()
-
-    def st_mode(self, path):
-        lstat = None
-        with suppress(FileNotFoundError):
-            lstat = self.sftp.lstat(path)
-
-        return lstat.st_mode if lstat else 0
-
-    def info(self, path):
-        st = self.sftp.stat(path)
-        return {
-            "size": st.st_size,
-            "type": "dir" if stat.S_ISDIR(st.st_mode) else "file",
-        }
-
-    def getsize(self, path):
-        return self.info(path)["size"]
-
-    def exists(self, path):
-        return bool(self.st_mode(path))
-
-    def isdir(self, path):
-        return stat.S_ISDIR(self.st_mode(path))
-
-    def isfile(self, path):
-        return stat.S_ISREG(self.st_mode(path))
-
-    def islink(self, path):
-        return stat.S_ISLNK(self.st_mode(path))
-
-    def makedirs(self, path):
-        # Single stat call will say whether this is a dir, a file or a link
-        st_mode = self.st_mode(path)
-
-        if stat.S_ISDIR(st_mode):
-            return
-
-        if stat.S_ISREG(st_mode) or stat.S_ISLNK(st_mode):
-            raise DvcException(
-                f"a file with the same name '{path}' already exists"
-            )
-
-        head, tail = posixpath.split(path)
-
-        if head:
-            self.makedirs(head)
-
-        if tail:
-            try:
-                self.sftp.mkdir(path)
-            except OSError as exc:
-                # Since paramiko errors are very vague we need to recheck
-                # whether it's because path already exists or something else
-                if exc.errno == errno.EACCES or not self.exists(path):
-                    raise DvcException(
-                        f"unable to create remote directory '{path}'"
-                    ) from exc
-
-    def walk(self, directory, topdown=True):
-        # NOTE: original os.walk() implementation [1] with default options was
-        # used as a template.
-        #
-        # [1] https://github.com/python/cpython/blob/master/Lib/os.py
-        try:
-            dir_entries = self.sftp.listdir_attr(directory)
-        except OSError as exc:
-            raise DvcException(
-                "couldn't get the '{}' remote directory files list".format(
-                    directory
-                )
-            ) from exc
-
-        dirs = []
-        nondirs = []
-        for entry in dir_entries:
-            name = entry.filename
-            if stat.S_ISDIR(entry.st_mode):
-                dirs.append(name)
-            else:
-                nondirs.append(name)
-
-        if topdown:
-            yield directory, dirs, nondirs
-
-        for dname in dirs:
-            newpath = posixpath.join(directory, dname)
-            yield from self.walk(newpath, topdown=topdown)
-
-        if not topdown:
-            yield directory, dirs, nondirs
-
-    def walk_files(self, directory):
-        for root, _, files in self.walk(directory):
-            for fname in files:
-                yield posixpath.join(root, fname)
-
-    def _remove_file(self, path):
-        with suppress(FileNotFoundError):
-            self.sftp.remove(path)
-
-    def _remove_dir(self, path):
-        for root, dirs, files in self.walk(path, topdown=False):
-            for fname in files:
-                with suppress(FileNotFoundError):
-                    self._remove_file(posixpath.join(root, fname))
-
-            for dname in dirs:
-                with suppress(FileNotFoundError):
-                    self.sftp.rmdir(posixpath.join(root, dname))
-
-        with suppress(FileNotFoundError):
-            self.sftp.rmdir(path)
-
-    def remove(self, path):
-        if self.isdir(path):
-            self._remove_dir(path)
-        else:
-            self._remove_file(path)
-
-    def download(self, src, dest, no_progress_bar=False, progress_title=None):
-        with Tqdm(
-            desc=progress_title or os.path.basename(src),
-            disable=no_progress_bar,
-            bytes=True,
-            total=self.getsize(src),
-        ) as pbar:
-            self.sftp.get(src, dest, callback=pbar.update_to)
-
-    def move(self, src, dst):
-        """Rename src to dst, if it is not possible (in case src and dst are
-        on different filesystems) and actual physical copying of data is
-        happening.
-        """
-        self.makedirs(posixpath.dirname(dst))
-
-        try:
-            self.sftp.rename(src, dst)
-        except OSError:
-            self.atomic_copy(src, dst)
-            self.remove(src)
-
-    def atomic_copy(self, src, dst):
-        tmp = tmp_fname(dst)
-
-        try:
-            self.copy(src, tmp)
-            self.sftp.rename(tmp, dst)
-        finally:
-            self.remove(tmp)
-
-    def upload(self, src, dest, no_progress_bar=False, progress_title=None):
-        self.makedirs(posixpath.dirname(dest))
-        tmp_file = tmp_fname(dest)
-        if not progress_title:
-            progress_title = posixpath.basename(dest)
-
-        with Tqdm(
-            desc=progress_title,
-            disable=no_progress_bar,
-            bytes=True,
-            total=os.path.getsize(src),
-        ) as pbar:
-            self.sftp.put(src, tmp_file, callback=pbar.update_to)
-
-        self.sftp.rename(tmp_file, dest)
-
-    def execute(self, cmd):
-        stdin, stdout, stderr = self._ssh.exec_command(cmd)
-        channel = stdout.channel
-
-        stdin.close()
-        channel.shutdown_write()
-
-        stdout_chunks = []
-        stderr_chunks = []
-        while (
-            not channel.closed
-            or channel.recv_ready()
-            or channel.recv_stderr_ready()
-        ):
-            import selectors
-
-            selector = selectors.DefaultSelector()
-            selector.register(stdout.channel, selectors.EVENT_READ)
-
-            got_chunk = False
-            events = selector.select(self.timeout)
-            for key, _ in events:
-                c = key.fileobj
-                if c.recv_ready():
-                    stdout_chunks.append(stdout.channel.recv(len(c.in_buffer)))
-                    got_chunk = True
-
-                if c.recv_stderr_ready():
-                    stderr_len = len(c.in_stderr_buffer)
-                    s = stderr.channel.recv_stderr(stderr_len)
-                    stderr_chunks.append(s)
-                    got_chunk = True
-
-            if (
-                not got_chunk
-                and stdout.channel.exit_status_ready()
-                and not stderr.channel.recv_stderr_ready()
-                and not stdout.channel.recv_ready()
-            ):
-                stdout.channel.shutdown_read()
-                stdout.channel.close()
-                break
-
-        stdout.close()
-        stderr.close()
-
-        ret = stdout.channel.recv_exit_status()
-        if ret != 0:
-            err = b"".join(stderr_chunks).decode("utf-8")
-            raise RemoteCmdError("ssh", cmd, ret, err)
-
-        return b"".join(stdout_chunks).decode("utf-8")
-
-    @cached_property
-    def uname(self):
-        return self.execute("uname").strip()
-
-    def md5(self, path):
-        """
-        Use different md5 commands depending on the OS:
-
-         - Darwin's `md5` returns BSD-style checksums by default
-         - Linux's `md5sum` needs the `--tag` flag for a similar output
-
-         Example:
-              MD5 (foo.txt) = f3d220a856b52aabbf294351e8a24300
-        """
-        path = shlex.quote(path)
-        if self.uname == "Linux":
-            md5 = self.execute("md5sum " + path).split()[0]
-        elif self.uname == "Darwin":
-            md5 = self.execute("md5 " + path).split()[-1]
-        else:
-            raise DvcException(
-                f"'{self.uname}' is not supported as a SSH remote"
-            )
-
-        assert len(md5) == 32
-        return md5
-
-    def copy(self, src, dest):
-        dest = shlex.quote(dest)
-        src = shlex.quote(src)
-        self.execute(f"cp {src} {dest}")
-
-    def open_max_sftp_channels(self):
-        # If there are more than 1 it means we've already opened max amount
-        if len(self._sftp_channels) <= 1:
-            while True:
-                try:
-                    self._sftp_channels.append(self._ssh.open_sftp())
-                except paramiko.ssh_exception.ChannelException:
-                    if not self._sftp_channels:
-                        raise
-                    break
-        return self._sftp_channels
-
-    def open(self, *args, **kwargs):
-        return self.sftp.open(*args, **kwargs)
-
-    def symlink(self, src, dest):
-        self.sftp.symlink(src, dest)
-
-    def reflink(self, src, dest):
-        dest = shlex.quote(dest)
-        src = shlex.quote(src)
-        if self.uname == "Linux":
-            return self.execute(f"cp --reflink {src} {dest}")
-
-        if self.uname == "Darwin":
-            return self.execute(f"cp -c {src} {dest}")
-
-        raise DvcException(f"'{self.uname}' is not supported as a SSH remote")
-
-    def hardlink(self, src, dest):
-        dest = shlex.quote(dest)
-        src = shlex.quote(src)
-        self.execute(f"ln {src} {dest}")

dvc/fs/webdav.py

@@ -1,270 +1,94 @@
-import io
 import logging
 import os
 import threading
-from collections import deque
+from functools import lru_cache
 
-from funcy import cached_property, nullcontext, wrap_prop
+from funcy import cached_property, wrap_prop
 
-from dvc.config import ConfigError
-from dvc.exceptions import DvcException
-from dvc.path_info import HTTPURLInfo, WebDAVURLInfo
-from dvc.progress import Tqdm
+from dvc.path_info import WebDAVURLInfo
 from dvc.scheme import Schemes
 
-from .base import BaseFileSystem
+from .fsspec_wrapper import FSSpecWrapper
 from .http import ask_password
 
 logger = logging.getLogger(__name__)
 
 
-class WebDAVConnectionError(DvcException):
-    def __init__(self, host):
-        super().__init__(f"Unable to connect to WebDAV {host}.")
-
-
-class WebDAVFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
-    # Use webdav scheme
+class WebDAVFileSystem(FSSpecWrapper):  # pylint:disable=abstract-method
     scheme = Schemes.WEBDAV
-
-    # URLInfo for Webdav ~ replaces webdav -> http
     PATH_CLS = WebDAVURLInfo
-
-    # Traversable as walk_files is implemented
     CAN_TRAVERSE = True
-
-    # Length of walk_files prefix
     TRAVERSE_PREFIX_LEN = 2
-
-    # Implementation based on webdav3.client
-    REQUIRES = {"webdavclient3": "webdav3.client"}
-
-    # Chunk size for buffered upload/download with progress bar
-    CHUNK_SIZE = 2 ** 16
-
+    REQUIRES = {"webdav4": "webdav4"}
     PARAM_CHECKSUM = "etag"
     DETAIL_FIELDS = frozenset(("etag", "size"))
 
-    # Constructor
     def __init__(self, **config):
-        # Call BaseFileSystem constructor
         super().__init__(**config)
 
-        # Get username from configuration
-        self.user = config.get("user", None)
+        cert_path = config.get("cert_path", None)
+        key_path = config.get("key_path", None)
+        self.prefix = config.get("prefix", "")
+        self.fs_args.update(
+            {
+                "base_url": config["url"],
+                "cert": cert_path if not key_path else (cert_path, key_path),
+                "verify": config.get("ssl_verify", True),
+                "timeout": config.get("timeout", 30),
+            }
+        )
 
-        # Get password from configuration (might be None ~ not set)
-        self.password = config.get("password", None)
-
-        # Whether to ask for password if it is not set
-        self.ask_password = config.get("ask_password", False)
-
-        # Use token for webdav auth
-        self.token = config.get("token", None)
-
-        # Path to certificate
-        self.cert_path = config.get("cert_path", None)
-
-        # Path to private key
-        self.key_path = config.get("key_path", None)
-
-        # Connection timeout
-        self.timeout = config.get("timeout", 30)
-
-        # Get URL from configuration
-        self.url = config.get("url", None)
-
-        # If URL in config parse path_info
-        if self.url:
-            self.path_info = self.PATH_CLS(self.url)
-
-            # If username not specified try to use from URL
-            if self.user is None and self.path_info.user is not None:
-                self.user = self.path_info.user
-
-            # Construct hostname from path_info by stripping path
-            http_info = HTTPURLInfo(self.path_info.url)
-            self.hostname = http_info.replace(path="").url
-
-    # Webdav client
-    @wrap_prop(threading.Lock())
-    @cached_property
-    def _client(self):
-        from webdav3.client import Client
-
-        # Set password or ask for it
-        if self.ask_password and self.password is None and self.token is None:
-            self.password = ask_password(self.hostname, self.user)
-
-        # Setup webdav client options dictionary
-        options = {
-            "webdav_hostname": self.hostname,
-            "webdav_login": self.user,
-            "webdav_password": self.password,
-            "webdav_token": self.token,
-            "webdav_cert_path": self.cert_path,
-            "webdav_key_path": self.key_path,
-            "webdav_timeout": self.timeout,
-            "webdav_chunk_size": self.CHUNK_SIZE,
+    @staticmethod
+    def _get_kwargs_from_urls(urlpath):
+        path_info = WebDAVURLInfo(urlpath)
+        return {
+            "prefix": path_info.path.rstrip("/"),
+            "host": path_info.replace(path="").url,
+            "url": path_info.url.rstrip("/"),
+            "user": path_info.user,
         }
 
-        client = Client(options)
+    def _prepare_credentials(self, **config):
+        user = config.get("user", None)
+        password = config.get("password", None)
 
-        # Check whether client options are valid
-        if not client.valid():
-            raise ConfigError(
-                f"Configuration for WebDAV {self.hostname} is invalid."
-            )
+        headers = {}
+        token = config.get("token")
+        auth = None
+        if token:
+            headers.update({"Authorization": f"Bearer {token}"})
+        elif user:
+            if not password and config.get("ask_password"):
+                password = ask_password(config["host"], user)
+            auth = (user, password)
 
-        # Check whether connection is valid (root should always exist)
-        if not client.check(self.path_info.path):
-            raise WebDAVConnectionError(self.hostname)
+        return {"headers": headers, "auth": auth}
 
-        return client
+    @wrap_prop(threading.Lock())
+    @cached_property
+    def fs(self):
+        from webdav4.fsspec import WebdavFileSystem
 
-    def open(self, path_info, mode="r", encoding=None, **kwargs):
-        from webdav3.exceptions import RemoteResourceNotFound
+        return WebdavFileSystem(**self.fs_args)
 
-        assert mode in {"r", "rt", "rb"}
+    def _upload_fobj(self, fobj, to_info, size: int = None):
+        rpath = self.translate_path_info(to_info)
+        self.makedirs(os.path.dirname(rpath))
+        # using upload_fileobj to directly upload fileobj rather than buffering
+        # and using overwrite=True to avoid check for an extra exists call,
+        # as caller should ensure that the file does not exist beforehand.
+        return self.fs.client.upload_fileobj(
+            fobj, rpath, overwrite=True, size=size
+        )
 
-        fobj = io.BytesIO()
+    @lru_cache(512)
+    def translate_path_info(self, path):
+        if isinstance(path, self.PATH_CLS):
+            return path.path[len(self.prefix) :].lstrip("/")
+        return path
 
-        try:
-            self._client.download_from(buff=fobj, remote_path=path_info.path)
-        except RemoteResourceNotFound as exc:
-            raise FileNotFoundError from exc
+    _with_bucket = translate_path_info
 
-        fobj.seek(0)
 
-        if "mode" == "rb":
-            return fobj
-
-        return io.TextIOWrapper(fobj, encoding=encoding)
-
-    # Checks whether file/directory exists at remote
-    def exists(self, path_info) -> bool:
-        # Use webdav check to test for file existence
-        return self._client.check(path_info.path)
-
-    # Checks whether path points to directory
-    def isdir(self, path_info):
-        # Use webdav is_dir to test whether path points to a directory
-        return self._client.is_dir(path_info.path)
-
-    def walk_files(self, path_info, **kwargs):
-        if not self.exists(path_info):
-            return
-
-        for file in self.find(path_info):
-            yield path_info.replace(path=file)
-
-    def ls(self, path_info, detail=False):
-        for entry in self._client.list(path_info.path):
-            path = entry["path"]
-            if detail:
-                if entry["isdir"]:
-                    yield {"type": "directory", "name": path}
-                else:
-                    yield {
-                        "type": "file",
-                        "name": path,
-                        "size": entry["size"],
-                        "etag": entry["etag"],
-                    }
-            else:
-                yield path
-
-    def find(self, path_info, detail=False):
-        dirs = deque([path_info.path])
-
-        while dirs:
-            for entry in self._client.list(dirs.pop(), get_info=True):
-                path = entry["path"]
-                if entry["isdir"]:
-                    dirs.append(path)
-                    continue
-
-                if detail:
-                    yield {
-                        "type": "file",
-                        "name": path,
-                        "size": entry["size"],
-                        "etag": entry["etag"],
-                    }
-                else:
-                    yield path
-
-    # Removes file/directory
-    def remove(self, path_info):
-        # Use webdav client clean (DELETE) method to remove file/directory
-        self._client.clean(path_info.path)
-
-    # Creates directories
-    def makedirs(self, path_info):
-        # Terminate recursion
-        if path_info.path == self.path_info.path or self.exists(path_info):
-            return
-
-        # Recursively descent to root
-        self.makedirs(path_info.parent)
-
-        # Construct directory at current recursion depth
-        self._client.mkdir(path_info.path)
-
-    # Moves file/directory at remote
-    def move(self, from_info, to_info):
-        # Webdav client move
-        self._client.move(from_info.path, to_info.path)
-
-    def _upload_fobj(self, fobj, to_info):
-        # In contrast to other upload_fobj implementations, this one does not
-        # exactly do a chunked-upload but rather put everything in one request.
-        self.makedirs(to_info.parent)
-        self._client.upload_to(buff=fobj, remote_path=to_info.path)
-
-    # Downloads file from remote to file
-    def _download(self, from_info, to_file, name=None, no_progress_bar=False):
-        # Progress from HTTPFileSystem
-        with open(to_file, "wb") as fd:
-            with Tqdm.wrapattr(
-                fd,
-                "write",
-                total=None if no_progress_bar else self.getsize(from_info),
-                leave=False,
-                desc=from_info.url if name is None else name,
-                disable=no_progress_bar,
-            ) as fd_wrapped:
-                # Download from WebDAV via buffer
-                self._client.download_from(
-                    buff=fd_wrapped, remote_path=from_info.path
-                )
-
-    # Uploads file to remote
-    def _upload(
-        self, from_file, to_info, name=None, no_progress_bar=False, **_kwargs
-    ):
-        # First try to create parent directories
-        self.makedirs(to_info.parent)
-
-        file_size = os.path.getsize(from_file)
-        with open(from_file, "rb") as fd:
-            progress_context = (
-                nullcontext(fd)
-                if file_size == 0
-                else Tqdm.wrapattr(
-                    fd,
-                    "read",
-                    total=None if no_progress_bar else file_size,
-                    leave=False,
-                    desc=to_info.url if name is None else name,
-                    disable=no_progress_bar,
-                )
-            )
-            with progress_context as fd_wrapped:
-                self._client.upload_to(
-                    buff=fd_wrapped, remote_path=to_info.path
-                )
-
-    def info(self, path_info):
-        info = self._client.info(path_info.path)
-        return {"size": int(info["size"]), "etag": info["etag"]}
+class WebDAVSFileSystem(WebDAVFileSystem):  # pylint:disable=abstract-method
+    scheme = Schemes.WEBDAVS

dvc/fs/webdavs.py

@@ -1,7 +0,0 @@
-from dvc.scheme import Schemes
-
-from .webdav import WebDAVFileSystem
-
-
-class WebDAVSFileSystem(WebDAVFileSystem):  # pylint:disable=abstract-method
-    scheme = Schemes.WEBDAVS

dvc/fs/webhdfs.py

@@ -4,7 +4,6 @@ import posixpath
 import shutil
 import threading
 from contextlib import contextmanager
-from urllib.parse import urlparse
 
 from funcy import cached_property, wrap_prop
 
@@ -40,23 +39,24 @@ class WebHDFSFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
     def __init__(self, **config):
         super().__init__(**config)
 
-        self.path_info = None
-        url = config.get("url")
-        if not url:
-            return
-
-        self.path_info = self.PATH_CLS(url)
-
-        parsed = urlparse(url)
-
-        self.host = parsed.hostname
-        self.user = parsed.username or config.get("user")
-        self.port = parsed.port
+        self.host = config["host"]
+        self.user = config.get("user")
+        self.port = config.get("port")
 
         self.hdfscli_config = config.get("hdfscli_config")
         self.token = config.get("webhdfs_token")
         self.alias = config.get("webhdfs_alias")
 
+    @staticmethod
+    def _get_kwargs_from_urls(urlpath):
+        from fsspec.implementations.webhdfs import WebHDFS
+
+        return (
+            WebHDFS._get_kwargs_from_urls(  # pylint:disable=protected-access
+                urlpath
+            )
+        )
+
     @wrap_prop(threading.Lock())
     @cached_property
     def hdfs_client(self):
@@ -123,7 +123,7 @@ class WebHDFSFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
 
     def info(self, path_info):
         st = self.hdfs_client.status(path_info.path)
-        return {"size": st["length"]}
+        return {"size": st["length"], "type": "file"}
 
     def checksum(self, path_info):
         return HashInfo(
@@ -141,7 +141,7 @@ class WebHDFSFileSystem(BaseFileSystem):  # pylint:disable=abstract-method
         self.hdfs_client.makedirs(to_info.parent.path)
         self.hdfs_client.rename(from_info.path, to_info.path)
 
-    def _upload_fobj(self, fobj, to_info):
+    def _upload_fobj(self, fobj, to_info, **kwargs):
         with self.hdfs_client.write(to_info.path) as fdest:
             shutil.copyfileobj(fobj, fdest)
 

dvc/hash_info.py

@@ -14,6 +14,7 @@ class HashInfo:
     value: Optional[str]
     size: Optional[int] = field(default=None, compare=False)
     nfiles: Optional[int] = field(default=None, compare=False)
+    obj_name: Optional[str] = field(default=None, compare=False)
 
     def __bool__(self):
         return bool(self.value)
@@ -21,6 +22,9 @@ class HashInfo:
     def __str__(self):
         return f"{self.name}: {self.value}"
 
+    def __hash__(self):
+        return hash((self.name, self.value))
+
     @classmethod
     def from_dict(cls, d):
         _d = d.copy() if d else {}

dvc/ignore.py

@@ -57,7 +57,7 @@ class DvcIgnorePatterns(DvcIgnore):
         dirname = os.path.normpath(os.path.dirname(path))
         with fs.open(path, encoding="utf-8") as fobj:
             path_spec_lines = [
-                PatternInfo(line, "{}:{}:{}".format(name, line_no + 1, line))
+                PatternInfo(line, f"{name}:{line_no + 1}:{line}")
                 for line_no, line in enumerate(
                     map(str.strip, fobj.readlines())
                 )
@@ -166,7 +166,7 @@ class DvcIgnoreFilter:
             ".hg/",
             ".git/",
             ".git",
-            "{}/".format(Repo.DVC_DIR),
+            f"{Repo.DVC_DIR}/",
         ]
 
         self.fs = fs

dvc/info.py

@@ -20,6 +20,11 @@ if PKG is None:
 else:
     package = f"({PKG})"
 
+try:
+    import importlib.metadata as importlib_metadata
+except ImportError:  # < 3.8
+    import importlib_metadata  # type: ignore[no-redef]
+
 
 def get_dvc_info():
     info = [
@@ -27,7 +32,7 @@ def get_dvc_info():
         "---------------------------------",
         f"Platform: Python {platform.python_version()} on "
         f"{platform.platform()}",
-        f"Supports: {_get_supported_remotes()}",
+        f"Supports:{_get_supported_remotes()}",
     ]
 
     try:
@@ -37,9 +42,7 @@ def get_dvc_info():
             # later decides to enable shared cache mode with
             # `dvc config cache.shared group`.
             if os.path.exists(repo.odb.local.cache_dir):
-                info.append(
-                    "Cache types: {}".format(_get_linktype_support_info(repo))
-                )
+                info.append(f"Cache types: {_get_linktype_support_info(repo)}")
                 fs_type = get_fs_type(repo.odb.local.cache_dir)
                 info.append(f"Cache directory: {fs_type}")
             else:
@@ -51,7 +54,7 @@ def get_dvc_info():
             root_directory = repo.root_dir
             fs_root = get_fs_type(os.path.abspath(root_directory))
             info.append(f"Workspace directory: {fs_root}")
-            info.append("Repo: {}".format(_get_dvc_repo_info(repo)))
+            info.append(f"Repo: {_get_dvc_repo_info(repo)}")
     except NotDvcRepoError:
         pass
     except SCMError:
@@ -113,19 +116,23 @@ def _get_linktype_support_info(repo):
 
 
 def _get_supported_remotes():
-
     supported_remotes = []
     for scheme, fs_cls in FS_MAP.items():
         if not fs_cls.get_missing_deps():
-            supported_remotes.append(scheme)
+            dependencies = []
+            for requirement in fs_cls.REQUIRES:
+                dependencies.append(
+                    f"{requirement} = "
+                    f"{importlib_metadata.version(requirement)}"
+                )
 
-    if len(supported_remotes) == len(FS_MAP):
-        return "All remotes"
+            remote_info = scheme
+            if dependencies:
+                remote_info += " (" + ", ".join(dependencies) + ")"
+            supported_remotes.append(remote_info)
 
-    if len(supported_remotes) == 1:
-        return supported_remotes
-
-    return ", ".join(supported_remotes)
+    assert len(supported_remotes) >= 1
+    return "\n\t" + ",\n\t".join(supported_remotes)
 
 
 def get_fs_type(path):

dvc/istextfile.py

@@ -5,6 +5,7 @@
 
 
 TEXT_CHARS = bytes(range(32, 127)) + b"\n\r\t\f\b"
+DEFAULT_CHUNK_SIZE = 512
 
 
 def istextblock(block):
@@ -22,7 +23,7 @@ def istextblock(block):
     return float(len(nontext)) / len(block) <= 0.30
 
 
-def istextfile(fname, fs, blocksize=512):
+def istextfile(fname, fs, blocksize=DEFAULT_CHUNK_SIZE):
     """Uses heuristics to guess whether the given file is text or binary,
     by reading a single block of bytes from the file.
     If more than 30% of the chars in the block are non-text, or there

dvc/logger.py

@@ -62,7 +62,7 @@ def addLoggingLevel(levelName, levelNum, methodName=None):
 
 class LoggingException(Exception):
     def __init__(self, record):
-        msg = "failed to log {}".format(str(record))
+        msg = f"failed to log {str(record)}"
         super().__init__(msg)
 
 

dvc/objects/__init__.py

@@ -1,45 +1,34 @@
 import logging
-from concurrent.futures import ThreadPoolExecutor, as_completed
-
-from dvc.progress import Tqdm
+from typing import TYPE_CHECKING, Iterator, Union
 
 from .tree import Tree
 
+if TYPE_CHECKING:
+    from dvc.hash_info import HashInfo
+
+    from .db.base import ObjectDB
+    from .file import HashFile
+
 logger = logging.getLogger(__name__)
 
 
-def save(odb, obj, jobs=None, **kwargs):
-    if isinstance(obj, Tree):
-        with ThreadPoolExecutor(max_workers=jobs) as executor:
-            for future in Tqdm(
-                as_completed(
-                    executor.submit(
-                        odb.add,
-                        entry.path_info,
-                        entry.fs,
-                        entry.hash_info,
-                        **kwargs
-                    )
-                    for _, entry in obj
-                ),
-                desc="Saving files",
-                total=len(obj),
-                unit="file",
-            ):
-                future.result()
-
-    odb.add(obj.path_info, obj.fs, obj.hash_info, **kwargs)
-
-
-def check(odb, obj):
-    odb.check(obj.hash_info)
-
+def check(odb: "ObjectDB", obj: "HashFile", **kwargs):
     if isinstance(obj, Tree):
         for _, entry in obj:
-            odb.check(entry.hash_info)
+            odb.check(entry.hash_info, **kwargs)
+
+    odb.check(obj.hash_info, **kwargs)
 
 
-def load(odb, hash_info):
+def load(odb: "ObjectDB", hash_info: "HashInfo") -> "HashFile":
     if hash_info.isdir:
         return Tree.load(odb, hash_info)
     return odb.get(hash_info)
+
+
+def iterobjs(
+    obj: Union["Tree", "HashFile"]
+) -> Iterator[Union["Tree", "HashFile"]]:
+    if isinstance(obj, Tree):
+        yield from (entry_obj for _, entry_obj in obj)
+    yield obj

dvc/objects/db/__init__.py

@@ -1,24 +1,27 @@
-from collections import defaultdict
+from typing import TYPE_CHECKING
 
 from dvc.scheme import Schemes
 
+if TYPE_CHECKING:
+    from .index import ObjectDBIndexBase
 
-def get_odb(fs, **config):
+
+def get_odb(fs, path_info, **config):
     from .base import ObjectDB
     from .gdrive import GDriveObjectDB
     from .local import LocalObjectDB
-    from .ssh import SSHObjectDB
+    from .oss import OSSObjectDB
 
     if fs.scheme == Schemes.LOCAL:
-        return LocalObjectDB(fs, **config)
-
-    if fs.scheme == Schemes.SSH:
-        return SSHObjectDB(fs, **config)
+        return LocalObjectDB(fs, path_info, **config)
 
     if fs.scheme == Schemes.GDRIVE:
-        return GDriveObjectDB(fs, **config)
+        return GDriveObjectDB(fs, path_info, **config)
 
-    return ObjectDB(fs, **config)
+    if fs.scheme == Schemes.OSS:
+        return OSSObjectDB(fs, path_info, **config)
+
+    return ObjectDB(fs, path_info, **config)
 
 
 def _get_odb(repo, settings):
@@ -27,8 +30,22 @@ def _get_odb(repo, settings):
     if not settings:
         return None
 
-    cls, config = get_cloud_fs(repo, **settings)
-    return get_odb(cls(**config), state=repo.state, **config)
+    cls, config, path_info = get_cloud_fs(repo, **settings)
+    config["tmp_dir"] = repo.tmp_dir
+    return get_odb(cls(**config), path_info, state=repo.state, **config)
+
+
+def get_index(odb) -> "ObjectDBIndexBase":
+    import hashlib
+
+    from .index import ObjectDBIndex, ObjectDBIndexNoop
+
+    cls = ObjectDBIndex if odb.tmp_dir else ObjectDBIndexNoop
+    return cls(
+        odb.tmp_dir,
+        hashlib.sha256(odb.path_info.url.encode("utf-8")).hexdigest(),
+        odb.fs.CHECKSUM_DIR_SUFFIX,
+    )
 
 
 class ODBManager:
@@ -80,99 +97,3 @@ class ODBManager:
     def by_scheme(self):
         self._init_odb(self.CLOUD_SCHEMES)
         yield from self._odb.items()
-
-
-class NamedCacheItem:
-    def __init__(self):
-        self.names = set()
-        self.children = defaultdict(NamedCacheItem)
-
-    def __eq__(self, other):
-        return self.names == other.names and self.children == other.children
-
-    def child_keys(self):
-        for key, child in self.children.items():
-            yield key
-            yield from child.child_keys()
-
-    def child_names(self):
-        for key, child in self.children.items():
-            yield key, child.names
-            yield from child.child_names()
-
-    def add(self, checksum, item):
-        self.children[checksum].update(item)
-
-    def update(self, item, suffix=""):
-        if suffix:
-            self.names.update(n + suffix for n in item.names)
-        else:
-            self.names.update(item.names)
-        for checksum, child_item in item.children.items():
-            self.children[checksum].update(child_item)
-
-
-class NamedCache:
-    # pylint: disable=protected-access
-    def __init__(self):
-        self._items = defaultdict(lambda: defaultdict(NamedCacheItem))
-        self.external = defaultdict(set)
-
-    @classmethod
-    def make(cls, scheme, checksum, name):
-        cache = cls()
-        cache.add(scheme, checksum, name)
-        return cache
-
-    def __getitem__(self, key):
-        return self._items[key]
-
-    def add(self, scheme, checksum, name):
-        """Add a mapped name for the specified checksum."""
-        self._items[scheme][checksum].names.add(name)
-
-    def add_child_cache(self, checksum, cache, suffix=""):
-        """Add/update child cache for the specified checksum."""
-        for scheme, src in cache._items.items():
-            dst = self._items[scheme][checksum].children
-            for child_checksum, item in src.items():
-                dst[child_checksum].update(item, suffix=suffix)
-
-        for repo_pair, files in cache.external.items():
-            self.external[repo_pair].update(files)
-
-    def add_external(self, url, rev, path):
-        self.external[url, rev].add(path)
-
-    def update(self, cache, suffix=""):
-        for scheme, src in cache._items.items():
-            dst = self._items[scheme]
-            for checksum, item in src.items():
-                dst[checksum].update(item, suffix=suffix)
-
-        for repo_pair, files in cache.external.items():
-            self.external[repo_pair].update(files)
-
-    def scheme_keys(self, scheme):
-        """Iterate over a flat list of all keys for the specified scheme,
-        including children.
-        """
-        for key, item in self._items[scheme].items():
-            yield key
-            yield from item.child_keys()
-
-    def scheme_names(self, scheme):
-        """Iterate over a flat list of checksum, names items for the specified
-        scheme, including children.
-        """
-        for key, item in self._items[scheme].items():
-            yield key, item.names
-            yield from item.child_names()
-
-    def dir_keys(self, scheme):
-        return (
-            key for key, item in self._items[scheme].items() if item.children
-        )
-
-    def child_keys(self, scheme, checksum):
-        return self._items[scheme][checksum].child_keys()

dvc/objects/db/base.py

@@ -2,13 +2,20 @@ import itertools
 import logging
 import os
 from concurrent.futures import ThreadPoolExecutor
+from contextlib import suppress
 from copy import copy
-from typing import Optional
+from typing import TYPE_CHECKING, Optional
 
-from dvc.objects.errors import ObjectFormatError
+from dvc.fs.local import LocalFileSystem
+from dvc.objects.errors import ObjectDBPermissionError, ObjectFormatError
 from dvc.objects.file import HashFile
 from dvc.progress import Tqdm
 
+if TYPE_CHECKING:
+    from dvc.fs.base import BaseFileSystem
+    from dvc.hash_info import HashInfo
+    from dvc.types import AnyPath, DvcPath
+
 logger = logging.getLogger(__name__)
 
 
@@ -18,15 +25,42 @@ class ObjectDB:
     DEFAULT_CACHE_TYPES = ["copy"]
     CACHE_MODE: Optional[int] = None
 
-    def __init__(self, fs, **config):
+    def __init__(self, fs: "BaseFileSystem", path_info: "AnyPath", **config):
         from dvc.state import StateNoop
 
         self.fs = fs
+        self.path_info = path_info
         self.state = config.get("state", StateNoop())
         self.verify = config.get("verify", self.DEFAULT_VERIFY)
         self.cache_types = config.get("type") or copy(self.DEFAULT_CACHE_TYPES)
         self.cache_type_confirmed = False
         self.slow_link_warning = config.get("slow_link_warning", True)
+        self.tmp_dir = config.get("tmp_dir")
+        self.read_only = config.get("read_only", False)
+
+    @property
+    def config(self):
+        return {
+            "state": self.state,
+            "verify": self.verify,
+            "type": self.cache_types,
+            "slow_link_warning": self.slow_link_warning,
+            "tmp_dir": self.tmp_dir,
+            "read_only": self.read_only,
+        }
+
+    def __eq__(self, other):
+        return (
+            self.fs == other.fs
+            and self.path_info == other.path_info
+            and self.read_only == other.read_only
+        )
+
+    def __hash__(self):
+        return hash((self.fs.scheme, self.path_info))
+
+    def exists(self, hash_info: "HashInfo"):
+        return self.fs.exists(self.hash_to_path_info(hash_info.value))
 
     def move(self, from_info, to_info):
         self.fs.move(from_info, to_info)
@@ -34,7 +68,7 @@ class ObjectDB:
     def makedirs(self, path_info):
         self.fs.makedirs(path_info)
 
-    def get(self, hash_info):
+    def get(self, hash_info: "HashInfo"):
         """get raw object"""
         return HashFile(
             # Prefer string path over PathInfo when possible due to performance
@@ -43,23 +77,28 @@ class ObjectDB:
             hash_info,
         )
 
-    def add(self, path_info, fs, hash_info, move=True, **kwargs):
-        try:
-            self.check(hash_info, check_hash=self.verify)
-            return
-        except (ObjectFormatError, FileNotFoundError):
-            pass
-
-        cache_info = self.hash_to_path_info(hash_info.value)
-        # using our makedirs to create dirs with proper permissions
-        self.makedirs(cache_info.parent)
-        use_move = isinstance(fs, type(self.fs)) and move
+    def _add_file(
+        self,
+        from_fs: "BaseFileSystem",
+        from_info: "AnyPath",
+        to_info: "DvcPath",
+        _hash_info: "HashInfo",
+        move: bool = False,
+    ):
+        self.makedirs(to_info.parent)
+        use_move = isinstance(from_fs, type(self.fs)) and move
         try:
             if use_move:
-                self.fs.move(path_info, cache_info)
+                self.fs.move(from_info, to_info)
+            elif isinstance(from_fs, LocalFileSystem):
+                if not isinstance(from_info, from_fs.PATH_CLS):
+                    from_info = from_fs.PATH_CLS(from_info)
+                self.fs.upload(from_info, to_info)
+            elif isinstance(self.fs, LocalFileSystem):
+                from_fs.download_file(from_info, to_info)
             else:
-                with fs.open(path_info, mode="rb") as fobj:
-                    self.fs.upload_fobj(fobj, cache_info)
+                with from_fs.open(from_info, mode="rb") as fobj:
+                    self.fs.upload_fobj(fobj, to_info)
         except OSError as exc:
             # If the target file already exists, we are going to simply
             # ignore the exception (#4992).
@@ -73,21 +112,49 @@ class ObjectDB:
                 and exc.__context__
                 and isinstance(exc.__context__, FileExistsError)
             ):
-                logger.debug("'%s' file already exists, skipping", path_info)
+                logger.debug("'%s' file already exists, skipping", to_info)
                 if use_move:
-                    fs.remove(path_info)
+                    from_fs.remove(from_info)
             else:
                 raise
 
-        self.protect(cache_info)
-        self.state.save(cache_info, self.fs, hash_info)
+    def add(
+        self,
+        path_info: "AnyPath",
+        fs: "BaseFileSystem",
+        hash_info: "HashInfo",
+        move: bool = True,
+        verify: Optional[bool] = None,
+        **kwargs,
+    ):
+        if self.read_only:
+            raise ObjectDBPermissionError("Cannot add to read-only ODB")
+
+        if verify is None:
+            verify = self.verify
+        try:
+            self.check(hash_info, check_hash=verify)
+            return
+        except (ObjectFormatError, FileNotFoundError):
+            pass
+
+        cache_info = self.hash_to_path_info(hash_info.value)
+        self._add_file(fs, path_info, cache_info, hash_info, move)
+
+        try:
+            if verify:
+                self.check(hash_info, check_hash=True)
+            self.protect(cache_info)
+            self.state.save(cache_info, self.fs, hash_info)
+        except (ObjectFormatError, FileNotFoundError):
+            pass
 
         callback = kwargs.get("download_callback")
         if callback:
             callback(1)
 
-    def hash_to_path_info(self, hash_):
-        return self.fs.path_info / hash_[0:2] / hash_[2:]
+    def hash_to_path_info(self, hash_) -> "DvcPath":
+        return self.path_info / hash_[0:2] / hash_[2:]
 
     # Override to return path as a string instead of PathInfo for clouds
     # which support string paths (see local)
@@ -106,7 +173,11 @@ class ObjectDB:
     def set_exec(self, path_info):  # pylint: disable=unused-argument
         pass
 
-    def check(self, hash_info, check_hash=True):
+    def check(
+        self,
+        hash_info: "HashInfo",
+        check_hash: bool = True,
+    ):
         """Compare the given hash with the (corresponding) actual one if
         check_hash is specified, or just verify the existence of the cache
         files on the filesystem.
@@ -123,7 +194,7 @@ class ObjectDB:
 
         obj = self.get(hash_info)
         if self.is_protected(obj.path_info):
-            logger.trace(
+            logger.trace(  # type: ignore[attr-defined]
                 "Assuming '%s' is unchanged since it is read-only",
                 obj.path_info,
             )
@@ -133,22 +204,24 @@ class ObjectDB:
             obj.check(self, check_hash=check_hash)
         except ObjectFormatError:
             logger.warning("corrupted cache file '%s'.", obj.path_info)
-            self.fs.remove(obj.path_info)
+            with suppress(FileNotFoundError):
+                self.fs.remove(obj.path_info)
             raise
 
-        # making cache file read-only so we don't need to check it
-        # next time
-        self.protect(obj.path_info)
+        if check_hash:
+            # making cache file read-only so we don't need to check it
+            # next time
+            self.protect(obj.path_info)
 
     def _list_paths(self, prefix=None, progress_callback=None):
         if prefix:
             if len(prefix) > 2:
-                path_info = self.fs.path_info / prefix[:2] / prefix[2:]
+                path_info = self.path_info / prefix[:2] / prefix[2:]
             else:
-                path_info = self.fs.path_info / prefix[:2]
+                path_info = self.path_info / prefix[:2]
             prefix = True
         else:
-            path_info = self.fs.path_info
+            path_info = self.path_info
             prefix = False
         if progress_callback:
             for file_info in self.fs.walk_files(path_info, prefix=prefix):
@@ -315,15 +388,30 @@ class ObjectDB:
     def _remove_unpacked_dir(self, hash_):
         pass
 
-    def gc(self, used, jobs=None):
+    def gc(self, used, jobs=None, cache_odb=None, shallow=True):
+        from ..tree import Tree
+
+        if self.read_only:
+            raise ObjectDBPermissionError("Cannot gc read-only ODB")
+        if not cache_odb:
+            cache_odb = self
+        used_hashes = set()
+        for hash_info in used:
+            used_hashes.add(hash_info.value)
+            if hash_info.isdir and not shallow:
+                tree = Tree.load(cache_odb, hash_info)
+                used_hashes.update(
+                    entry_obj.hash_info.value for _, entry_obj in tree
+                )
+
         removed = False
         # hashes must be sorted to ensure we always remove .dir files first
         for hash_ in sorted(
-            self.all(jobs, str(self.fs.path_info)),
+            self.all(jobs, str(self.path_info)),
             key=self.fs.is_dir_hash,
             reverse=True,
         ):
-            if hash_ in used:
+            if hash_ in used_hashes:
                 continue
             path_info = self.hash_to_path_info(hash_)
             if self.fs.is_dir_hash(hash_):
@@ -339,9 +427,7 @@ class ObjectDB:
         """Return list of the specified hashes which exist in this fs.
         Hashes will be queried individually.
         """
-        logger.debug(
-            "Querying {} hashes via object_exists".format(len(hashes))
-        )
+        logger.debug(f"Querying {len(hashes)} hashes via object_exists")
         with Tqdm(
             desc="Querying "
             + ("cache in " + name if name else "remote cache"),
@@ -399,8 +485,15 @@ class ObjectDB:
         # hashes_exist() (see ssh, local)
         assert self.fs.TRAVERSE_PREFIX_LEN >= 2
 
+        # During the tests, for ensuring that the traverse behavior
+        # is working we turn on this option. It will ensure the
+        # list_hashes_traverse() is called.
+        always_traverse = getattr(self.fs, "_ALWAYS_TRAVERSE", False)
+
         hashes = set(hashes)
-        if len(hashes) == 1 or not self.fs.CAN_TRAVERSE:
+        if (
+            len(hashes) == 1 or not self.fs.CAN_TRAVERSE
+        ) and not always_traverse:
             remote_hashes = self.list_hashes_exists(hashes, jobs, name)
             return remote_hashes
 
@@ -418,7 +511,7 @@ class ObjectDB:
             )
         else:
             traverse_weight = traverse_pages
-        if len(hashes) < traverse_weight:
+        if len(hashes) < traverse_weight and not always_traverse:
             logger.debug(
                 "Large remote ('{}' hashes < '{}' traverse weight), "
                 "using object_exists for remaining hashes".format(
@@ -429,7 +522,7 @@ class ObjectDB:
                 hashes - remote_hashes, jobs, name
             )
 
-        logger.debug("Querying '{}' hashes via traverse".format(len(hashes)))
+        logger.debug(f"Querying '{len(hashes)}' hashes via traverse")
         remote_hashes = set(
             self.list_hashes_traverse(remote_size, remote_hashes, jobs, name)
         )

dvc/objects/db/index.py

@@ -0,0 +1,137 @@
+import logging
+import os
+from abc import ABC, abstractmethod
+from typing import TYPE_CHECKING, Iterable, Optional, Set
+
+from ..errors import ObjectDBError
+
+if TYPE_CHECKING:
+    from dvc.types import StrPath
+
+logger = logging.getLogger(__name__)
+
+
+class ObjectDBIndexBase(ABC):
+    @abstractmethod
+    def __init__(
+        self, tmp_dir: "StrPath", name: str, dir_suffix: Optional[str] = None
+    ):
+        pass
+
+    @abstractmethod
+    def __iter__(self):
+        pass
+
+    @abstractmethod
+    def __contains__(self, hash_):
+        pass
+
+    def hashes(self):
+        return iter(self)
+
+    @abstractmethod
+    def dir_hashes(self):
+        pass
+
+    @abstractmethod
+    def clear(self):
+        pass
+
+    @abstractmethod
+    def update(self, dir_hashes: Iterable[str], file_hashes: Iterable[str]):
+        pass
+
+    @abstractmethod
+    def intersection(self, hashes: Set[str]):
+        pass
+
+
+class ObjectDBIndexNoop(ObjectDBIndexBase):
+    """No-op class for ODBs which are not indexed."""
+
+    def __init__(
+        self, tmp_dir: "StrPath", name: str, dir_suffix: Optional[str] = None
+    ):  # pylint: disable=super-init-not-called
+        pass
+
+    def __iter__(self):
+        return iter([])
+
+    def __contains__(self, hash_):
+        return False
+
+    def dir_hashes(self):
+        return []
+
+    def clear(self):
+        pass
+
+    def update(self, dir_hashes: Iterable[str], file_hashes: Iterable[str]):
+        pass
+
+    def intersection(self, hashes: Set[str]):
+        return []
+
+
+class ObjectDBIndex(ObjectDBIndexBase):
+    """Class for indexing hashes in an ODB."""
+
+    INDEX_SUFFIX = ".idx"
+    INDEX_DIR = "index"
+
+    def __init__(
+        self, tmp_dir: "StrPath", name: str, dir_suffix: Optional[str] = None
+    ):  # pylint: disable=super-init-not-called
+        from diskcache import Index
+
+        from dvc.fs.local import LocalFileSystem
+        from dvc.utils.fs import makedirs
+
+        self.index_dir = os.path.join(tmp_dir, self.INDEX_DIR, name)
+        makedirs(self.index_dir, exist_ok=True)
+        self.fs = LocalFileSystem()
+        self.index = Index(self.index_dir)
+
+        if not dir_suffix:
+            dir_suffix = self.fs.CHECKSUM_DIR_SUFFIX
+        self.dir_suffix = dir_suffix
+
+    def __iter__(self):
+        return iter(self.index)
+
+    def __contains__(self, hash_):
+        return hash_ in self.index
+
+    def dir_hashes(self):
+        """Iterate over .dir hashes stored in the index."""
+        yield from (hash_ for hash_, is_dir in self.index.items() if is_dir)
+
+    def is_dir_hash(self, hash_: str):
+        return hash_.endswith(self.dir_suffix)
+
+    def clear(self):
+        """Clear this index (to force re-indexing later)."""
+        from diskcache import Timeout
+
+        try:
+            self.index.clear()
+        except Timeout as exc:
+            raise ObjectDBError("Failed to clear ODB index") from exc
+
+    def update(self, dir_hashes: Iterable[str], file_hashes: Iterable[str]):
+        """Update this index, adding the specified hashes."""
+        from diskcache import Timeout
+
+        try:
+            with self.index.transact():
+                for hash_ in dir_hashes:
+                    self.index[hash_] = True
+            with self.index.transact():
+                for hash_ in file_hashes:
+                    self.index[hash_] = False
+        except Timeout as exc:
+            raise ObjectDBError("Failed to update ODB index") from exc
+
+    def intersection(self, hashes: Set[str]):
+        """Iterate over values from `hashes` which exist in the index."""
+        yield from hashes.intersection(self.index.keys())

dvc/objects/db/local.py

@@ -22,9 +22,9 @@ class LocalObjectDB(ObjectDB):
     CACHE_MODE = 0o444
     UNPACKED_DIR_SUFFIX = ".unpacked"
 
-    def __init__(self, fs, **config):
-        super().__init__(fs, **config)
-        self.cache_dir = config.get("url")
+    def __init__(self, fs, path_info, **config):
+        super().__init__(fs, path_info, **config)
+        self.cache_dir = path_info.fspath
 
         shared = config.get("shared")
         if shared:
@@ -36,11 +36,11 @@ class LocalObjectDB(ObjectDB):
 
     @property
     def cache_dir(self):
-        return self.fs.path_info.fspath if self.fs.path_info else None
+        return self.path_info.fspath if self.path_info else None
 
     @cache_dir.setter
     def cache_dir(self, value):
-        self.fs.path_info = PathInfo(value) if value else None
+        self.path_info = PathInfo(value) if value else None
 
     @cached_property
     def cache_path(self):
@@ -81,13 +81,13 @@ class LocalObjectDB(ObjectDB):
         return ret
 
     def _list_paths(self, prefix=None, progress_callback=None):
-        assert self.fs.path_info is not None
+        assert self.path_info is not None
         if prefix:
-            path_info = self.fs.path_info / prefix[:2]
+            path_info = self.path_info / prefix[:2]
             if not self.fs.exists(path_info):
                 return
         else:
-            path_info = self.fs.path_info
+            path_info = self.path_info
         # NOTE: use utils.fs walk_files since fs.walk_files will not follow
         # symlinks
         if progress_callback:
@@ -112,7 +112,7 @@ class LocalObjectDB(ObjectDB):
             # would get only the part of file. So, at first, the file should be
             # copied with the temporary name, and then original file should be
             # replaced by new.
-            copyfile(path, tmp, name="Unprotecting '{}'".format(relpath(path)))
+            copyfile(path, tmp, name=f"Unprotecting '{relpath(path)}'")
             remove(path)
             os.rename(tmp, path)
 

dvc/objects/db/oss.py

@@ -0,0 +1,9 @@
+from .base import ObjectDB
+
+
+class OSSObjectDB(ObjectDB):
+    """
+    Temporary extra verification
+    """
+
+    DEFAULT_VERIFY = True

dvc/objects/db/reference.py

@@ -0,0 +1,84 @@
+import io
+import logging
+import os
+from typing import TYPE_CHECKING, Dict
+
+from dvc.scheme import Schemes
+
+from ..errors import ObjectFormatError
+from ..reference import ReferenceHashFile
+from .base import ObjectDB
+
+if TYPE_CHECKING:
+    from dvc.fs.base import BaseFileSystem
+    from dvc.hash_info import HashInfo
+    from dvc.types import AnyPath, DvcPath
+
+logger = logging.getLogger(__name__)
+
+
+class ReferenceObjectDB(ObjectDB):
+    """Reference ODB.
+
+    File objects are stored as ReferenceHashFiles which reference paths outside
+    of the staging ODB fs. Tree objects are stored natively.
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self._fs_cache: Dict[tuple, "BaseFileSystem"] = {}
+        self._obj_cache: Dict["HashInfo", "ReferenceHashFile"] = {}
+
+    def get(self, hash_info: "HashInfo"):
+        if hash_info.isdir:
+            return super().get(hash_info)
+        try:
+            return self._obj_cache[hash_info]
+        except KeyError:
+            pass
+        path_info = self.hash_to_path(hash_info.value)
+        try:
+            with self.fs.open(path_info, "rb") as fobj:
+                ref_file = ReferenceHashFile.from_bytes(
+                    fobj.read(), fs_cache=self._fs_cache
+                )
+        except OSError:
+            raise FileNotFoundError
+        try:
+            ref_file.check(self, check_hash=False)
+        except ObjectFormatError:
+            self.fs.remove(path_info)
+            raise
+        self._obj_cache[hash_info] = ref_file
+        return ref_file
+
+    def _add_file(
+        self,
+        from_fs: "BaseFileSystem",
+        from_info: "AnyPath",
+        to_info: "DvcPath",
+        hash_info: "HashInfo",
+        move: bool = False,
+    ):
+        self.makedirs(to_info.parent)
+        if hash_info.isdir:
+            return super()._add_file(
+                from_fs, from_info, to_info, hash_info, move
+            )
+        ref_file = ReferenceHashFile(from_info, from_fs, hash_info)
+        self._obj_cache[hash_info] = ref_file
+        ref_fobj = io.BytesIO(ref_file.to_bytes())
+        ref_fobj.seek(0)
+        try:
+            self.fs.upload_fobj(ref_fobj, to_info)
+        except OSError as exc:
+            if isinstance(exc, FileExistsError) or (
+                os.name == "nt"
+                and exc.__context__
+                and isinstance(exc.__context__, FileExistsError)
+            ):
+                logger.debug("'%s' file already exists, skipping", to_info)
+            else:
+                raise
+        if from_fs.scheme != Schemes.LOCAL:
+            self._fs_cache[ReferenceHashFile.config_tuple(from_fs)] = from_fs

dvc/objects/db/slow_link_detection.py

@@ -25,15 +25,15 @@ this.message = (
 
 def slow_link_guard(f):
     @wraps(f)
-    def wrapper(remote, *args, **kwargs):
+    def wrapper(odb, *args, **kwargs):
         if this.already_displayed:
-            return f(remote, *args, **kwargs)
+            return f(odb, *args, **kwargs)
 
-        if not remote.slow_link_warning or remote.cache_types:
-            return f(remote, *args, **kwargs)
+        if not odb.slow_link_warning or odb.cache_types:
+            return f(odb, *args, **kwargs)
 
         start = time.time()
-        result = f(remote, *args, **kwargs)
+        result = f(odb, *args, **kwargs)
         delta = time.time() - start
 
         if delta >= this.timeout_seconds:

dvc/objects/db/ssh.py

@@ -1,90 +0,0 @@
-import errno
-import itertools
-import logging
-import posixpath
-from concurrent.futures import ThreadPoolExecutor
-
-from dvc.progress import Tqdm
-from dvc.utils import to_chunks
-
-from .base import ObjectDB
-
-logger = logging.getLogger(__name__)
-
-
-class SSHObjectDB(ObjectDB):
-    def batch_exists(self, path_infos, callback):
-        def _exists(chunk_and_channel):
-            chunk, channel = chunk_and_channel
-            ret = []
-            for path in chunk:
-                try:
-                    channel.stat(path)
-                    ret.append(True)
-                except OSError as exc:
-                    if exc.errno != errno.ENOENT:
-                        raise
-                    ret.append(False)
-                callback(path)
-            return ret
-
-        with self.fs.ssh(path_infos[0]) as ssh:
-            channels = ssh.open_max_sftp_channels()
-            max_workers = len(channels)
-
-            with ThreadPoolExecutor(max_workers=max_workers) as executor:
-                paths = [path_info.path for path_info in path_infos]
-                chunks = to_chunks(paths, num_chunks=max_workers)
-                chunks_and_channels = zip(chunks, channels)
-                outcome = executor.map(_exists, chunks_and_channels)
-                results = list(itertools.chain.from_iterable(outcome))
-
-            return results
-
-    def hashes_exist(self, hashes, jobs=None, name=None):
-        """This is older implementation used in remote/base.py
-        We are reusing it in RemoteSSH, because SSH's batch_exists proved to be
-        faster than current approach (relying on exists(path_info)) applied in
-        remote/base.
-        """
-        if not self.fs.CAN_TRAVERSE:
-            return list(set(hashes) & set(self.all()))
-
-        # possibly prompt for credentials before "Querying" progress output
-        self.fs.ensure_credentials()
-
-        with Tqdm(
-            desc="Querying "
-            + ("cache in " + name if name else "remote cache"),
-            total=len(hashes),
-            unit="file",
-        ) as pbar:
-
-            def exists_with_progress(chunks):
-                return self.batch_exists(chunks, callback=pbar.update_msg)
-
-            with ThreadPoolExecutor(
-                max_workers=jobs or self.fs.jobs
-            ) as executor:
-                path_infos = [self.hash_to_path_info(x) for x in hashes]
-                chunks = to_chunks(path_infos, num_chunks=self.fs.jobs)
-                results = executor.map(exists_with_progress, chunks)
-                in_remote = itertools.chain.from_iterable(results)
-                ret = list(itertools.compress(hashes, in_remote))
-                return ret
-
-    def _list_paths(self, prefix=None, progress_callback=None):
-        if prefix:
-            root = posixpath.join(self.fs.path_info.path, prefix[:2])
-        else:
-            root = self.fs.path_info.path
-        with self.fs.ssh(self.fs.path_info) as ssh:
-            if prefix and not ssh.exists(root):
-                return
-            # If we simply return an iterator then with above closes instantly
-            if progress_callback:
-                for path in ssh.walk_files(root):
-                    progress_callback()
-                    yield path
-            else:
-                yield from ssh.walk_files(root)

dvc/objects/errors.py

@@ -7,3 +7,11 @@ class ObjectError(DvcException):
 
 class ObjectFormatError(ObjectError):
     pass
+
+
+class ObjectDBError(DvcException):
+    pass
+
+
+class ObjectDBPermissionError(ObjectDBError):
+    pass

dvc/objects/file.py

@@ -1,23 +1,36 @@
 import errno
 import logging
 import os
+from typing import TYPE_CHECKING, Optional
 
 from .errors import ObjectFormatError
 
+if TYPE_CHECKING:
+    from dvc.fs.base import BaseFileSystem
+    from dvc.hash_info import HashInfo
+    from dvc.types import AnyPath
+
+    from .db.base import ObjectDB
+
 logger = logging.getLogger(__name__)
 
 
 class HashFile:
-    def __init__(self, path_info, fs, hash_info):
+    def __init__(
+        self,
+        path_info: Optional["AnyPath"],
+        fs: Optional["BaseFileSystem"],
+        hash_info: "HashInfo",
+        name: Optional[str] = None,
+    ):
         self.path_info = path_info
         self.fs = fs
         self.hash_info = hash_info
+        self.name = name
 
     @property
     def size(self):
-        if not (self.path_info and self.fs):
-            return None
-        return self.fs.getsize(self.path_info)
+        return self.hash_info.size
 
     def __len__(self):
         return 1
@@ -37,10 +50,18 @@ class HashFile:
             and self.hash_info == other.hash_info
         )
 
-    def check(self, odb, check_hash=True):
-        from .stage import get_file_hash
+    def __hash__(self):
+        return hash(
+            (
+                self.hash_info,
+                self.path_info,
+                self.fs.scheme if self.fs else None,
+            )
+        )
 
+    def check(self, odb: "ObjectDB", check_hash: bool = True):
         if not check_hash:
+            assert self.fs
             if not self.fs.exists(self.path_info):
                 raise FileNotFoundError(
                     errno.ENOENT, os.strerror(errno.ENOENT), self.path_info
@@ -48,6 +69,11 @@ class HashFile:
             else:
                 return None
 
+        self._check_hash(odb)
+
+    def _check_hash(self, odb):
+        from .stage import get_file_hash
+
         actual = get_file_hash(
             self.path_info, self.fs, self.hash_info.name, odb.state
         )

dvc/objects/reference.py

@@ -0,0 +1,135 @@
+import errno
+import logging
+import os
+import pickle
+from typing import TYPE_CHECKING, Optional, Tuple
+
+from .errors import ObjectFormatError
+from .file import HashFile
+
+if TYPE_CHECKING:
+    from dvc.fs.base import BaseFileSystem
+    from dvc.hash_info import HashInfo
+    from dvc.types import AnyPath
+
+    from .db.base import ObjectDB
+
+logger = logging.getLogger(__name__)
+
+
+class ReferenceHashFile(HashFile):
+    PARAM_PATH = "path"
+    PARAM_HASH = "hash"
+    PARAM_MTIME = "mtime"
+    PARAM_SIZE = "size"
+    PARAM_FS_CONFIG = "fs_config"
+
+    def __init__(
+        self,
+        path_info: "AnyPath",
+        fs: "BaseFileSystem",
+        hash_info: "HashInfo",
+        mtime: Optional[str] = None,
+        size: Optional[int] = None,
+        **kwargs,
+    ):
+        super().__init__(path_info, fs, hash_info, **kwargs)
+        cur_mtime, cur_size = self._get_mtime_and_size()
+        self.mtime = cur_mtime if mtime is None else mtime
+        self.hash_info.size = cur_size if size is None else size
+
+    def __str__(self):
+        return f"ref object {self.hash_info} -> {self.path_info}"
+
+    def check(self, odb: "ObjectDB", check_hash: bool = True):
+        assert self.fs
+        if not self.fs.exists(self.path_info):
+            raise FileNotFoundError(
+                errno.ENOENT, os.strerror(errno.ENOENT), self.path_info
+            )
+        if not (self.mtime, self.size) == self._get_mtime_and_size():
+            raise ObjectFormatError(f"{self} is changed")
+        if check_hash:
+            self._check_hash(odb)
+
+    def _get_mtime_and_size(self) -> Tuple[Optional[str], int]:
+        from dvc.utils.fs import get_mtime_and_size
+
+        assert self.fs
+        if hasattr(self.fs, "stat"):
+            try:
+                return get_mtime_and_size(self.path_info, self.fs)
+            except FileNotFoundError:
+                pass
+        return None, self.fs.getsize(self.path_info)
+
+    def to_bytes(self):
+        from dvc.fs.repo import RepoFileSystem
+        from dvc.path_info import PathInfo
+
+        # NOTE: dumping reference FS's this way is insecure, as the
+        # fully parsed remote FS config will include credentials
+        #
+        # ReferenceHashFiles should currently only be serialized in
+        # memory and not to disk
+        path_info = self.path_info
+        if isinstance(self.fs, RepoFileSystem):
+            path_info = PathInfo(path_info).relative_to(self.fs.root_dir)
+
+        dict_ = {
+            self.PARAM_PATH: path_info,
+            self.PARAM_HASH: self.hash_info,
+            self.PARAM_MTIME: self.mtime,
+            self.PARAM_SIZE: self.size,
+            self.PARAM_FS_CONFIG: self.config_tuple(self.fs),
+        }
+        try:
+            return pickle.dumps(dict_)
+        except pickle.PickleError as exc:
+            raise ObjectFormatError(f"Could not pickle {self}") from exc
+
+    @classmethod
+    def from_bytes(cls, data: bytes, fs_cache: Optional[dict] = None):
+        from dvc.fs import get_fs_cls
+        from dvc.fs.repo import RepoFileSystem
+
+        try:
+            dict_ = pickle.loads(data)
+        except pickle.PickleError as exc:
+            raise ObjectFormatError("ReferenceHashFile is corrupted") from exc
+
+        try:
+            path_info = dict_[cls.PARAM_PATH]
+            hash_info = dict_[cls.PARAM_HASH]
+        except KeyError as exc:
+            raise ObjectFormatError("ReferenceHashFile is corrupted") from exc
+
+        scheme, config_pairs = dict_.get(cls.PARAM_FS_CONFIG)
+        fs = fs_cache.get((scheme, config_pairs)) if fs_cache else None
+        if not fs:
+            config = dict(config_pairs)
+            if RepoFileSystem.PARAM_REPO_URL in config:
+                fs = RepoFileSystem(**config)
+                path_info = fs.root_dir / path_info
+            else:
+                fs_cls = get_fs_cls(config, scheme=path_info.scheme)
+                fs = fs_cls(**config)
+        return ReferenceHashFile(
+            path_info,
+            fs,
+            hash_info,
+            mtime=dict_.get(cls.PARAM_MTIME),
+            size=dict_.get(cls.PARAM_SIZE),
+        )
+
+    @staticmethod
+    def config_tuple(fs: "BaseFileSystem"):
+        return (
+            fs.scheme,
+            tuple(
+                (key, value)
+                for key, value in sorted(
+                    fs.config.items(), key=lambda item: item[0]
+                )
+            ),
+        )

dvc/objects/stage.py

@@ -1,29 +1,45 @@
-import errno
+import hashlib
+import logging
 import os
+import pathlib
 from concurrent.futures import ThreadPoolExecutor
 from functools import partial
+from typing import TYPE_CHECKING, Dict, Tuple
 
 from dvc.exceptions import DvcIgnoreInCollectedDirError
 from dvc.hash_info import HashInfo
 from dvc.ignore import DvcIgnore
-from dvc.objects.file import HashFile
 from dvc.progress import Tqdm
 from dvc.utils import file_md5
 
+from .db.reference import ReferenceObjectDB
+from .file import HashFile
 
-def _upload_file(path_info, fs, odb):
+if TYPE_CHECKING:
+    from dvc.fs.base import BaseFileSystem
+    from dvc.types import AnyPath, DvcPath
+
+    from .db.base import ObjectDB
+
+logger = logging.getLogger(__name__)
+
+
+_STAGING_MEMFS_PATH = "dvc-staging"
+
+
+def _upload_file(path_info, fs, odb, upload_odb):
     from dvc.utils import tmp_fname
     from dvc.utils.stream import HashedStreamReader
 
-    tmp_info = odb.fs.path_info / tmp_fname()
+    tmp_info = upload_odb.path_info / tmp_fname()
     with fs.open(path_info, mode="rb", chunk_size=fs.CHUNK_SIZE) as stream:
         stream = HashedStreamReader(stream)
-        odb.fs.upload_fobj(
-            stream, tmp_info, desc=path_info.name, total=fs.getsize(path_info)
+        upload_odb.fs.upload_fobj(
+            stream, tmp_info, desc=path_info.name, size=fs.getsize(path_info)
         )
 
-    obj = HashFile(tmp_info, odb.fs, stream.hash_info)
-    return path_info, obj
+    odb.add(tmp_info, upload_odb.fs, stream.hash_info)
+    return path_info, odb.get(stream.hash_info)
 
 
 def _get_file_hash(path_info, fs, name):
@@ -52,11 +68,6 @@ def get_file_hash(path_info, fs, name, state=None):
         if hash_info:
             return hash_info
 
-    if not fs.exists(path_info):
-        raise FileNotFoundError(
-            errno.ENOENT, os.strerror(errno.ENOENT), path_info
-        )
-
     hash_info = _get_file_hash(path_info, fs, name)
 
     if state:
@@ -66,19 +77,31 @@ def get_file_hash(path_info, fs, name, state=None):
     return hash_info
 
 
-def _get_file_obj(path_info, fs, name, odb=None, state=None, upload=False):
-    if upload:
+def _get_file_obj(
+    path_info, fs, name, odb=None, upload_odb=None, dry_run=False
+):
+    state = odb.state if odb else None
+    hash_info = get_file_hash(path_info, fs, name, state=state)
+    if upload_odb and not dry_run:
         assert odb and name == "md5"
-        return _upload_file(path_info, fs, odb)
+        return _upload_file(path_info, fs, odb, upload_odb)
 
-    obj = HashFile(
-        path_info, fs, get_file_hash(path_info, fs, name, state=state)
-    )
+    if dry_run:
+        obj = HashFile(path_info, fs, hash_info)
+    else:
+        odb.add(path_info, fs, hash_info, move=False)
+        obj = odb.get(hash_info)
     return path_info, obj
 
 
 def _build_objects(
-    path_info, fs, name, odb, state, upload, dvcignore=None, **kwargs
+    path_info,
+    fs,
+    name,
+    dvcignore=None,
+    jobs=None,
+    no_progress_bar=False,
+    **kwargs,
 ):
     if dvcignore:
         walk_iterator = dvcignore.walk_files(fs, path_info)
@@ -87,45 +110,31 @@ def _build_objects(
     with Tqdm(
         unit="md5",
         desc="Computing file/dir hashes (only done once)",
-        disable=kwargs.pop("no_progress_bar", False),
+        disable=no_progress_bar,
     ) as pbar:
         worker = pbar.wrap_fn(
             partial(
                 _get_file_obj,
                 fs=fs,
                 name=name,
-                odb=odb,
-                state=state,
-                upload=upload,
+                **kwargs,
             )
         )
         with ThreadPoolExecutor(
-            max_workers=kwargs.pop("jobs", fs.hash_jobs)
+            max_workers=jobs if jobs is not None else fs.hash_jobs
         ) as executor:
             yield from executor.map(worker, walk_iterator)
 
 
-def _iter_objects(path_info, fs, name, odb, state, upload, **kwargs):
-    if not upload and name in fs.DETAIL_FIELDS:
-        for details in fs.find(path_info, detail=True):
-            file_info = path_info.replace(path=details["name"])
-            hash_info = HashInfo(name, details[name], size=details.get("size"))
-            yield file_info, HashFile(file_info, fs, hash_info)
-
-        return None
-
-    yield from _build_objects(
-        path_info, fs, name, odb, state, upload, **kwargs
-    )
+def _iter_objects(path_info, fs, name, **kwargs):
+    yield from _build_objects(path_info, fs, name, **kwargs)
 
 
-def _build_tree(path_info, fs, name, odb, state, upload, **kwargs):
+def _build_tree(path_info, fs, name, **kwargs):
     from .tree import Tree
 
     tree = Tree(None, None, None)
-    for file_info, obj in _iter_objects(
-        path_info, fs, name, odb, state, upload, **kwargs
-    ):
+    for file_info, obj in _iter_objects(path_info, fs, name, **kwargs):
         if DvcIgnore.DVCIGNORE_FILE == file_info.name:
             raise DvcIgnoreInCollectedDirError(file_info.parent)
 
@@ -138,89 +147,184 @@ def _build_tree(path_info, fs, name, odb, state, upload, **kwargs):
         # Yes, this is a BUG, as long as we permit "/" in
         # filenames on Windows and "\" on Unix
         tree.add(file_info.relative_to(path_info).parts, obj)
-
-    tree.digest()
-
     return tree
 
 
-def _get_tree_obj(path_info, fs, name, odb, state, upload, **kwargs):
+def _get_tree_obj(path_info, fs, fs_info, name, odb=None, **kwargs):
     from .tree import Tree
 
-    value = fs.info(path_info).get(name)
-    if value:
+    value = fs_info.get(name)
+    if odb and value:
         hash_info = HashInfo(name, value)
         try:
-            return Tree.load(odb, hash_info)
+            tree = Tree.load(odb, hash_info)
+            # NOTE: loaded entries are naive objects with hash_infos but no
+            # path_info. For staging trees, obj.path_info should be relative
+            # to the staging src `path_info` and src fs
+            for key, entry in tree:
+                entry.fs = fs
+                entry.path_info = path_info.joinpath(*key)
+            return tree
         except FileNotFoundError:
             pass
 
-    tree = _build_tree(path_info, fs, name, odb, state, upload, **kwargs)
-
-    odb.add(tree.path_info, tree.fs, tree.hash_info)
-    if name != "md5":
-        # NOTE: used only for external outputs. Initial reasoning was to be
-        # able to validate .dir files right in the workspace (e.g. check s3
-        # etag), but could be dropped for manual validation with regular md5,
-        # that would be universal for all clouds.
-        raw = odb.get(tree.hash_info)
-        hash_info = get_file_hash(raw.path_info, raw.fs, name, state)
-        tree.hash_info.name = hash_info.name
-        tree.hash_info.value = hash_info.value
-        if not tree.hash_info.value.endswith(".dir"):
-            tree.hash_info.value += ".dir"
-        odb.add(tree.path_info, tree.fs, tree.hash_info)
-
+    tree = _build_tree(path_info, fs, name, odb=odb, **kwargs)
+    state = odb.state if odb and odb.state else None
+    hash_info = None
+    if state:
+        hash_info = state.get(  # pylint: disable=assignment-from-none
+            path_info, fs
+        )
+    tree.digest(hash_info=hash_info)
+    odb.add(tree.path_info, tree.fs, tree.hash_info, move=True)
+    raw = odb.get(tree.hash_info)
+    # cleanup unneeded memfs tmpfile and return tree based on the
+    # ODB fs/path
+    if odb.fs != tree.fs:
+        tree.fs.remove(tree.path_info)
+    tree.fs = raw.fs
+    tree.path_info = raw.path_info
     return tree
 
 
-def stage(odb, path_info, fs, name, upload=False, **kwargs):
-    assert path_info and (
-        isinstance(path_info, str) or path_info.scheme == fs.scheme
-    )
+_url_cache: Dict[str, str] = {}
 
-    if not fs.exists(path_info):
-        raise FileNotFoundError(
-            errno.ENOENT, os.strerror(errno.ENOENT), path_info
-        )
+
+def _make_staging_url(path_info: "AnyPath"):
+    from dvc.path_info import CloudURLInfo
+    from dvc.scheme import Schemes
+
+    url = CloudURLInfo(f"{Schemes.MEMORY}://{_STAGING_MEMFS_PATH}")
+    if path_info:
+        if isinstance(path_info, (str, pathlib.PurePath)):
+            path = os.path.abspath(path_info)
+        else:
+            path = str(path_info)
+        if path not in _url_cache:
+            _url_cache[path] = hashlib.sha256(path.encode("utf-8")).hexdigest()
+        url /= _url_cache[path]
+    return url
+
+
+def _get_staging(odb: "ObjectDB") -> "ObjectDB":
+    """Return an ODB that can be used for staging objects.
+
+    Staging will be a reference ODB stored in the the global memfs.
+    """
+
+    from dvc.fs.memory import MemoryFileSystem
+
+    fs = MemoryFileSystem()
+    path_info = _make_staging_url(odb.path_info)
+    state = odb.state
+    return ReferenceObjectDB(fs, path_info, state=state)
+
+
+def _load_from_state(odb, staging, path_info, fs, name):
+    from . import check, load
+    from .errors import ObjectFormatError
+    from .tree import Tree
 
     state = odb.state
-    # pylint: disable=assignment-from-none
     hash_info = state.get(path_info, fs)
-
-    # If we have dir hash in state db, but dir cache file is lost,
-    # then we need to recollect the dir via .get_dir_hash() call below,
-    # see https://github.com/iterative/dvc/issues/2219 for context
-    if (
-        hash_info
-        and hash_info.isdir
-        and not odb.fs.exists(odb.hash_to_path_info(hash_info.value))
-    ):
-        hash_info = None
-
     if hash_info:
-        from . import load
-        from .tree import Tree
+        for odb_ in (odb, staging):
+            if odb_.exists(hash_info):
+                try:
+                    obj = load(odb_, hash_info)
+                    check(odb_, obj, check_hash=False)
+                    if isinstance(obj, Tree):
+                        obj.hash_info.nfiles = len(obj)
+                        for key, entry in obj:
+                            entry.fs = fs
+                            entry.path_info = path_info.joinpath(*key)
+                    else:
+                        obj.fs = fs
+                        obj.path_info = path_info
+                    assert obj.hash_info.name == name
+                    obj.hash_info.size = hash_info.size
+                    return odb_, obj
+                except (ObjectFormatError, FileNotFoundError):
+                    pass
+    raise FileNotFoundError
 
-        obj = load(odb, hash_info)
-        if isinstance(obj, Tree):
-            obj.hash_info.nfiles = len(obj)
-            for key, entry in obj:
-                entry.fs = fs
-                entry.path_info = path_info.joinpath(*key)
-        else:
-            obj.fs = fs
-            obj.path_info = path_info
-        assert obj.hash_info.name == name
-        obj.hash_info.size = hash_info.size
-        return obj
 
-    if fs.isdir(path_info):
-        obj = _get_tree_obj(path_info, fs, name, odb, state, upload, **kwargs)
+def _stage_external_tree_info(odb, tree, name):
+    # NOTE: used only for external outputs. Initial reasoning was to be
+    # able to validate .dir files right in the workspace (e.g. check s3
+    # etag), but could be dropped for manual validation with regular md5,
+    # that would be universal for all clouds.
+    assert odb and name != "md5"
+
+    odb.add(tree.path_info, tree.fs, tree.hash_info)
+    raw = odb.get(tree.hash_info)
+    hash_info = get_file_hash(raw.path_info, raw.fs, name, state=odb.state)
+    tree.path_info = raw.path_info
+    tree.fs = raw.fs
+    tree.hash_info.name = hash_info.name
+    tree.hash_info.value = hash_info.value
+    if not tree.hash_info.value.endswith(".dir"):
+        tree.hash_info.value += ".dir"
+    return tree
+
+
+def stage(
+    odb: "ObjectDB",
+    path_info: "DvcPath",
+    fs: "BaseFileSystem",
+    name: str,
+    upload: bool = False,
+    dry_run: bool = False,
+    **kwargs,
+) -> Tuple["ObjectDB", "HashFile"]:
+    """Stage (prepare) objects from the given path for addition to an ODB.
+
+    Returns at tuple of (staging_odb, object) where addition to the ODB can
+    be completed by transferring the object from staging to the dest ODB.
+
+    If dry_run is True, object hashes will be computed and returned, but file
+    objects themselves will not be added to the staging ODB (i.e. the resulting
+    file objects cannot transferred from staging to another ODB).
+
+    If upload is True, files will be uploaded to a temporary path on the dest
+    ODB filesystem, and staged objects will reference the uploaded path rather
+    than the original source path.
+    """
+    assert path_info and path_info.scheme == fs.scheme
+
+    details = fs.info(path_info)
+    staging = _get_staging(odb)
+    if odb:
+        try:
+            return _load_from_state(odb, staging, path_info, fs, name)
+        except FileNotFoundError:
+            pass
+
+    if details["type"] == "directory":
+        obj = _get_tree_obj(
+            path_info,
+            fs,
+            details,
+            name,
+            odb=staging,
+            upload_odb=odb if upload else None,
+            dry_run=dry_run,
+            **kwargs,
+        )
+        logger.debug("staged tree '%s'", obj)
+        if name != "md5":
+            obj = _stage_external_tree_info(odb, obj, name)
     else:
-        _, obj = _get_file_obj(path_info, fs, name, odb, state, upload)
+        _, obj = _get_file_obj(
+            path_info,
+            fs,
+            name,
+            odb=staging,
+            upload_odb=odb if upload else None,
+            dry_run=dry_run,
+        )
 
-    if obj.hash_info and fs.exists(path_info):
-        state.save(path_info, fs, obj.hash_info)
+    if odb and odb.state and obj.hash_info:
+        odb.state.save(path_info, fs, obj.hash_info)
 
-    return obj
+    return staging, obj

dvc/objects/status.py

@@ -0,0 +1,187 @@
+import logging
+from typing import TYPE_CHECKING, Dict, Iterable, NamedTuple, Optional, Set
+
+from dvc.hash_info import HashInfo
+from dvc.scheme import Schemes
+
+from .tree import Tree
+
+if TYPE_CHECKING:
+    from .db.base import ObjectDB
+    from .db.index import ObjectDBIndexBase
+    from .file import HashFile
+
+logger = logging.getLogger(__name__)
+
+
+class StatusResult(NamedTuple):
+    exists: Set["HashInfo"]
+    missing: Set["HashInfo"]
+
+
+class CompareStatusResult(NamedTuple):
+    ok: Set["HashInfo"]
+    missing: Set["HashInfo"]
+    new: Set["HashInfo"]
+    deleted: Set["HashInfo"]
+
+
+def _indexed_dir_hashes(odb, index, dir_objs, name, cache_odb):
+    # Validate our index by verifying all indexed .dir hashes
+    # still exist on the remote
+    dir_hashes = set(dir_objs.keys())
+    indexed_dirs = set(index.dir_hashes())
+    indexed_dir_exists = set()
+    if indexed_dirs:
+        indexed_dir_exists.update(odb.list_hashes_exists(indexed_dirs))
+        missing_dirs = indexed_dirs.difference(indexed_dir_exists)
+        if missing_dirs:
+            logger.debug(
+                "Remote cache missing indexed .dir hashes '%s', "
+                "clearing remote index",
+                ", ".join(missing_dirs),
+            )
+            index.clear()
+
+    # Check if non-indexed (new) dir hashes exist on remote
+    dir_exists = dir_hashes.intersection(indexed_dir_exists)
+    dir_exists.update(odb.list_hashes_exists(dir_hashes - dir_exists))
+
+    # If .dir hash exists in the ODB, assume directory contents
+    # also exists
+    for dir_hash in dir_exists:
+        tree = dir_objs.get(dir_hash)
+        if not tree:
+            try:
+                tree = Tree.load(cache_odb, HashInfo(name, dir_hash))
+            except FileNotFoundError:
+                continue
+        file_hashes = [entry.hash_info.value for _, entry in tree]
+        if dir_hash not in index:
+            logger.debug(
+                "Indexing new .dir '%s' with '%s' nested files",
+                dir_hash,
+                len(file_hashes),
+            )
+            index.update([dir_hash], file_hashes)
+        yield from file_hashes
+        yield tree.hash_info.value
+
+
+def status(
+    odb: "ObjectDB",
+    obj_ids: Iterable["HashInfo"],
+    name: Optional[str] = None,
+    index: Optional["ObjectDBIndexBase"] = None,
+    cache_odb: Optional["ObjectDB"] = None,
+    shallow: bool = True,
+    **kwargs,
+) -> "StatusResult":
+    """Return status of whether or not the specified objects exist odb.
+
+    If cache_odb is set, trees will be loaded from cache_odb instead of odb
+    when needed.
+
+    Status is returned as a tuple of:
+        exists: objs that exist in odb
+        missing: objs that do not exist in ODB
+    """
+    logger.debug("Preparing to collect status from '%s'", odb.path_info)
+    if not name:
+        name = odb.fs.PARAM_CHECKSUM
+
+    if cache_odb is None:
+        cache_odb = odb
+
+    hash_infos: Dict[str, "HashInfo"] = {}
+    dir_objs: Dict[str, Optional["HashFile"]] = {}
+    for hash_info in obj_ids:
+        assert hash_info.value
+        if hash_info.isdir:
+            if shallow:
+                tree = None
+            else:
+                tree = Tree.load(cache_odb, hash_info)
+                for _, entry in tree:
+                    assert entry.hash_info and entry.hash_info.value
+                    hash_infos[entry.hash_info.value] = entry.hash_info
+            if index:
+                dir_objs[hash_info.value] = tree
+        hash_infos[hash_info.value] = hash_info
+
+    if odb.fs.scheme == Schemes.MEMORY:
+        # assume memfs staged objects already exist
+        return StatusResult(set(hash_infos.values()), set())
+
+    hashes: Set[str] = set(hash_infos.keys())
+    exists: Set[str] = set()
+
+    logger.debug("Collecting status from '%s'", odb.path_info)
+    if index and hashes:
+        if dir_objs:
+            exists = hashes.intersection(
+                _indexed_dir_hashes(odb, index, dir_objs, name, cache_odb)
+            )
+            hashes.difference_update(exists)
+        if hashes:
+            exists.update(index.intersection(hashes))
+            hashes.difference_update(exists)
+
+    if hashes:
+        exists.update(
+            odb.hashes_exist(hashes, name=str(odb.path_info), **kwargs)
+        )
+    return StatusResult(
+        {hash_infos[hash_] for hash_ in exists},
+        {hash_infos[hash_] for hash_ in (hashes - exists)},
+    )
+
+
+def compare_status(
+    src: "ObjectDB",
+    dest: "ObjectDB",
+    obj_ids: Iterable["HashInfo"],
+    log_missing: bool = True,
+    check_deleted: bool = True,
+    src_index: Optional["ObjectDBIndexBase"] = None,
+    dest_index: Optional["ObjectDBIndexBase"] = None,
+    **kwargs,
+) -> "CompareStatusResult":
+    """Compare status for the specified objects between two ODBs.
+
+    Status is returned as a tuple of:
+        ok: hashes that exist in both src and dest
+        missing: hashes that do not exist in neither src nor dest
+        new: hashes that only exist in src
+        deleted: hashes that only exist in dest
+    """
+    if "cache_odb" not in kwargs:
+        kwargs["cache_odb"] = src
+    dest_exists, dest_missing = status(
+        dest, obj_ids, index=dest_index, **kwargs
+    )
+    # for transfer operations we can skip src status check when all objects
+    # already exist in dest
+    if dest_missing or check_deleted:
+        src_exists, src_missing = status(
+            src, obj_ids, index=src_index, **kwargs
+        )
+    else:
+        src_exists = dest_exists
+        src_missing = set()
+    result = CompareStatusResult(
+        src_exists & dest_exists,
+        src_missing & dest_missing,
+        src_exists - dest_exists,
+        dest_exists - src_exists,
+    )
+    if log_missing and result.missing:
+        missing_desc = "\n".join(
+            f"name: {hash_info.obj_name}, {hash_info}"
+            for hash_info in result.missing
+        )
+        logger.warning(
+            "Some of the cache files do not exist neither locally "
+            f"nor on remote. Missing cache files:\n{missing_desc}"
+        )
+    return result

dvc/objects/transfer.py

@@ -0,0 +1,239 @@
+import errno
+import itertools
+import logging
+from concurrent import futures
+from concurrent.futures import ThreadPoolExecutor
+from functools import partial, wraps
+from typing import TYPE_CHECKING, Callable, Iterable, Optional
+
+from funcy import split
+
+from dvc.progress import Tqdm
+
+if TYPE_CHECKING:
+    from dvc.hash_info import HashInfo
+
+    from .db.base import ObjectDB
+    from .db.index import ObjectDBIndexBase
+
+logger = logging.getLogger(__name__)
+
+
+def _log_exceptions(func):
+    @wraps(func)
+    def wrapper(path_info, *args, **kwargs):
+        try:
+            func(path_info, *args, **kwargs)
+            return 0
+        except Exception as exc:  # pylint: disable=broad-except
+            # NOTE: this means we ran out of file descriptors and there is no
+            # reason to try to proceed, as we will hit this error anyways.
+            # pylint: disable=no-member
+            if isinstance(exc, OSError) and exc.errno == errno.EMFILE:
+                raise
+
+            logger.exception("failed to transfer '%s'", path_info)
+            return 1
+
+    return wrapper
+
+
+def _transfer(
+    src, dest, dir_ids, file_ids, missing_ids, jobs, verify, move, **kwargs
+):
+    func = _log_exceptions(dest.add)
+    total = len(dir_ids) + len(file_ids)
+    if total == 0:
+        return 0
+    with Tqdm(total=total, unit="file", desc="Transferring") as pbar:
+        func = pbar.wrap_fn(func)
+        with ThreadPoolExecutor(max_workers=jobs) as executor:
+            processor = partial(
+                _create_tasks,
+                executor,
+                jobs,
+                func,
+                src,
+                verify,
+                move,
+            )
+            processor.add_func = func
+            _do_transfer(
+                src,
+                dest,
+                dir_ids,
+                file_ids,
+                missing_ids,
+                processor,
+                verify=verify,
+                move=move,
+                **kwargs,
+            )
+    return total
+
+
+def _create_tasks(executor, jobs, func, src, verify, move, obj_ids):
+    fails = 0
+    hash_iter = iter(obj_ids)
+
+    def submit(hash_info):
+        obj = src.get(hash_info)
+        return executor.submit(
+            func,
+            obj.path_info,
+            obj.fs,
+            obj.hash_info,
+            verify=verify,
+            move=move,
+        )
+
+    def create_taskset(amount):
+        return {
+            submit(hash_info)
+            for hash_info in itertools.islice(hash_iter, amount)
+        }
+
+    tasks = create_taskset(jobs * 5)
+    while tasks:
+        done, tasks = futures.wait(tasks, return_when=futures.FIRST_COMPLETED)
+        fails += sum(task.result() for task in done)
+        tasks.update(create_taskset(len(done)))
+    return fails
+
+
+def _do_transfer(
+    src: "ObjectDB",
+    dest: "ObjectDB",
+    dir_ids: Iterable["HashInfo"],
+    file_ids: Iterable["HashInfo"],
+    missing_ids: Iterable["HashInfo"],
+    processor: Callable,
+    src_index: Optional["ObjectDBIndexBase"] = None,
+    dest_index: Optional["ObjectDBIndexBase"] = None,
+    cache_odb: Optional["ObjectDB"] = None,
+    **kwargs,
+):
+    from dvc.exceptions import FileTransferError
+    from dvc.objects.errors import ObjectFormatError
+
+    total_fails = 0
+    succeeded_dir_objs = []
+    all_file_ids = set(file_ids)
+
+    for dir_hash in dir_ids:
+        from .tree import Tree
+
+        bound_file_ids = set()
+        dir_obj: Optional["Tree"] = None
+        for odb in (cache_odb, src):
+            if odb is not None:
+                try:
+                    dir_obj = Tree.load(odb, dir_hash)
+                    break
+                except (FileNotFoundError, ObjectFormatError):
+                    pass
+        assert dir_obj
+        entry_ids = {entry.hash_info for _, entry in dir_obj}
+
+        for file_hash in all_file_ids.copy():
+            if file_hash in entry_ids:
+                bound_file_ids.add(file_hash)
+                all_file_ids.remove(file_hash)
+
+        dir_fails = processor(bound_file_ids)
+        if dir_fails:
+            logger.debug(
+                "failed to upload full contents of '%s', "
+                "aborting .dir file upload",
+                dir_obj.name,
+            )
+            logger.error(
+                "failed to upload '%s' to '%s'",
+                src.get(dir_obj.hash_info).path_info,
+                dest.get(dir_obj.hash_info).path_info,
+            )
+            total_fails += dir_fails + 1
+        elif entry_ids.intersection(missing_ids):
+            # if for some reason a file contained in this dir is
+            # missing both locally and in the remote, we want to
+            # push whatever file content we have, but should not
+            # push .dir file
+            logger.debug(
+                "directory '%s' contains missing files,"
+                "skipping .dir file upload",
+                dir_obj.name,
+            )
+        else:
+            raw_obj = src.get(dir_obj.hash_info)
+            is_dir_failed = processor.add_func(  # type: ignore[attr-defined]
+                raw_obj.path_info,
+                raw_obj.fs,
+                raw_obj.hash_info,
+                **kwargs,
+            )
+            total_fails += is_dir_failed
+            if not is_dir_failed:
+                succeeded_dir_objs.append(dir_obj)
+
+    # insert the rest
+    total_fails += processor(all_file_ids)
+    if total_fails:
+        if src_index:
+            src_index.clear()
+        raise FileTransferError(total_fails)
+
+    # index successfully pushed dirs
+    if dest_index:
+        for dir_obj in succeeded_dir_objs:
+            file_hashes = {entry.hash_info.value for _, entry in dir_obj}
+            logger.debug(
+                "Indexing pushed dir '%s' with '%s' nested files",
+                dir_obj.hash_info,
+                len(file_hashes),
+            )
+            assert dir_obj.hash_info and dir_obj.hash_info.value
+            dest_index.update([dir_obj.hash_info.value], file_hashes)
+
+
+def transfer(
+    src: "ObjectDB",
+    dest: "ObjectDB",
+    obj_ids: Iterable["HashInfo"],
+    jobs: Optional[int] = None,
+    verify: bool = False,
+    move: bool = False,
+    **kwargs,
+) -> int:
+    """Transfer (copy) the specified objects from one ODB to another.
+
+    Returns the number of successfully transferred objects
+    """
+    from .status import compare_status
+
+    logger.debug(
+        "Preparing to transfer data from '%s' to '%s'",
+        src.path_info,
+        dest.path_info,
+    )
+    if src == dest:
+        return 0
+
+    status = compare_status(src, dest, obj_ids, check_deleted=False, **kwargs)
+    if not status.new:
+        return 0
+
+    dir_ids, file_ids = split(lambda hash_info: hash_info.isdir, status.new)
+    if jobs is None:
+        jobs = dest.fs.jobs
+
+    return _transfer(
+        src,
+        dest,
+        set(dir_ids),
+        set(file_ids),
+        status.missing,
+        jobs,
+        verify,
+        move,
+        **kwargs,
+    )

dvc/objects/tree.py

@@ -1,6 +1,7 @@
 import json
 import logging
 import posixpath
+from typing import TYPE_CHECKING, Dict, Optional, Tuple
 
 from funcy import cached_property
 
@@ -8,6 +9,9 @@ from .errors import ObjectFormatError
 from .file import HashFile
 from .stage import get_file_hash
 
+if TYPE_CHECKING:
+    from dvc.hash_info import HashInfo
+
 logger = logging.getLogger(__name__)
 
 
@@ -16,7 +20,7 @@ class Tree(HashFile):
 
     def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)
-        self._dict = {}
+        self._dict: Dict[Tuple[str], "HashFile"] = {}
 
     @cached_property
     def trie(self):
@@ -24,31 +28,31 @@ class Tree(HashFile):
 
         return Trie(self._dict)
 
-    @property
-    def size(self):
-        try:
-            return sum(obj.size for _, obj in self)
-        except TypeError:
-            return None
-
     def add(self, key, obj):
         self.__dict__.pop("trie", None)
         self._dict[key] = obj
 
-    def digest(self):
+    def digest(self, hash_info: Optional["HashInfo"] = None):
         from dvc.fs.memory import MemoryFileSystem
-        from dvc.path_info import PathInfo
+        from dvc.path_info import CloudURLInfo
         from dvc.utils import tmp_fname
 
         memfs = MemoryFileSystem()
-        path_info = PathInfo(tmp_fname(""))
+        path_info = CloudURLInfo("memory://{}".format(tmp_fname("")))
         with memfs.open(path_info, "wb") as fobj:
             fobj.write(self.as_bytes())
         self.fs = memfs
         self.path_info = path_info
-        self.hash_info = get_file_hash(path_info, memfs, "md5")
-        self.hash_info.value += ".dir"
-        self.hash_info.size = self.size
+        if hash_info:
+            self.hash_info = hash_info
+        else:
+            self.hash_info = get_file_hash(path_info, memfs, "md5")
+            assert self.hash_info.value
+            self.hash_info.value += ".dir"
+        try:
+            self.hash_info.size = sum(obj.size for _, obj in self)
+        except TypeError:
+            self.hash_info.size = None
         self.hash_info.nfiles = len(self)
 
     def __len__(self):
@@ -96,7 +100,6 @@ class Tree(HashFile):
 
     @classmethod
     def load(cls, odb, hash_info):
-
         obj = odb.get(hash_info)
 
         try:
@@ -115,24 +118,46 @@ class Tree(HashFile):
         tree = cls.from_list(raw)
         tree.path_info = obj.path_info
         tree.fs = obj.fs
+        for _, entry_obj in tree:
+            entry_obj.fs = obj.fs
         tree.hash_info = hash_info
 
         return tree
 
-    def filter(self, odb, prefix):
+    def filter(self, prefix: Tuple[str]) -> Optional["Tree"]:
+        """Return a filtered copy of this tree that only contains entries
+        inside prefix.
+
+        The returned tree will contain the original tree's hash_info and
+        path_info.
+
+        Returns an empty tree if no object exists at the specified prefix.
+        """
+        tree = Tree(self.path_info, self.fs, self.hash_info)
+        try:
+            for key, obj in self.trie.items(prefix):
+                tree.add(key, obj)
+        except KeyError:
+            pass
+        return tree
+
+    def get(self, prefix: Tuple[str]) -> Optional[HashFile]:
+        """Return object at the specified prefix in this tree.
+
+        Returns None if no object exists at the specified prefix.
+        """
         obj = self._dict.get(prefix)
         if obj:
             return obj
 
-        depth = len(prefix)
         tree = Tree(None, None, None)
+        depth = len(prefix)
         try:
             for key, obj in self.trie.items(prefix):
                 tree.add(key[depth:], obj)
         except KeyError:
             return None
         tree.digest()
-        odb.add(tree.path_info, tree.fs, tree.hash_info)
         return tree
 
 

dvc/output.py

@@ -2,7 +2,7 @@ import logging
 import os
 from collections import defaultdict
 from copy import copy
-from typing import Type
+from typing import TYPE_CHECKING, Dict, Optional, Set, Type
 from urllib.parse import urlparse
 
 from funcy import collecting, project
@@ -24,14 +24,17 @@ from .fs.local import LocalFileSystem
 from .fs.s3 import S3FileSystem
 from .hash_info import HashInfo
 from .istextfile import istextfile
-from .objects import save as osave
-from .objects.db import NamedCache
+from .objects import Tree
 from .objects.errors import ObjectFormatError
 from .objects.stage import stage as ostage
+from .objects.transfer import transfer as otransfer
 from .scheme import Schemes
 from .utils import relpath
 from .utils.fs import path_isin
 
+if TYPE_CHECKING:
+    from .objects.db.base import ObjectDB
+
 logger = logging.getLogger(__name__)
 
 
@@ -41,18 +44,24 @@ CHECKSUM_SCHEMA = Any(
     And(Any(str, And(int, Coerce(str))), Length(min=3), Lower),
 )
 
+CASE_SENSITIVE_CHECKSUM_SCHEMA = Any(
+    None,
+    And(str, Length(max=0), SetTo(None)),
+    And(Any(str, And(int, Coerce(str))), Length(min=3)),
+)
+
 # NOTE: currently there are only 3 possible checksum names:
 #
 #    1) md5 (LOCAL, SSH);
-#    2) etag (S3);
+#    2) etag (S3, GS, OSS, AZURE, HTTP);
 #    3) checksum (HDFS);
 #
 # so when a few types of outputs share the same name, we only need
 # specify it once.
 CHECKSUMS_SCHEMA = {
     LocalFileSystem.PARAM_CHECKSUM: CHECKSUM_SCHEMA,
-    S3FileSystem.PARAM_CHECKSUM: CHECKSUM_SCHEMA,
     HDFSFileSystem.PARAM_CHECKSUM: CHECKSUM_SCHEMA,
+    S3FileSystem.PARAM_CHECKSUM: CASE_SENSITIVE_CHECKSUM_SCHEMA,
 }
 
 
@@ -261,8 +270,6 @@ class Output:
     IsStageFileError = OutputIsStageFileError  # type: Type[DvcException]
     IsIgnoredError = OutputIsIgnoredError  # type: Type[DvcException]
 
-    sep = "/"
-
     def __init__(
         self,
         stage,
@@ -279,9 +286,8 @@ class Output:
     ):
         self.repo = stage.repo if stage else None
 
-        fs_cls, fs_config = get_cloud_fs(self.repo, url=path)
+        fs_cls, fs_config, path_info = get_cloud_fs(self.repo, url=path)
         self.fs = fs_cls(**fs_config)
-        path_info = self.fs.path_info
 
         if (
             self.fs.scheme == "local"
@@ -320,6 +326,8 @@ class Output:
         self.obj = None
         self.isexec = False if self.IS_DEPENDENCY else isexec
 
+        self.def_remote = None
+
     def _parse_path(self, fs, path_info):
         if fs.scheme != "local":
             return path_info
@@ -395,21 +403,21 @@ class Output:
         return self.odb.hash_to_path_info(self.hash_info.value).url
 
     def get_hash(self):
-        if not self.use_cache:
-            return ostage(
-                self.repo.odb.local,
-                self.path_info,
-                self.fs,
-                self.fs.PARAM_CHECKSUM,
-                dvcignore=self.dvcignore,
-            ).hash_info
-        return ostage(
-            self.odb,
+        if self.use_cache:
+            odb = self.odb
+            name = self.odb.fs.PARAM_CHECKSUM
+        else:
+            odb = self.repo.odb.local
+            name = self.fs.PARAM_CHECKSUM
+        _, obj = ostage(
+            odb,
             self.path_info,
             self.fs,
-            self.odb.fs.PARAM_CHECKSUM,
+            name,
             dvcignore=self.dvcignore,
-        ).hash_info
+            dry_run=not self.use_cache,
+        )
+        return obj.hash_info
 
     @property
     def is_dir_checksum(self):
@@ -536,7 +544,7 @@ class Output:
             logger.debug("Output '%s' didn't change. Skipping saving.", self)
             return
 
-        self.obj = ostage(
+        _, self.obj = ostage(
             self.odb,
             self.path_info,
             self.fs,
@@ -557,14 +565,28 @@ class Output:
         assert self.hash_info
 
         if self.use_cache:
-            obj = ostage(
-                self.odb,
-                filter_info or self.path_info,
-                self.fs,
-                self.odb.fs.PARAM_CHECKSUM,
-                dvcignore=self.dvcignore,
+            granular = (
+                self.is_dir_checksum
+                and filter_info
+                and filter_info != self.path_info
             )
-            objects.save(self.odb, obj)
+            if granular:
+                obj = self._commit_granular_dir(filter_info)
+            else:
+                staging, obj = ostage(
+                    self.odb,
+                    filter_info or self.path_info,
+                    self.fs,
+                    self.odb.fs.PARAM_CHECKSUM,
+                    dvcignore=self.dvcignore,
+                )
+                otransfer(
+                    staging,
+                    self.odb,
+                    {obj.hash_info},
+                    shallow=False,
+                    move=True,
+                )
             checkout(
                 filter_info or self.path_info,
                 self.fs,
@@ -576,6 +598,26 @@ class Output:
             )
             self.set_exec()
 
+    def _commit_granular_dir(self, filter_info):
+        prefix = filter_info.relative_to(self.path_info).parts
+        staging, save_obj = ostage(
+            self.odb,
+            self.path_info,
+            self.fs,
+            self.odb.fs.PARAM_CHECKSUM,
+            dvcignore=self.dvcignore,
+        )
+        save_obj = save_obj.filter(prefix)
+        checkout_obj = save_obj.get(prefix)
+        otransfer(
+            staging,
+            self.odb,
+            {save_obj.hash_info} | {entry.hash_info for _, entry in save_obj},
+            shallow=True,
+            move=True,
+        )
+        return checkout_obj
+
     def dumpd(self):
         ret = copy(self.hash_info.to_dict())
 
@@ -648,7 +690,7 @@ class Output:
     def download(self, to, jobs=None):
         self.fs.download(self.path_info, to.path_info, jobs=jobs)
 
-    def get_obj(self, filter_info=None):
+    def get_obj(self, filter_info=None, **kwargs):
         if self.obj:
             obj = self.obj
         elif self.hash_info:
@@ -661,7 +703,7 @@ class Output:
 
         if filter_info and filter_info != self.path_info:
             prefix = filter_info.relative_to(self.path_info).parts
-            obj = obj.filter(self.odb, prefix)
+            obj = obj.get(prefix)
 
         return obj
 
@@ -741,9 +783,8 @@ class Output:
         if odb is None:
             odb = self.odb
 
-        cls, config = get_cloud_fs(self.repo, url=source)
+        cls, config, from_info = get_cloud_fs(self.repo, url=source)
         from_fs = cls(**config)
-        from_info = from_fs.path_info
 
         # When running import-url --to-remote / add --to-remote/-o ... we
         # assume that it is unlikely that the odb will contain majority of the
@@ -756,7 +797,7 @@ class Output:
 
         upload = not (update and from_fs.isdir(from_info))
         jobs = jobs or min((from_fs.jobs, odb.fs.jobs))
-        obj = ostage(
+        staging, obj = ostage(
             odb,
             from_info,
             from_fs,
@@ -765,7 +806,14 @@ class Output:
             jobs=jobs,
             no_progress_bar=no_progress_bar,
         )
-        osave(odb, obj, jobs=jobs, move=upload)
+        otransfer(
+            staging,
+            odb,
+            {obj.hash_info},
+            jobs=jobs,
+            move=upload,
+            shallow=False,
+        )
 
         self.hash_info = obj.hash_info
         return obj
@@ -788,18 +836,17 @@ class Output:
             self.odb.unprotect(self.path_info)
 
     def get_dir_cache(self, **kwargs):
-
         if not self.is_dir_checksum:
             raise DvcException("cannot get dir cache for file checksum")
 
+        obj = self.odb.get(self.hash_info)
         try:
-            objects.check(self.odb, self.odb.get(self.hash_info))
-        except (FileNotFoundError, ObjectFormatError):
-            self.repo.cloud.pull(
-                NamedCache.make("local", self.hash_info.value, str(self)),
-                show_checksums=False,
-                **kwargs,
-            )
+            objects.check(self.odb, obj)
+        except FileNotFoundError:
+            self.repo.cloud.pull([obj.hash_info], **kwargs)
+
+        if self.obj:
+            return self.obj
 
         try:
             self.obj = objects.load(self.odb, self.hash_info)
@@ -810,27 +857,8 @@ class Output:
 
     def collect_used_dir_cache(
         self, remote=None, force=False, jobs=None, filter_info=None
-    ):
-        """Get a list of `info`s related to the given directory.
-
-        - Pull the directory entry from the remote cache if it was changed.
-
-        Example:
-
-            Given the following commands:
-
-            $ echo "foo" > directory/foo
-            $ echo "bar" > directory/bar
-            $ dvc add directory
-
-            It will return a NamedCache like:
-
-            nc = NamedCache()
-            nc.add(self.scheme, 'c157a79031e1', 'directory/foo')
-            nc.add(self.scheme, 'd3b07384d113', 'directory/bar')
-        """
-
-        cache = NamedCache()
+    ) -> Dict[Optional["ObjectDB"], Set["HashInfo"]]:
+        """Fetch dir cache and return used object IDs for this out."""
 
         try:
             self.get_dir_cache(jobs=jobs, remote=remote)
@@ -839,7 +867,7 @@ class Output:
 
         try:
             objects.check(self.odb, self.odb.get(self.hash_info))
-        except (FileNotFoundError, ObjectFormatError):
+        except FileNotFoundError:
             msg = (
                 "Missing cache for directory '{}'. "
                 "Cache for files inside will be lost. "
@@ -850,38 +878,24 @@ class Output:
                     "unable to fully collect used cache"
                     " without cache for directory '{}'".format(self)
                 )
-            return cache
+            return {}
 
-        path = str(self.path_info)
-        filter_path = str(filter_info) if filter_info else None
-        for entry_key, entry_obj in self.obj:
-            entry_path = os.path.join(path, *entry_key)
-            if (
-                not filter_path
-                or entry_path == filter_path
-                or entry_path.startswith(filter_path + os.sep)
-            ):
-                cache.add(self.scheme, entry_obj.hash_info.value, entry_path)
+        obj = self.get_obj()
+        if filter_info and filter_info != self.path_info:
+            prefix = filter_info.relative_to(self.path_info).parts
+            obj = obj.filter(prefix)
+        return {None: set(self._named_obj_ids(obj))}
 
-        return cache
-
-    def get_used_cache(self, **kwargs):
-        """Get a dumpd of the given `out`, with an entry including the branch.
-
-        The `used_cache` of an output is no more than its `info`.
-
-        In case that the given output is a directory, it will also
-        include the `info` of its files.
-        """
+    def get_used_objs(
+        self, **kwargs
+    ) -> Dict[Optional["ObjectDB"], Set["HashInfo"]]:
+        """Return filtered set of used object IDs for this out."""
 
         if not self.use_cache:
-            return NamedCache()
+            return {}
 
         if self.stage.is_repo_import:
-            cache = NamedCache()
-            (dep,) = self.stage.deps
-            cache.external[dep.repo_pair].add(dep.def_path)
-            return cache
+            return self.get_used_external(**kwargs)
 
         if not self.hash_info:
             msg = (
@@ -900,18 +914,34 @@ class Output:
                     )
                 )
             logger.warning(msg)
-            return NamedCache()
+            return {}
 
-        ret = NamedCache.make(self.scheme, self.hash_info.value, str(self))
+        if self.is_dir_checksum:
+            return self.collect_used_dir_cache(**kwargs)
 
-        if not self.is_dir_checksum:
-            return ret
+        obj = self.get_obj(filter_info=kwargs.get("filter_info"))
+        if not obj:
+            obj = self.odb.get(self.hash_info)
 
-        ret.add_child_cache(
-            self.hash_info.value, self.collect_used_dir_cache(**kwargs)
-        )
+        return {None: set(self._named_obj_ids(obj))}
 
-        return ret
+    def _named_obj_ids(self, obj):
+        name = str(self)
+        obj.hash_info.obj_name = name
+        yield obj.hash_info
+        if isinstance(obj, Tree):
+            for key, entry_obj in obj:
+                entry_obj.hash_info.obj_name = self.fs.sep.join([name, *key])
+                yield entry_obj.hash_info
+
+    def get_used_external(
+        self, **kwargs
+    ) -> Dict[Optional["ObjectDB"], Set["HashInfo"]]:
+        if not self.use_cache or not self.stage.is_repo_import:
+            return {}
+
+        (dep,) = self.stage.deps
+        return dep.get_used_objs(**kwargs)
 
     def _validate_output_path(self, path, stage=None):
         from dvc.dvcfile import is_valid_filename
@@ -974,6 +1004,18 @@ class Output:
     def fspath(self):
         return self.path_info.fspath
 
+    @property
+    def is_decorated(self) -> bool:
+        return self.is_metric or self.is_plot
+
+    @property
+    def is_metric(self) -> bool:
+        return bool(self.metric) or bool(self.live)
+
+    @property
+    def is_plot(self) -> bool:
+        return bool(self.plot)
+
 
 ARTIFACT_SCHEMA = {
     **CHECKSUMS_SCHEMA,

dvc/path_info.py

@@ -52,7 +52,7 @@ class PathInfo(pathlib.PurePath, _BasePath):
         return relpath(path)
 
     def __repr__(self):
-        return "{}: '{}'".format(type(self).__name__, self)
+        return f"{type(self).__name__}: '{self}'"
 
     # This permits passing it to file utils directly in Python 3.6+
     def __fspath__(self):
@@ -64,6 +64,8 @@ class PathInfo(pathlib.PurePath, _BasePath):
 
     url = fspath
 
+    path = fspath
+
     def relpath(self, other):
         return self.__class__(relpath(self, other))
 
@@ -174,7 +176,7 @@ class URLInfo(_BasePath):
         return self.url
 
     def __repr__(self):
-        return "{}: '{}'".format(type(self).__name__, self)
+        return f"{type(self).__name__}: '{self}'"
 
     def __eq__(self, other):
         if isinstance(other, (str, bytes)):

dvc/progress.py

@@ -4,8 +4,10 @@ import logging
 import sys
 from threading import RLock
 
+import fsspec
 from tqdm import tqdm
 
+from dvc.env import DVC_IGNORE_ISATTY
 from dvc.utils import env2bool
 
 logger = logging.getLogger(__name__)
@@ -50,7 +52,7 @@ class Tqdm(tqdm):
         file=None,
         total=None,
         postfix=None,
-        **kwargs
+        **kwargs,
     ):
         """
         bytes   : shortcut for
@@ -77,7 +79,7 @@ class Tqdm(tqdm):
         # auto-disable based on TTY
         if (
             not disable
-            and not env2bool("DVC_IGNORE_ISATTY")
+            and not env2bool(DVC_IGNORE_ISATTY)
             and hasattr(file, "isatty")
         ):
             disable = not file.isatty()
@@ -89,7 +91,7 @@ class Tqdm(tqdm):
             bar_format="!",
             lock_args=(False,),
             total=total,
-            **kwargs
+            **kwargs,
         )
         self.postfix = postfix or {"info": ""}
         if bar_format is None:
@@ -105,13 +107,16 @@ class Tqdm(tqdm):
             self.bar_format = bar_format
         self.refresh()
 
-    def update_msg(self, msg, n=1):
+    def update_msg(self, msg: str, n: int = 1) -> None:
         """
         Sets `msg` as a postfix and calls `update(n)`.
         """
-        self.postfix["info"] = " %s |" % msg
+        self.set_msg(msg)
         self.update(n)
 
+    def set_msg(self, msg: str) -> None:
+        self.postfix["info"] = f" {msg} |"
+
     def update_to(self, current, total=None):
         if total:
             self.total = total
@@ -132,6 +137,9 @@ class Tqdm(tqdm):
 
         return wrapped
 
+    def as_callback(self, fs, path_info):
+        return FsspecCallback(fs, path_info, self)
+
     def close(self):
         self.postfix["info"] = ""
         # remove ETA (either unknown or zero); remove completed bar
@@ -157,3 +165,26 @@ class Tqdm(tqdm):
             d["ncols_desc"] = d["ncols_info"] = 1
             d["prefix"] = ""
         return d
+
+
+class FsspecCallback(fsspec.Callback):
+    def __init__(self, fs, path_info, progress_bar):
+        self.fs = fs
+        self.path_info = path_info
+        self.progress_bar = progress_bar
+        super().__init__()
+
+    def set_size(self, size):
+        """``set_size()`` is an API that might be called with ``None`` if the
+        information is not already present on the caller. In that case, we'll
+        retrieve the size from an ``info()`` call."""
+        if size is None:
+            size = self.fs.info(self.path_info)["size"]
+        self.progress_bar.total = size
+        self.progress_bar.refresh()
+
+    def relative_update(self, inc=1):
+        self.progress_bar.update(inc)
+
+    def absolute_update(self, value):
+        self.progress_bar.update_to(value)

dvc/remote/__init__.py

@@ -1,11 +0,0 @@
-from ..fs import get_cloud_fs
-from .base import Remote
-from .local import LocalRemote
-
-
-def get_remote(repo, **kwargs):
-    cls, config = get_cloud_fs(repo, **kwargs)
-    fs = cls(**config)
-    if fs.scheme == "local":
-        return LocalRemote(fs, repo.tmp_dir, **config)
-    return Remote(fs, repo.tmp_dir, **config)

dvc/remote/base.py

@@ -1,522 +0,0 @@
-import errno
-import hashlib
-import itertools
-import logging
-from concurrent import futures
-from concurrent.futures import ThreadPoolExecutor
-from functools import partial, wraps
-
-from dvc.exceptions import DownloadError, UploadError
-from dvc.hash_info import HashInfo
-
-from ..progress import Tqdm
-from .index import RemoteIndex, RemoteIndexNoop
-
-logger = logging.getLogger(__name__)
-
-STATUS_OK = 1
-STATUS_MISSING = 2
-STATUS_NEW = 3
-STATUS_DELETED = 4
-
-STATUS_MAP = {
-    # (local_exists, remote_exists)
-    (True, True): STATUS_OK,
-    (False, False): STATUS_MISSING,
-    (True, False): STATUS_NEW,
-    (False, True): STATUS_DELETED,
-}
-
-
-def _log_exceptions(func, operation):
-    @wraps(func)
-    def wrapper(from_info, to_info, *args, **kwargs):
-        try:
-            func(from_info, to_info, *args, **kwargs)
-            return 0
-        except Exception as exc:  # pylint: disable=broad-except
-            # NOTE: this means we ran out of file descriptors and there is no
-            # reason to try to proceed, as we will hit this error anyways.
-            # pylint: disable=no-member
-            if isinstance(exc, OSError) and exc.errno == errno.EMFILE:
-                raise
-
-            logger.exception(
-                "failed to %s '%s' to '%s'", operation, from_info, to_info
-            )
-            return 1
-
-    return wrapper
-
-
-def index_locked(f):
-    @wraps(f)
-    def wrapper(obj, *args, **kwargs):
-        with obj.index:
-            return f(obj, *args, **kwargs)
-
-    return wrapper
-
-
-class Remote:
-    """Cloud remote class.
-
-    Provides methods for indexing and garbage collecting fss which contain
-    DVC remotes.
-    """
-
-    INDEX_CLS = RemoteIndex
-
-    def __init__(self, fs, tmp_dir, **config):
-        from dvc.objects.db import get_odb
-
-        self.fs = fs
-        self.odb = get_odb(self.fs, **config)
-
-        url = config.get("url")
-        if url:
-            index_name = hashlib.sha256(url.encode("utf-8")).hexdigest()
-            self.index = self.INDEX_CLS(
-                tmp_dir, index_name, dir_suffix=self.fs.CHECKSUM_DIR_SUFFIX
-            )
-        else:
-            self.index = RemoteIndexNoop()
-
-    def __repr__(self):
-        return "{class_name}: '{path_info}'".format(
-            class_name=type(self).__name__,
-            path_info=self.fs.path_info or "No path",
-        )
-
-    @index_locked
-    def gc(self, *args, **kwargs):
-        removed = self.odb.gc(*args, **kwargs)
-
-        if removed:
-            self.index.clear()
-
-        return removed
-
-    @index_locked
-    def status(
-        self,
-        cache,
-        named_cache,
-        jobs=None,
-        show_checksums=False,
-        download=False,
-        log_missing=True,
-    ):
-        # Return flattened dict containing all status info
-        dir_status, file_status, _ = self._status(
-            cache,
-            named_cache,
-            jobs=jobs,
-            show_checksums=show_checksums,
-            download=download,
-            log_missing=log_missing,
-        )
-        return dict(dir_status, **file_status)
-
-    def hashes_exist(self, hashes, **kwargs):
-        hashes = set(hashes)
-        indexed_hashes = set(self.index.intersection(hashes))
-        hashes -= indexed_hashes
-        indexed_hashes = list(indexed_hashes)
-        logger.debug("Matched '{}' indexed hashes".format(len(indexed_hashes)))
-        if not hashes:
-            return indexed_hashes
-
-        return indexed_hashes + self.odb.hashes_exist(list(hashes), **kwargs)
-
-    def _status(
-        self,
-        cache,
-        named_cache,
-        jobs=None,
-        show_checksums=False,
-        download=False,
-        log_missing=True,
-    ):
-        """Return a tuple of (dir_status_info, file_status_info, dir_contents).
-
-        dir_status_info contains status for .dir files, file_status_info
-        contains status for all other files, and dir_contents is a dict of
-        {dir_hash: set(file_hash, ...)} which can be used to map
-        a .dir file to its file contents.
-        """
-        logger.debug(f"Preparing to collect status from {self.fs.path_info}")
-        md5s = set(named_cache.scheme_keys(cache.fs.scheme))
-
-        logger.debug("Collecting information from local cache...")
-        local_exists = frozenset(
-            cache.hashes_exist(md5s, jobs=jobs, name=cache.cache_dir)
-        )
-
-        # This is a performance optimization. We can safely assume that,
-        # if the resources that we want to fetch are already cached,
-        # there's no need to check the remote storage for the existence of
-        # those files.
-        if download and local_exists == md5s:
-            remote_exists = local_exists
-        else:
-            logger.debug("Collecting information from remote cache...")
-            remote_exists = set()
-            dir_md5s = set(named_cache.dir_keys(cache.fs.scheme))
-            if dir_md5s:
-                remote_exists.update(
-                    self._indexed_dir_hashes(cache, named_cache, dir_md5s)
-                )
-                md5s.difference_update(remote_exists)
-            if md5s:
-                remote_exists.update(
-                    self.hashes_exist(
-                        md5s, jobs=jobs, name=str(self.fs.path_info)
-                    )
-                )
-        return self._make_status(
-            cache,
-            named_cache,
-            show_checksums,
-            local_exists,
-            remote_exists,
-            log_missing,
-        )
-
-    def _make_status(
-        self,
-        cache,
-        named_cache,
-        show_checksums,
-        local_exists,
-        remote_exists,
-        log_missing,
-    ):
-        def make_names(hash_, names):
-            return {"name": hash_ if show_checksums else " ".join(names)}
-
-        dir_status = {}
-        file_status = {}
-        dir_contents = {}
-        for hash_, item in named_cache[cache.fs.scheme].items():
-            if item.children:
-                dir_status[hash_] = make_names(hash_, item.names)
-                dir_contents[hash_] = set()
-                for child_hash, child in item.children.items():
-                    file_status[child_hash] = make_names(
-                        child_hash, child.names
-                    )
-                    dir_contents[hash_].add(child_hash)
-            else:
-                file_status[hash_] = make_names(hash_, item.names)
-
-        self._fill_statuses(dir_status, local_exists, remote_exists)
-        self._fill_statuses(file_status, local_exists, remote_exists)
-
-        if log_missing:
-            self._log_missing_caches(dict(dir_status, **file_status))
-
-        return dir_status, file_status, dir_contents
-
-    def _indexed_dir_hashes(self, cache, named_cache, dir_md5s):
-        # Validate our index by verifying all indexed .dir hashes
-        # still exist on the remote
-        indexed_dirs = set(self.index.dir_hashes())
-        indexed_dir_exists = set()
-        if indexed_dirs:
-            indexed_dir_exists.update(
-                self.odb.list_hashes_exists(indexed_dirs)
-            )
-            missing_dirs = indexed_dirs.difference(indexed_dir_exists)
-            if missing_dirs:
-                logger.debug(
-                    "Remote cache missing indexed .dir hashes '{}', "
-                    "clearing remote index".format(", ".join(missing_dirs))
-                )
-                self.index.clear()
-
-        # Check if non-indexed (new) dir hashes exist on remote
-        dir_exists = dir_md5s.intersection(indexed_dir_exists)
-        dir_exists.update(self.odb.list_hashes_exists(dir_md5s - dir_exists))
-
-        # If .dir hash exists on the remote, assume directory contents
-        # still exists on the remote
-        for dir_hash in dir_exists:
-            file_hashes = list(
-                named_cache.child_keys(cache.fs.scheme, dir_hash)
-            )
-            if dir_hash not in self.index:
-                logger.debug(
-                    "Indexing new .dir '{}' with '{}' nested files".format(
-                        dir_hash, len(file_hashes)
-                    )
-                )
-                self.index.update([dir_hash], file_hashes)
-            yield dir_hash
-            yield from file_hashes
-
-    @staticmethod
-    def _fill_statuses(hash_info_dir, local_exists, remote_exists):
-        # Using sets because they are way faster for lookups
-        local = set(local_exists)
-        remote = set(remote_exists)
-
-        for md5, info in hash_info_dir.items():
-            status = STATUS_MAP[(md5 in local, md5 in remote)]
-            info["status"] = status
-
-    def _get_plans(self, cache_obj, download, status_info, status):
-        cache = []
-        path_infos = []
-        names = []
-        hashes = []
-        missing = []
-        for md5, info in Tqdm(
-            status_info.items(), desc="Analysing status", unit="file"
-        ):
-            if info["status"] == status:
-                cache.append(cache_obj.hash_to_path_info(md5))
-                path_infos.append(self.odb.hash_to_path_info(md5))
-                names.append(info["name"])
-                hashes.append(md5)
-            elif info["status"] == STATUS_MISSING:
-                missing.append(md5)
-
-        if download:
-            to_infos = cache
-            from_infos = path_infos
-        else:
-            to_infos = path_infos
-            from_infos = cache
-
-        return (from_infos, to_infos, names, hashes), missing
-
-    def _process(
-        self,
-        cache,
-        named_cache,
-        jobs=None,
-        show_checksums=False,
-        download=False,
-    ):
-        logger.debug(
-            "Preparing to {} '{}'".format(
-                "download data from" if download else "upload data to",
-                self.fs.path_info,
-            )
-        )
-
-        if download:
-            func = _log_exceptions(self.fs.download_file, "download")
-            status = STATUS_DELETED
-            desc = "Downloading"
-        else:
-            func = _log_exceptions(self.fs.upload, "upload")
-            status = STATUS_NEW
-            desc = "Uploading"
-
-        if jobs is None:
-            jobs = self.fs.jobs
-
-        dir_status, file_status, dir_contents = self._status(
-            cache,
-            named_cache,
-            jobs=jobs,
-            show_checksums=show_checksums,
-            download=download,
-        )
-
-        dir_plans, _ = self._get_plans(cache, download, dir_status, status)
-        file_plans, missing_files = self._get_plans(
-            cache, download, file_status, status
-        )
-
-        total = len(dir_plans[0]) + len(file_plans[0])
-        if total == 0:
-            return 0
-
-        with Tqdm(total=total, unit="file", desc=desc) as pbar:
-            func = pbar.wrap_fn(func)
-            self._process_plans(
-                download,
-                dir_plans,
-                file_plans,
-                dir_contents,
-                missing_files,
-                jobs,
-                func,
-            )
-
-        return total
-
-    def _process_plans(
-        self,
-        download,
-        dir_plans,
-        file_plans,
-        dir_contents,
-        missing_files,
-        jobs,
-        func,
-    ):
-        with ThreadPoolExecutor(max_workers=jobs) as executor:
-            processor = partial(self._create_tasks, executor, jobs, func)
-            processor.transfer_func = func
-            if download:
-                self._download_plans(dir_plans, file_plans, processor)
-            else:
-                self._upload_plans(
-                    dir_plans,
-                    file_plans,
-                    dir_contents,
-                    missing_files,
-                    processor,
-                )
-
-    def _create_tasks(self, executor, jobs, func, file_plans):
-        fails = 0
-        file_plan_iterator = iter(file_plans)
-
-        def create_taskset(amount):
-            return {
-                executor.submit(func, from_info, to_info, name)
-                for from_info, to_info, name, _ in itertools.islice(
-                    file_plan_iterator, amount
-                )
-            }
-
-        tasks = create_taskset(jobs * 5)
-        while tasks:
-            done, tasks = futures.wait(
-                tasks, return_when=futures.FIRST_COMPLETED
-            )
-            fails += sum(task.result() for task in done)
-            tasks.update(create_taskset(len(done)))
-        return fails
-
-    def _download_plans(self, dir_plans, file_plans, processor):
-        plans = [*zip(*dir_plans), *zip(*file_plans)]
-        fails = processor(plans)
-        if fails:
-            self.index.clear()
-            raise DownloadError(fails)
-
-    def _upload_plans(
-        self, dir_plans, file_plans, dir_contents, missing_files, processor
-    ):
-        total_fails = 0
-        succeeded_dir_hashes = []
-        all_file_plans = list(zip(*file_plans))
-        for dir_from_info, dir_to_info, dir_name, dir_hash in zip(*dir_plans):
-            bound_file_plans = []
-            directory_hashes = dir_contents[dir_hash]
-
-            for file_plan in all_file_plans.copy():
-                if file_plan[-1] in directory_hashes:
-                    bound_file_plans.append(file_plan)
-                    all_file_plans.remove(file_plan)
-
-            dir_fails = processor(bound_file_plans)
-            if dir_fails:
-                logger.debug(
-                    "failed to upload full contents of '{}', "
-                    "aborting .dir file upload".format(dir_name)
-                )
-                logger.error(
-                    f"failed to upload '{dir_from_info}'"
-                    f" to '{dir_to_info}'"
-                )
-                total_fails += dir_fails + 1
-            elif directory_hashes.intersection(missing_files):
-                # if for some reason a file contained in this dir is
-                # missing both locally and in the remote, we want to
-                # push whatever file content we have, but should not
-                # push .dir file
-                logger.debug(
-                    "directory '%s' contains missing files,"
-                    "skipping .dir file upload",
-                    dir_name,
-                )
-            else:
-                is_dir_failed = processor.transfer_func(
-                    dir_from_info, dir_to_info, dir_name
-                )
-                total_fails += is_dir_failed
-                if not is_dir_failed:
-                    succeeded_dir_hashes.append(dir_hash)
-
-        # insert the rest
-        total_fails += processor(all_file_plans)
-        if total_fails:
-            raise UploadError(total_fails)
-
-        # index successfully pushed dirs
-        for dir_hash in succeeded_dir_hashes:
-            file_hashes = dir_contents[dir_hash]
-            logger.debug(
-                "Indexing pushed dir '{}' with "
-                "'{}' nested files".format(dir_hash, len(file_hashes))
-            )
-            self.index.update([dir_hash], file_hashes)
-
-    @index_locked
-    def push(self, cache, named_cache, jobs=None, show_checksums=False):
-        ret = self._process(
-            cache,
-            named_cache,
-            jobs=jobs,
-            show_checksums=show_checksums,
-            download=False,
-        )
-
-        if self.fs.scheme == "local":
-            for checksum in named_cache.scheme_keys("local"):
-                cache_file = self.odb.hash_to_path_info(checksum)
-                if self.fs.exists(cache_file):
-                    hash_info = HashInfo(self.fs.PARAM_CHECKSUM, checksum)
-                    self.odb.state.save(cache_file, self.fs, hash_info)
-                    self.odb.protect(cache_file)
-
-        return ret
-
-    @index_locked
-    def pull(self, cache, named_cache, jobs=None, show_checksums=False):
-        ret = self._process(
-            cache,
-            named_cache,
-            jobs=jobs,
-            show_checksums=show_checksums,
-            download=True,
-        )
-
-        if not self.odb.verify:
-            for checksum in named_cache.scheme_keys("local"):
-                cache_file = cache.hash_to_path_info(checksum)
-                if cache.fs.exists(cache_file):
-                    # We can safely save here, as existing corrupted files
-                    # will be removed upon status, while files corrupted
-                    # during download will not be moved from tmp_file
-                    # (see `BaseFileSystem.download()`)
-                    hash_info = HashInfo(cache.fs.PARAM_CHECKSUM, checksum)
-                    cache.state.save(cache_file, cache.fs, hash_info)
-                    cache.protect(cache_file)
-
-        return ret
-
-    @staticmethod
-    def _log_missing_caches(hash_info_dict):
-        missing_caches = [
-            (md5, info)
-            for md5, info in hash_info_dict.items()
-            if info["status"] == STATUS_MISSING
-        ]
-        if missing_caches:
-            missing_desc = "\n".join(
-                "name: {}, md5: {}".format(info["name"], md5)
-                for md5, info in missing_caches
-            )
-            msg = (
-                "Some of the cache files do not exist neither locally "
-                "nor on remote. Missing cache files:\n{}".format(missing_desc)
-            )
-            logger.warning(msg)

dvc/remote/index.py

@@ -1,213 +0,0 @@
-import logging
-import os
-import sqlite3
-import threading
-
-from funcy import lchunks
-
-from dvc.state import _connect_sqlite
-from dvc.utils.fs import makedirs
-
-logger = logging.getLogger(__name__)
-
-
-class RemoteIndexNoop:
-    """No-op class for remotes which are not indexed (i.e. local)."""
-
-    def __init__(self, *args, **kwargs):
-        pass
-
-    def __enter__(self):
-        pass
-
-    def __exit__(self, typ, value, tbck):
-        pass
-
-    def __iter__(self):
-        return iter([])
-
-    def __contains__(self, hash_):
-        return False
-
-    @staticmethod
-    def hashes():
-        return []
-
-    @staticmethod
-    def dir_hashes():
-        return []
-
-    def load(self):
-        pass
-
-    def dump(self):
-        pass
-
-    def clear(self):
-        pass
-
-    def update(self, *args):
-        pass
-
-    @staticmethod
-    def intersection(*args):
-        return []
-
-
-class RemoteIndex:
-    """Class for indexing remote hashes in a sqlite3 database.
-
-    Args:
-        repo: repo for this remote index.
-        name: name for this index. Index db will be loaded from and saved to
-            ``.dvc/tmp/index/{name}.idx``.
-        dir_suffix: suffix used for naming directory hashes
-    """
-
-    INDEX_SUFFIX = ".idx"
-    VERSION = 1
-    INDEX_TABLE = "remote_index"
-    INDEX_TABLE_LAYOUT = "checksum TEXT PRIMARY KEY, dir INTEGER NOT NULL"
-    INDEX_DIR = "index"
-
-    def __init__(self, tmp_dir, name, dir_suffix=".dir"):
-        self.path = os.path.join(
-            tmp_dir, self.INDEX_DIR, f"{name}{self.INDEX_SUFFIX}"
-        )
-
-        self.dir_suffix = dir_suffix
-        self.database = None
-        self.cursor = None
-        self.modified = False
-        self.lock = threading.Lock()
-
-    def __iter__(self):
-        cmd = f"SELECT checksum FROM {self.INDEX_TABLE}"
-        for (hash_,) in self._execute(cmd):
-            yield hash_
-
-    def __enter__(self):
-        self.lock.acquire()
-        self.load()
-
-    def __exit__(self, typ, value, tbck):
-        self.dump()
-        self.lock.release()
-
-    def __contains__(self, hash_):
-        cmd = "SELECT checksum FROM {} WHERE checksum = (?)".format(
-            self.INDEX_TABLE
-        )
-        self._execute(cmd, (hash_,))
-        return self.cursor.fetchone() is not None
-
-    def hashes(self):
-        """Iterate over hashes stored in the index."""
-        return iter(self)
-
-    def dir_hashes(self):
-        """Iterate over .dir hashes stored in the index."""
-        cmd = f"SELECT checksum FROM {self.INDEX_TABLE} WHERE dir = 1"
-        for (hash_,) in self._execute(cmd):
-            yield hash_
-
-    def is_dir_hash(self, hash_):
-        return hash_.endswith(self.dir_suffix)
-
-    def _execute(self, cmd, parameters=()):
-        return self.cursor.execute(cmd, parameters)
-
-    def _executemany(self, cmd, seq_of_parameters):
-        return self.cursor.executemany(cmd, seq_of_parameters)
-
-    def _prepare_db(self, empty=False):
-        if not empty:
-            cmd = "PRAGMA user_version;"
-            self._execute(cmd)
-            ret = self.cursor.fetchall()
-            assert len(ret) == 1
-            assert len(ret[0]) == 1
-            assert isinstance(ret[0][0], int)
-            version = ret[0][0]
-
-            if version != self.VERSION:
-                logger.error(
-                    "Index file version '{}' will be reformatted "
-                    "to the current version '{}'.".format(
-                        version, self.VERSION
-                    )
-                )
-                cmd = "DROP TABLE IF EXISTS {};"
-                self._execute(cmd.format(self.INDEX_TABLE))
-
-        cmd = "CREATE TABLE IF NOT EXISTS {} ({})"
-        self._execute(cmd.format(self.INDEX_TABLE, self.INDEX_TABLE_LAYOUT))
-
-        cmd = "PRAGMA user_version = {};"
-        self._execute(cmd.format(self.VERSION))
-
-    def load(self):
-        """(Re)load this index database."""
-        retries = 1
-        while True:
-            assert self.database is None
-            assert self.cursor is None
-
-            empty = not os.path.isfile(self.path)
-            makedirs(os.path.dirname(self.path), exist_ok=True)
-            self.database = _connect_sqlite(self.path, {"nolock": 1})
-            self.cursor = self.database.cursor()
-
-            try:
-                self._prepare_db(empty=empty)
-                return
-            except sqlite3.DatabaseError:
-                self.cursor.close()
-                self.database.close()
-                self.database = None
-                self.cursor = None
-                if retries > 0:
-                    os.unlink(self.path)
-                    retries -= 1
-                else:
-                    raise
-
-    def dump(self):
-        """Save this index database."""
-        assert self.database is not None
-
-        self.database.commit()
-        self.cursor.close()
-        self.database.close()
-        self.database = None
-        self.cursor = None
-
-    def clear(self):
-        """Clear this index (to force re-indexing later).
-
-        Changes to the index will not committed until dump() is called.
-        """
-        cmd = f"DELETE FROM {self.INDEX_TABLE}"
-        self._execute(cmd)
-
-    def update(self, dir_hashes, file_hashes):
-        """Update this index, adding the specified hashes.
-
-        Changes to the index will not committed until dump() is called.
-        """
-        cmd = "INSERT OR IGNORE INTO {} (checksum, dir) VALUES (?, ?)".format(
-            self.INDEX_TABLE
-        )
-        self._executemany(cmd, ((hash_, True) for hash_ in dir_hashes))
-        self._executemany(cmd, ((hash_, False) for hash_ in file_hashes))
-
-    def intersection(self, hashes):
-        """Iterate over values from `hashes` which exist in the index."""
-        # sqlite has a compile time limit of 999, see:
-        # https://www.sqlite.org/c3ref/c_limit_attached.html#sqlitelimitvariablenumber
-        for chunk in lchunks(999, hashes):
-            cmd = "SELECT checksum FROM {} WHERE checksum IN ({})".format(
-                self.INDEX_TABLE, ",".join("?" for hash_ in chunk)
-            )
-            for (hash_,) in self._execute(cmd, chunk):
-                yield hash_

dvc/remote/local.py

@@ -1,6 +0,0 @@
-from .base import Remote
-from .index import RemoteIndexNoop
-
-
-class LocalRemote(Remote):
-    INDEX_CLS = RemoteIndexNoop  # type: ignore[assignment]

dvc/repo/__init__.py

@@ -1,10 +1,11 @@
 import logging
 import os
+from collections import defaultdict
 from contextlib import contextmanager
 from functools import wraps
-from typing import TYPE_CHECKING, Callable, Optional
+from typing import TYPE_CHECKING, Callable, Optional, Set
 
-from funcy import cached_property, cat
+from funcy import cached_property
 
 from dvc.exceptions import FileMissingError
 from dvc.exceptions import IsADirectoryError as DvcIsADirectoryError
@@ -13,11 +14,9 @@ from dvc.ignore import DvcIgnoreFilter
 from dvc.path_info import PathInfo
 from dvc.utils.fs import path_isin
 
-from .graph import build_graph, build_outs_graph, get_pipelines
-from .trie import build_outs_trie
-
 if TYPE_CHECKING:
     from dvc.fs.base import BaseFileSystem
+    from dvc.objects.file import HashFile
     from dvc.scm import Base
 
 
@@ -56,7 +55,6 @@ class Repo:
     DVC_DIR = ".dvc"
 
     from dvc.repo.add import add
-    from dvc.repo.brancher import brancher
     from dvc.repo.checkout import checkout
     from dvc.repo.commit import commit
     from dvc.repo.destroy import destroy
@@ -206,6 +204,12 @@ class Repo:
     def __str__(self):
         return self.url or self.root_dir
 
+    @cached_property
+    def index(self):
+        from dvc.repo.index import Index
+
+        return Index(self)
+
     @staticmethod
     def open(url, *args, **kwargs):
         if url is None:
@@ -322,25 +326,12 @@ class Repo:
 
         self.scm.ignore_list(flist)
 
-    def check_modified_graph(self, new_stages, old_stages=None):
-        """Generate graph including the new stage to check for errors"""
-        # Building graph might be costly for the ones with many DVC-files,
-        # so we provide this undocumented hack to skip it. See [1] for
-        # more details. The hack can be used as:
-        #
-        #     repo = Repo(...)
-        #     repo._skip_graph_checks = True
-        #     repo.add(...)
-        #
-        # A user should care about not duplicating outs and not adding cycles,
-        # otherwise DVC might have an undefined behaviour.
-        #
-        # [1] https://github.com/iterative/dvc/issues/2671
-        if not getattr(self, "_skip_graph_checks", False):
-            existing_stages = self.stages if old_stages is None else old_stages
-            build_graph(existing_stages + new_stages)
+    def brancher(self, *args, **kwargs):
+        from dvc.repo.brancher import brancher
 
-    def used_cache(
+        return brancher(self, *args, **kwargs)
+
+    def used_objs(
         self,
         targets=None,
         all_branches=False,
@@ -366,13 +357,20 @@ class Repo:
         the scope.
 
         Returns:
-            A dictionary with Schemes (representing output's location) mapped
-            to items containing the output's `dumpd` names and the output's
-            children (if the given output is a directory).
+            A dict mapping (remote) ODB instances to sets of objects that
+            belong to each ODB. If the ODB instance is None, the objects
+            are naive and do not belong to a specific remote ODB.
         """
-        from dvc.objects.db import NamedCache
+        used = defaultdict(set)
 
-        cache = NamedCache()
+        def _add_suffix(objs: Set["HashFile"], suffix: str) -> None:
+            from itertools import chain
+
+            from dvc.objects import iterobjs
+
+            for obj in chain.from_iterable(map(iterobjs, objs)):
+                if obj.name is not None:
+                    obj.name += suffix
 
         for branch in self.brancher(
             revs=revs,
@@ -381,66 +379,33 @@ class Repo:
             all_commits=all_commits,
             all_experiments=all_experiments,
         ):
-            targets = targets or [None]
-
-            pairs = cat(
-                self.stage.collect_granular(
-                    target, recursive=recursive, with_deps=with_deps
-                )
-                for target in targets
-            )
-
-            suffix = f"({branch})" if branch else ""
-            for stage, filter_info in pairs:
-                used_cache = stage.get_used_cache(
-                    remote=remote,
-                    force=force,
-                    jobs=jobs,
-                    filter_info=filter_info,
-                )
-                cache.update(used_cache, suffix=suffix)
+            for odb, objs in self.index.used_objs(
+                targets,
+                remote=remote,
+                force=force,
+                jobs=jobs,
+                recursive=recursive,
+                with_deps=with_deps,
+            ).items():
+                if branch:
+                    _add_suffix(objs, f" ({branch})")
+                used[odb].update(objs)
 
         if used_run_cache:
-            used_cache = self.stage_cache.get_used_cache(
+            for odb, objs in self.stage_cache.get_used_objs(
                 used_run_cache, remote=remote, force=force, jobs=jobs
-            )
-            cache.update(used_cache)
+            ).items():
+                used[odb].update(objs)
 
-        return cache
+        return used
 
-    @cached_property
-    def outs_trie(self):
-        return build_outs_trie(self.stages)
-
-    @cached_property
-    def graph(self):
-        return build_graph(self.stages, self.outs_trie)
-
-    @cached_property
-    def outs_graph(self):
-        return build_outs_graph(self.graph, self.outs_trie)
-
-    @cached_property
-    def pipelines(self):
-        return get_pipelines(self.graph)
-
-    @cached_property
-    def stages(self):
-        """
-        Walks down the root directory looking for Dvcfiles,
-        skipping the directories that are related with
-        any SCM (e.g. `.git`), DVC itself (`.dvc`), or directories
-        tracked by DVC (e.g. `dvc add data` would skip `data/`)
-
-        NOTE: For large repos, this could be an expensive
-              operation. Consider using some memoization.
-        """
-        error_handler = self.stage_collection_error_handler
-        return self.stage.collect_repo(onerror=error_handler)
+    @property
+    def stages(self):  # obsolete, only for backward-compatibility
+        return self.index.stages
 
     def find_outs_by_path(self, path, outs=None, recursive=False, strict=True):
         # using `outs_graph` to ensure graph checks are run
-        outs = outs or self.outs_graph
+        outs = outs or self.index.outs_graph
 
         abs_path = os.path.abspath(path)
         path_info = PathInfo(abs_path)
@@ -501,11 +466,7 @@ class Repo:
     def _reset(self):
         self.state.close()
         self.scm._reset()  # pylint: disable=protected-access
-        self.__dict__.pop("outs_trie", None)
-        self.__dict__.pop("outs_graph", None)
-        self.__dict__.pop("graph", None)
-        self.__dict__.pop("stages", None)
-        self.__dict__.pop("pipelines", None)
+        self.__dict__.pop("index", None)
         self.__dict__.pop("dvcignore", None)
 
     def __enter__(self):

dvc/repo/add.py

@@ -1,9 +1,13 @@
 import logging
 import os
-from typing import TYPE_CHECKING
+from contextlib import contextmanager
+from itertools import tee
+from typing import TYPE_CHECKING, Any, Iterator, List
 
 import colorama
 
+from dvc.ui import ui
+
 from ..exceptions import (
     CacheLinkError,
     InvalidArgumentError,
@@ -11,46 +15,58 @@ from ..exceptions import (
     OverlappingOutputPathsError,
     RecursiveAddingWhileUsingFilename,
 )
-from ..progress import Tqdm
 from ..repo.scm_context import scm_context
 from ..utils import LARGE_DIR_SIZE, glob_targets, resolve_output, resolve_paths
+from ..utils.collections import ensure_list, validate
 from . import locked
 
 if TYPE_CHECKING:
+    from dvc.repo import Repo
+    from dvc.stage import Stage
     from dvc.types import TargetType
 
-
+Stages = List["Stage"]
 logger = logging.getLogger(__name__)
 
 
-@locked
-@scm_context
-def add(  # noqa: C901
-    repo,
-    targets: "TargetType",
-    recursive=False,
-    no_commit=False,
-    fname=None,
-    to_remote=False,
-    **kwargs,
-):
-    from dvc.utils.collections import ensure_list
+OVERLAPPING_OUTPUT_FMT = (
+    "Cannot add '{out}', because it is overlapping with other "
+    "DVC tracked output: '{parent}'.\n"
+    "To include '{out}' in '{parent}', run "
+    "'dvc commit {parent_stage}'"
+)
 
-    if recursive and fname:
+
+def check_recursive_and_fname(args):
+    if args.recursive and args.fname:
         raise RecursiveAddingWhileUsingFilename()
 
-    targets = ensure_list(targets)
 
-    to_cache = kwargs.get("out") and not to_remote
+def transform_targets(args):
+    from funcy import count_reps
+
+    counts = count_reps(ensure_list(args.targets))
+    dupes = [key for key, count in counts.items() if count > 1]
+    if dupes:
+        msg = ", ".join(f"[b]{key}[/]" for key in dupes)
+        ui.error_write(f"ignoring duplicated targets: {msg}", styled=True)
+    args.targets = list(counts)
+
+
+def check_arg_combinations(args):
+    kwargs = args.kwargs
     invalid_opt = None
+    to_remote = args.to_remote
+    to_cache = kwargs.get("out") and not to_remote
+
     if to_remote or to_cache:
         message = "{option} can't be used with "
         message += "--to-remote" if to_remote else "-o"
-        if len(targets) != 1:
+        if len(args.targets) != 1:
             invalid_opt = "multiple targets"
-        elif no_commit:
+        elif args.no_commit:
             invalid_opt = "--no-commit option"
-        elif recursive:
+        elif args.recursive:
             invalid_opt = "--recursive option"
         elif kwargs.get("external"):
             invalid_opt = "--external option"
@@ -64,214 +80,190 @@ def add(  # noqa: C901
     if invalid_opt is not None:
         raise InvalidArgumentError(message.format(option=invalid_opt))
 
-    link_failures = []
-    stages_list = []
-    num_targets = len(targets)
-    with Tqdm(total=num_targets, desc="Add", unit="file", leave=True) as pbar:
-        if num_targets == 1:
-            # clear unneeded top-level progress bar for single target
-            pbar.bar_format = "Adding..."
-            pbar.refresh()
-        for target in targets:
-            sub_targets = _find_all_targets(repo, target, recursive)
-            pbar.total += len(sub_targets) - 1
 
-            if os.path.isdir(target) and len(sub_targets) > LARGE_DIR_SIZE:
-                logger.warning(
-                    "You are adding a large directory '{target}' recursively,"
-                    " consider tracking it as a whole instead.\n"
-                    "{purple}HINT:{nc} Remove the generated DVC file and then"
-                    " run `{cyan}dvc add {target}{nc}`".format(
-                        purple=colorama.Fore.MAGENTA,
-                        cyan=colorama.Fore.CYAN,
-                        nc=colorama.Style.RESET_ALL,
-                        target=target,
-                    )
-                )
-
-            stages = _create_stages(
-                repo,
-                sub_targets,
-                fname,
-                pbar=pbar,
-                transfer=to_remote or to_cache,
-                **kwargs,
-            )
-
-            try:
-                repo.check_modified_graph(stages)
-            except OverlappingOutputPathsError as exc:
-                msg = (
-                    "Cannot add '{out}', because it is overlapping with other "
-                    "DVC tracked output: '{parent}'.\n"
-                    "To include '{out}' in '{parent}', run "
-                    "'dvc commit {parent_stage}'"
-                ).format(
-                    out=exc.overlapping_out.path_info,
-                    parent=exc.parent.path_info,
-                    parent_stage=exc.parent.stage.addressing,
-                )
-                raise OverlappingOutputPathsError(
-                    exc.parent, exc.overlapping_out, msg
-                )
-            except OutputDuplicationError as exc:
-                raise OutputDuplicationError(
-                    exc.output, list(set(exc.stages) - set(stages))
-                )
-
-            link_failures.extend(
-                _process_stages(
-                    repo,
-                    sub_targets,
-                    stages,
-                    no_commit,
-                    pbar,
-                    to_remote,
-                    to_cache,
-                    **kwargs,
-                )
-            )
-            stages_list += stages
-
-        if num_targets == 1:  # restore bar format for stats
-            pbar.bar_format = pbar.BAR_FMT_DEFAULT
-
-    if link_failures:
-        msg = (
-            "Some targets could not be linked from cache to workspace.\n{}\n"
-            "To re-link these targets, reconfigure cache types and then run:\n"
-            "\n\tdvc checkout {}"
-        ).format(
-            CacheLinkError.SUPPORT_LINK,
-            " ".join([str(stage.relpath) for stage in link_failures]),
+@contextmanager
+def translate_graph_error(stages: Stages) -> Iterator[None]:
+    try:
+        yield
+    except OverlappingOutputPathsError as exc:
+        msg = OVERLAPPING_OUTPUT_FMT.format(
+            out=exc.overlapping_out.path_info,
+            parent=exc.parent.path_info,
+            parent_stage=exc.parent.stage.addressing,
+        )
+        raise OverlappingOutputPathsError(exc.parent, exc.overlapping_out, msg)
+    except OutputDuplicationError as exc:
+        raise OutputDuplicationError(
+            exc.output, list(set(exc.stages) - set(stages))
         )
-        logger.warning(msg)
-
-    return stages_list
 
 
-def _process_stages(
-    repo, sub_targets, stages, no_commit, pbar, to_remote, to_cache, **kwargs
+def progress_iter(stages: Stages) -> Iterator["Stage"]:
+    total = len(stages)
+    desc = "Adding..."
+    with ui.progress(
+        stages, total=total, desc=desc, unit="file", leave=True
+    ) as pbar:
+        if total == 1:
+            pbar.bar_format = desc
+            pbar.refresh()
+
+        for stage in pbar:
+            if total > 1:
+                pbar.set_msg(f"{stage.outs[0]}")
+            yield stage
+            if total == 1:  # restore bar format for stats
+                # pylint: disable=no-member
+                pbar.bar_format = pbar.BAR_FMT_DEFAULT
+
+
+LINK_FAILURE_MESSAGE = (
+    "\nSome targets could not be linked from cache to workspace.\n{}\n"
+    "To re-link these targets, reconfigure cache types and then run:\n"
+    "\n\tdvc checkout {}"
+)
+
+
+@contextmanager
+def warn_link_failures() -> Iterator[List[str]]:
+    link_failures: List[str] = []
+    try:
+        yield link_failures
+    finally:
+        if link_failures:
+            msg = LINK_FAILURE_MESSAGE.format(
+                CacheLinkError.SUPPORT_LINK,
+                " ".join(link_failures),
+            )
+            ui.error_write(msg)
+
+
+VALIDATORS = (
+    check_recursive_and_fname,
+    transform_targets,
+    check_arg_combinations,
+)
+
+
+@validate(*VALIDATORS)
+@locked
+@scm_context
+def add(  # noqa: C901
+    repo: "Repo",
+    targets: "TargetType",
+    recursive: bool = False,
+    no_commit: bool = False,
+    fname: str = None,
+    to_remote: bool = False,
+    **kwargs: Any,
 ):
-    link_failures = []
-    from dvc.dvcfile import Dvcfile
+    to_cache = bool(kwargs.get("out")) and not to_remote
+    transfer = to_remote or to_cache
 
-    from ..output import OutputDoesNotExistError
+    glob = kwargs.get("glob", False)
+    add_targets = collect_targets(repo, targets, recursive, glob)
+    # pass one for creating stages, other one is used for iterating here
+    add_targets, sources = tee(add_targets)
 
-    if to_remote or to_cache:
-        # Already verified in the add()
-        (stage,) = stages
-        (target,) = sub_targets
-        (out,) = stage.outs
+    # collect targets and build stages as we go
+    desc = "Collecting targets"
+    stages_it = create_stages(repo, add_targets, fname, transfer, **kwargs)
+    stages = list(ui.progress(stages_it, desc=desc, unit="file"))
+    msg = "Collecting stages from the workspace"
+    with translate_graph_error(stages), ui.status(msg) as status:
+        # remove existing stages that are to-be replaced with these
+        # new stages for the graph checks.
+        new_index = repo.index.update(stages)
+        status.update("Checking graph")
+        new_index.check_graph()
 
-        odb = None
-        if to_remote:
-            remote = repo.cloud.get_remote(kwargs.get("remote"), "add")
-            odb = remote.odb
+    odb = None
+    if to_remote:
+        odb = repo.cloud.get_remote_odb(kwargs.get("remote"), "add")
 
-        out.transfer(target, odb=odb, jobs=kwargs.get("jobs"))
-
-        if to_cache:
-            out.checkout()
-
-        Dvcfile(repo, stage.path).dump(stage)
-        return link_failures
-
-    with Tqdm(
-        total=len(stages),
-        desc="Processing",
-        unit="file",
-        disable=len(stages) == 1,
-    ) as pbar_stages:
-        for stage in stages:
-            try:
-                stage.save()
-            except OutputDoesNotExistError:
-                pbar.n -= 1
-                raise
-
-            try:
-                if not no_commit:
-                    stage.commit()
-            except CacheLinkError:
-                link_failures.append(stage)
-
-            Dvcfile(repo, stage.path).dump(stage)
-            pbar_stages.update()
-
-    return link_failures
+    with warn_link_failures() as link_failures:
+        for stage, source in zip(progress_iter(stages), sources):
+            if to_remote or to_cache:
+                stage.transfer(source, to_remote=to_remote, odb=odb, **kwargs)
+            else:
+                try:
+                    stage.save()
+                    if not no_commit:
+                        stage.commit()
+                except CacheLinkError:
+                    link_failures.append(str(stage.relpath))
+            stage.dump()
+    return stages
 
 
-def _find_all_targets(repo, target, recursive):
+LARGE_DIR_RECURSIVE_ADD_WARNING = (
+    "You are adding a large directory '{target}' recursively.\n"
+    "Consider tracking it as a whole instead with "
+    "`{cyan}dvc add {target}{nc}`."
+)
+
+
+def collect_targets(
+    repo: "Repo",
+    targets: "TargetType",
+    recursive: bool = False,
+    glob: bool = False,
+) -> Iterator[str]:
+    for target in glob_targets(ensure_list(targets), glob=glob):
+        expanded_targets = _find_all_targets(repo, target, recursive=recursive)
+        for index, path in enumerate(expanded_targets):
+            if index == LARGE_DIR_SIZE:
+                msg = LARGE_DIR_RECURSIVE_ADD_WARNING.format(
+                    cyan=colorama.Fore.CYAN,
+                    nc=colorama.Style.RESET_ALL,
+                    target=target,
+                )
+                ui.error_write(msg)
+            yield path
+
+
+def _find_all_targets(
+    repo: "Repo", target: str, recursive: bool = False
+) -> Iterator[str]:
     from dvc.dvcfile import is_dvc_file
 
     if os.path.isdir(target) and recursive:
-        return [
-            os.fspath(path)
-            for path in Tqdm(
-                repo.dvcignore.walk_files(repo.fs, target),
-                desc="Searching " + target,
-                bar_format=Tqdm.BAR_FMT_NOTOTAL,
-                unit="file",
-            )
-            if not repo.is_dvc_internal(os.fspath(path))
-            if not is_dvc_file(os.fspath(path))
-            if not repo.scm.belongs_to_scm(os.fspath(path))
-            if not repo.scm.is_tracked(os.fspath(path))
-        ]
-    return [target]
-
-
-def _create_stages(
-    repo,
-    targets,
-    fname,
-    pbar=None,
-    external=False,
-    glob=False,
-    desc=None,
-    transfer=False,
-    **kwargs,
-):
-    from dvc.dvcfile import Dvcfile
-    from dvc.stage import Stage, create_stage, restore_meta
-
-    expanded_targets = glob_targets(targets, glob=glob)
-
-    stages = []
-    for out in Tqdm(
-        expanded_targets,
-        desc="Creating DVC files",
-        disable=len(expanded_targets) < LARGE_DIR_SIZE,
-        unit="file",
-    ):
-        if kwargs.get("out"):
-            out = resolve_output(out, kwargs["out"])
-        path, wdir, out = resolve_paths(
-            repo, out, always_local=transfer and not kwargs.get("out")
+        files = map(os.fspath, repo.dvcignore.walk_files(repo.fs, target))
+        yield from (
+            path
+            for path in files
+            if not repo.is_dvc_internal(path)
+            if not is_dvc_file(path)
+            if not repo.scm.belongs_to_scm(path)
+            if not repo.scm.is_tracked(path)
         )
-        stage = create_stage(
-            Stage,
-            repo,
-            fname or path,
+    else:
+        yield target
+
+
+def create_stages(
+    repo: "Repo",
+    targets: Iterator[str],
+    fname: str = None,
+    transfer: bool = False,
+    external: bool = False,
+    **kwargs: Any,
+) -> Iterator["Stage"]:
+    for target in targets:
+        if kwargs.get("out"):
+            target = resolve_output(target, kwargs["out"])
+        path, wdir, out = resolve_paths(
+            repo, target, always_local=transfer and not kwargs.get("out")
+        )
+
+        stage = repo.stage.create(
+            single_stage=True,
+            validate=False,
+            fname=fname or path,
             wdir=wdir,
             outs=[out],
             external=external,
         )
-        restore_meta(stage)
-        Dvcfile(repo, stage.path).remove()
-        if desc:
-            stage.outs[0].desc = desc
-
-        repo._reset()  # pylint: disable=protected-access
-
-        if not stage:
-            if pbar is not None:
-                pbar.total -= 1
-            continue
-
-        stages.append(stage)
-        if pbar is not None:
-            pbar.update_msg(out)
-
-    return stages
+        if kwargs.get("desc"):
+            stage.outs[0].desc = kwargs["desc"]
+        yield stage

dvc/repo/checkout.py

@@ -28,13 +28,7 @@ def _fspath_dir(path):
 
 
 def _remove_unused_links(repo):
-    used = [
-        out.fspath
-        for stage in repo.stages
-        for out in stage.outs
-        if out.scheme == "local"
-    ]
-
+    used = [out.fspath for out in repo.index.outs if out.scheme == "local"]
     unused = repo.state.get_unused_links(used, repo.fs)
     ret = [_fspath_dir(u) for u in unused]
     repo.state.remove_links(unused, repo.fs)

dvc/repo/collect.py

@@ -20,12 +20,9 @@ DvcPaths = List[DvcPath]
 def _collect_outs(
     repo: "Repo", output_filter: FilterFn = None, deps: bool = False
 ) -> Outputs:
-    outs = [
-        out
-        for stage in repo.graph  # using `graph` to ensure graph checks run
-        for out in (stage.deps if deps else stage.outs)
-    ]
-    return list(filter(output_filter, outs)) if output_filter else outs
+    index = repo.index
+    index.check_graph()  # ensure graph is correct
+    return list(filter(output_filter, index.deps if deps else index.outs))
 
 
 def _collect_paths(
@@ -46,15 +43,12 @@ def _collect_paths(
             target_infos.extend(repo.dvcignore.walk_files(fs, path_info))
 
         if not fs.exists(path_info):
-            if not recursive:
-                if rev == "workspace" or rev == "":
-                    logger.warning(
-                        "'%s' was not found in current workspace.", path_info
-                    )
-                else:
-                    logger.warning(
-                        "'%s' was not found at: '%s'.", path_info, rev
-                    )
+            if rev == "workspace" or rev == "":
+                logger.warning(
+                    "'%s' was not found in current workspace.", path_info
+                )
+            else:
+                logger.warning("'%s' was not found at: '%s'.", path_info, rev)
             continue
         target_infos.append(path_info)
     return target_infos

dvc/repo/destroy.py

@@ -5,7 +5,7 @@ from . import locked
 
 @locked
 def _destroy_stages(repo):
-    for stage in repo.stages:
+    for stage in repo.index.stages:
         stage.unprotect_outs()
         stage.dvcfile.remove(force=True)
 

dvc/repo/diff.py

@@ -5,7 +5,6 @@ from typing import Dict, List
 
 from dvc.exceptions import PathMissingError
 from dvc.objects.stage import get_file_hash
-from dvc.objects.stage import stage as ostage
 from dvc.repo import locked
 from dvc.repo.experiments.utils import fix_exp_head
 
@@ -116,6 +115,7 @@ def _paths_checksums(repo, repo_fs, targets):
 
 def _output_paths(repo, repo_fs, targets):
     from dvc.fs.local import LocalFileSystem
+    from dvc.objects.stage import stage as ostage
 
     on_working_fs = isinstance(repo.fs, LocalFileSystem)
 
@@ -134,25 +134,28 @@ def _output_paths(repo, repo_fs, targets):
     def _to_checksum(output):
         if on_working_fs:
             return ostage(
-                repo.odb.local, output.path_info, repo.odb.local.fs, "md5"
-            ).hash_info.value
+                repo.odb.local,
+                output.path_info,
+                repo.odb.local.fs,
+                "md5",
+                dry_run=True,
+            )[1].hash_info.value
         return output.hash_info.value
 
-    for stage in repo.stages:
-        for output in stage.outs:
-            if _exists(output):
-                yield_output = targets is None or any(
-                    output.path_info.isin_or_eq(target) for target in targets
-                )
+    for output in repo.index.outs:
+        if _exists(output):
+            yield_output = targets is None or any(
+                output.path_info.isin_or_eq(target) for target in targets
+            )
 
-                if yield_output:
-                    yield _to_path(output), _to_checksum(output)
+            if yield_output:
+                yield _to_path(output), _to_checksum(output)
 
-                if output.is_dir_checksum and (
-                    yield_output
-                    or any(target.isin(output.path_info) for target in targets)
-                ):
-                    yield from _dir_output_paths(repo_fs, output, targets)
+            if output.is_dir_checksum and (
+                yield_output
+                or any(target.isin(output.path_info) for target in targets)
+            ):
+                yield from _dir_output_paths(repo_fs, output, targets)
 
 
 def _dir_output_paths(repo_fs, output, targets=None):

dvc/repo/experiments/__init__.py

@@ -81,6 +81,7 @@ class Experiments:
         r"(?P<checkpoint>-checkpoint)?$"
     )
     EXEC_TMP_DIR = "exps"
+    EXEC_PID_DIR = os.path.join(EXEC_TMP_DIR, "run")
 
     StashEntry = namedtuple(
         "StashEntry", ["index", "rev", "baseline_rev", "branch", "name"]
@@ -351,8 +352,7 @@ class Experiments:
 
     def _update_params(self, params: dict):
         """Update experiment params files with the specified values."""
-        from benedict import benedict
-
+        from dvc.utils.collections import merge_params
         from dvc.utils.serialize import MODIFIERS
 
         logger.debug("Using experiment params '%s'", params)
@@ -362,7 +362,7 @@ class Experiments:
             suffix = path.suffix.lower()
             modify_data = MODIFIERS[suffix]
             with modify_data(path, fs=self.repo.fs) as data:
-                benedict(data).merge(params[params_fname], overwrite=True)
+                merge_params(data, params[params_fname])
 
         # Force params file changes to be staged in git
         # Otherwise in certain situations the changes to params file may be
@@ -455,7 +455,7 @@ class Experiments:
             )
         logger.info(
             "\nTo promote an experiment to a Git branch run:\n\n"
-            "\tdvc exp branch <exp>\n"
+            "\tdvc exp branch <exp> <branch>\n"
         )
 
     @scm_locked
@@ -584,30 +584,25 @@ class Experiments:
 
         logger.debug(
             "Reproducing experiment revs '%s'",
-            ", ".join((rev[:7] for rev in to_run)),
+            ", ".join(rev[:7] for rev in to_run),
         )
 
         executors = self._init_executors(to_run)
-        exec_results = self._executors_repro(executors, **kwargs)
-
-        if keep_stash:
+        try:
+            exec_results = {}
+            exec_results.update(self._executors_repro(executors, **kwargs))
+        finally:
             # only drop successfully run stashed experiments
-            to_drop = sorted(
-                (
-                    stash_revs[rev][0]
-                    for rev in exec_results
-                    if rev in stash_revs
-                ),
-                reverse=True,
-            )
-        else:
-            # drop all stashed experiments
-            to_drop = sorted(
-                (stash_revs[rev][0] for rev in to_run if rev in stash_revs),
-                reverse=True,
-            )
-        for index in to_drop:
-            self.stash.drop(index)
+            to_drop = [
+                entry.index
+                for rev, entry in to_run.items()
+                if (
+                    entry.index is not None
+                    and (not keep_stash or rev in exec_results)
+                )
+            ]
+            for index in sorted(to_drop, reverse=True):
+                self.stash.drop(index)
 
         result: Dict[str, str] = {}
         for _, exp_result in exec_results.items():
@@ -623,6 +618,9 @@ class Experiments:
         base_tmp_dir = os.path.join(self.repo.tmp_dir, self.EXEC_TMP_DIR)
         if not os.path.exists(base_tmp_dir):
             makedirs(base_tmp_dir)
+        pid_dir = os.path.join(self.repo.tmp_dir, self.EXEC_PID_DIR)
+        if not os.path.exists(pid_dir):
+            makedirs(pid_dir)
         for stash_rev, item in to_run.items():
             self.scm.set_ref(EXEC_HEAD, item.rev)
             self.scm.set_ref(EXEC_MERGE, stash_rev)
@@ -649,9 +647,10 @@ class Experiments:
 
         return executors
 
+    @unlocked_repo
     def _executors_repro(
         self, executors: dict, jobs: Optional[int] = 1
-    ) -> Mapping[str, Mapping[str, str]]:
+    ) -> Dict[str, Dict[str, str]]:
         """Run dvc repro for the specified BaseExecutors in parallel.
 
         Returns:
@@ -667,6 +666,11 @@ class Experiments:
         with ProcessPoolExecutor(max_workers=jobs) as workers:
             futures = {}
             for rev, executor in executors.items():
+                pidfile = os.path.join(
+                    self.repo.tmp_dir,
+                    self.EXEC_PID_DIR,
+                    f"{rev}{executor.PIDFILE_EXT}",
+                )
                 future = workers.submit(
                     executor.reproduce,
                     executor.dvc_dir,
@@ -675,6 +679,8 @@ class Experiments:
                     name=executor.name,
                     rel_cwd=rel_cwd,
                     log_level=logger.getEffectiveLevel(),
+                    pidfile=pidfile,
+                    git_url=executor.git_url,
                 )
                 futures[future] = (rev, executor)
 
@@ -683,12 +689,15 @@ class Experiments:
             except KeyboardInterrupt:
                 # forward SIGINT to any running executor processes and
                 # cancel any remaining futures
+                workers.shutdown(wait=False)
                 pids = {}
-                while not pid_q.empty():
-                    rev, pid = pid_q.get()
-                    pids[rev] = pid
                 for future, (rev, _) in futures.items():
                     if future.running():
+                        # if future has already been started by the scheduler
+                        # we still have to wait until it tells us its PID
+                        while rev not in pids:
+                            rev, pid = pid_q.get()
+                            pids[rev] = pid
                         os.kill(pids[rev], signal.SIGINT)
                     elif not future.done():
                         future.cancel()
@@ -735,7 +744,10 @@ class Experiments:
             raise ExperimentExistsError(ref_info.name)
 
         for ref in executor.fetch_exps(
-            self.scm, force=exec_result.force, on_diverged=on_diverged
+            self.scm,
+            executor.git_url,
+            force=exec_result.force,
+            on_diverged=on_diverged,
         ):
             exp_rev = self.scm.get_ref(ref)
             if exp_rev:
@@ -747,13 +759,15 @@ class Experiments:
     @unlocked_repo
     def _workspace_repro(self) -> Mapping[str, str]:
         """Run the most recently stashed experiment in the workspace."""
+        from dvc.utils.fs import makedirs
+
         from .executor.base import BaseExecutor
 
         entry = first(self.stash_revs.values())
         assert entry.index == 0
 
         # NOTE: the stash commit to be popped already contains all the current
-        # workspace changes plus CLI modifed --params changes.
+        # workspace changes plus CLI modified --params changes.
         # `checkout --force` here will not lose any data (popping stash commit
         # will result in conflict between workspace params and stashed CLI
         # params, but we always want the stashed version).
@@ -766,12 +780,19 @@ class Experiments:
                 self.scm.remove_ref(EXEC_BRANCH)
             try:
                 orig_checkpoint = self.scm.get_ref(EXEC_CHECKPOINT)
+                pid_dir = os.path.join(self.repo.tmp_dir, self.EXEC_PID_DIR)
+                if not os.path.exists(pid_dir):
+                    makedirs(pid_dir)
+                pidfile = os.path.join(
+                    pid_dir, f"workspace{BaseExecutor.PIDFILE_EXT}"
+                )
                 exec_result = BaseExecutor.reproduce(
                     None,
                     rev,
                     name=entry.name,
                     rel_cwd=relpath(os.getcwd(), self.scm.root_dir),
                     log_errors=False,
+                    pidfile=pidfile,
                 )
 
                 if not exec_result.exp_hash:
@@ -858,9 +879,58 @@ class Experiments:
         exclude = f"{EXEC_NAMESPACE}/*"
         ref = self.scm.describe(rev, base=EXPS_NAMESPACE, exclude=exclude)
         if ref:
-            return ExpRefInfo.from_ref(ref).name
+            name = ExpRefInfo.from_ref(ref).name
+            if name:
+                return name
+        if rev in self.stash_revs:
+            return self.stash_revs[rev].name
         return None
 
+    def get_running_exps(self) -> Dict[str, int]:
+        """Return info for running experiments."""
+        from dvc.utils.serialize import load_yaml
+
+        from .executor.base import BaseExecutor, ExecutorInfo
+
+        result = {}
+        for pidfile in self.repo.fs.walk_files(
+            os.path.join(self.repo.tmp_dir, self.EXEC_PID_DIR)
+        ):
+            rev, _ = os.path.splitext(os.path.basename(pidfile))
+
+            try:
+                info = ExecutorInfo.from_dict(load_yaml(pidfile))
+                if rev == "workspace":
+                    # If we are appending to a checkpoint branch in a workspace
+                    # run, show the latest checkpoint as running.
+                    last_rev = self.scm.get_ref(EXEC_BRANCH)
+                    if last_rev:
+                        result[last_rev] = info.to_dict()
+                    else:
+                        result[rev] = info.to_dict()
+                else:
+                    result[rev] = info.to_dict()
+                    if info.git_url:
+
+                        def on_diverged(_ref: str, _checkpoint: bool):
+                            return False
+
+                        for ref in BaseExecutor.fetch_exps(
+                            self.scm,
+                            info.git_url,
+                            on_diverged=on_diverged,
+                        ):
+                            logger.debug(
+                                "Updated running experiment '%s'.", ref
+                            )
+                            last_rev = self.scm.get_ref(ref)
+                            result[rev]["last"] = last_rev
+                            if last_rev:
+                                result[last_rev] = info.to_dict()
+            except OSError:
+                pass
+        return result
+
     def apply(self, *args, **kwargs):
         from dvc.repo.experiments.apply import apply
 

dvc/repo/experiments/diff.py

@@ -35,8 +35,8 @@ def diff(repo, *args, a_rev=None, b_rev=None, param_deps=False, **kwargs):
 
     return {
         key: _diff(
-            format_dict(old[key]),
-            format_dict(new[key]),
+            format_dict(old.get("data", {}).get(key, {})),
+            format_dict(new.get("data", {}).get(key, {})),
             with_unchanged=with_unchanged,
         )
         for key in ["metrics", "params"]

dvc/repo/experiments/executor/base.py

@@ -3,6 +3,7 @@ import os
 import pickle
 from abc import ABC, abstractmethod
 from contextlib import contextmanager
+from dataclasses import dataclass
 from functools import partial
 from typing import (
     TYPE_CHECKING,
@@ -15,8 +16,10 @@ from typing import (
 
 from funcy import cached_property
 
+from dvc.env import DVC_EXP_AUTO_PUSH, DVC_EXP_GIT_REMOTE
 from dvc.exceptions import DvcException
 from dvc.path_info import PathInfo
+from dvc.repo import Repo
 from dvc.repo.experiments.base import (
     EXEC_BASELINE,
     EXEC_BRANCH,
@@ -35,7 +38,7 @@ from dvc.scm import SCM
 from dvc.stage import PipelineStage
 from dvc.stage.monitor import CheckpointKilledError
 from dvc.stage.serialize import to_lockfile
-from dvc.utils import dict_sha256
+from dvc.utils import dict_sha256, env2bool
 from dvc.utils.fs import remove
 
 if TYPE_CHECKING:
@@ -52,6 +55,36 @@ class ExecutorResult(NamedTuple):
     force: bool
 
 
+@dataclass
+class ExecutorInfo:
+    PARAM_PID = "pid"
+    PARAM_GIT_URL = "git"
+    PARAM_BASELINE_REV = "baseline"
+    PARAM_LOCATION = "location"
+
+    pid: Optional[int]
+    git_url: Optional[str]
+    baseline_rev: Optional[str]
+    location: Optional[str]
+
+    @classmethod
+    def from_dict(cls, d):
+        return cls(
+            d.get(cls.PARAM_PID),
+            d.get(cls.PARAM_GIT_URL),
+            d.get(cls.PARAM_BASELINE_REV),
+            d.get(cls.PARAM_LOCATION),
+        )
+
+    def to_dict(self):
+        return {
+            self.PARAM_PID: self.pid,
+            self.PARAM_GIT_URL: self.git_url,
+            self.PARAM_BASELINE_REV: self.baseline_rev,
+            self.PARAM_LOCATION: self.location,
+        }
+
+
 class BaseExecutor(ABC):
     """Base class for executing experiments in parallel.
 
@@ -66,6 +99,8 @@ class BaseExecutor(ABC):
     PACKED_ARGS_FILE = "repro.dat"
     WARN_UNTRACKED = False
     QUIET = False
+    PIDFILE_EXT = ".run"
+    DEFAULT_LOCATION: Optional[str] = "workspace"
 
     def __init__(
         self,
@@ -161,9 +196,11 @@ class BaseExecutor(ABC):
             data = pickle.load(fobj)
         return data["args"], data["kwargs"]
 
+    @classmethod
     def fetch_exps(
-        self,
+        cls,
         dest_scm: "Git",
+        url: str,
         force: bool = False,
         on_diverged: Callable[[str, bool], None] = None,
     ) -> Iterable[str]:
@@ -171,13 +208,17 @@ class BaseExecutor(ABC):
 
         Args:
             dest_scm: Destination Git instance.
+            url: Git remote URL to fetch from.
             force: If True, diverged refs will be overwritten
             on_diverged: Callback in the form on_diverged(ref, is_checkpoint)
                 to be called when an experiment ref has diverged.
         """
         refs = []
-        for ref in self.scm.iter_refs(base=EXPS_NAMESPACE):
-            if not ref.startswith(EXEC_NAMESPACE) and ref != EXPS_STASH:
+        has_checkpoint = False
+        for ref in dest_scm.iter_remote_refs(url, base=EXPS_NAMESPACE):
+            if ref == EXEC_CHECKPOINT:
+                has_checkpoint = True
+            elif not ref.startswith(EXEC_NAMESPACE) and ref != EXPS_STASH:
                 refs.append(ref)
 
         def on_diverged_ref(orig_ref: str, new_rev: str):
@@ -185,29 +226,43 @@ class BaseExecutor(ABC):
                 logger.debug("Replacing existing experiment '%s'", orig_ref)
                 return True
 
-            checkpoint = self.scm.get_ref(EXEC_CHECKPOINT) is not None
-            self._raise_ref_conflict(dest_scm, orig_ref, new_rev, checkpoint)
+            cls._raise_ref_conflict(
+                dest_scm, orig_ref, new_rev, has_checkpoint
+            )
             if on_diverged:
-                on_diverged(orig_ref, checkpoint)
+                on_diverged(orig_ref, has_checkpoint)
             logger.debug("Reproduced existing experiment '%s'", orig_ref)
             return False
 
         # fetch experiments
         dest_scm.fetch_refspecs(
-            self.git_url,
+            url,
             [f"{ref}:{ref}" for ref in refs],
             on_diverged=on_diverged_ref,
             force=force,
         )
         # update last run checkpoint (if it exists)
-        if self.scm.get_ref(EXEC_CHECKPOINT):
+        if has_checkpoint:
             dest_scm.fetch_refspecs(
-                self.git_url,
+                url,
                 [f"{EXEC_CHECKPOINT}:{EXEC_CHECKPOINT}"],
                 force=True,
             )
         return refs
 
+    @classmethod
+    def _validate_remotes(cls, dvc: "Repo", git_remote: Optional[str]):
+
+        if git_remote == dvc.root_dir:
+            logger.warning(
+                f"'{git_remote}' points to the current Git repo, experiment "
+                "Git refs will not be pushed. But DVC cache and run cache "
+                "will automatically be pushed to the default DVC remote "
+                "(if any) on each experiment commit."
+            )
+        dvc.scm.validate_git_remote(git_remote)
+        dvc.cloud.get_remote_odb()
+
     @classmethod
     def reproduce(
         cls,
@@ -218,6 +273,7 @@ class BaseExecutor(ABC):
         name: Optional[str] = None,
         log_errors: bool = True,
         log_level: Optional[int] = None,
+        **kwargs,
     ) -> "ExecutorResult":
         """Run dvc repro and return the result.
 
@@ -229,6 +285,9 @@ class BaseExecutor(ABC):
         from dvc.repo.checkout import checkout as dvc_checkout
         from dvc.repo.reproduce import reproduce as dvc_reproduce
 
+        auto_push = env2bool(DVC_EXP_AUTO_PUSH)
+        git_remote = os.getenv(DVC_EXP_GIT_REMOTE, None)
+
         unchanged = []
 
         if queue is not None:
@@ -245,7 +304,15 @@ class BaseExecutor(ABC):
         exp_ref: Optional["ExpRefInfo"] = None
         repro_force: bool = False
 
-        with cls._repro_dvc(dvc_dir, rel_cwd, log_errors) as dvc:
+        with cls._repro_dvc(
+            dvc_dir,
+            rel_cwd,
+            log_errors,
+            **kwargs,
+        ) as dvc:
+            if auto_push:
+                cls._validate_remotes(dvc, git_remote)
+
             args, kwargs = cls._repro_args(dvc)
             if args:
                 targets: Optional[Union[list, str]] = args[0]
@@ -285,6 +352,7 @@ class BaseExecutor(ABC):
 
             checkpoint_func = partial(
                 cls.checkpoint_callback,
+                dvc,
                 dvc.scm,
                 name,
                 repro_force or checkpoint_reset,
@@ -315,6 +383,8 @@ class BaseExecutor(ABC):
                         force=repro_force,
                         checkpoint=is_checkpoint,
                     )
+                    if auto_push:
+                        cls._auto_push(dvc, dvc.scm, git_remote)
                 except UnchangedExperimentError:
                     pass
                 ref = dvc.scm.get_ref(EXEC_BRANCH, follow=False)
@@ -339,9 +409,15 @@ class BaseExecutor(ABC):
     @classmethod
     @contextmanager
     def _repro_dvc(
-        cls, dvc_dir: Optional[str], rel_cwd: Optional[str], log_errors: bool
+        cls,
+        dvc_dir: Optional[str],
+        rel_cwd: Optional[str],
+        log_errors: bool,
+        pidfile: Optional[str] = None,
+        git_url: Optional[str] = None,
+        **kwargs,
     ):
-        from dvc.repo import Repo
+        from dvc.utils.serialize import modify_yaml
 
         dvc = Repo(dvc_dir)
         if cls.QUIET:
@@ -354,6 +430,15 @@ class BaseExecutor(ABC):
                 os.chdir(dvc.root_dir)
         else:
             old_cwd = None
+        if pidfile is not None:
+            info = ExecutorInfo(
+                os.getpid(),
+                git_url,
+                dvc.scm.get_rev(),
+                cls.DEFAULT_LOCATION,
+            )
+            with modify_yaml(pidfile) as d:
+                d.update(info.to_dict())
         logger.debug("Running repro in '%s'", os.getcwd())
 
         try:
@@ -369,6 +454,8 @@ class BaseExecutor(ABC):
                 logger.exception("unexpected error")
             raise
         finally:
+            if pidfile is not None:
+                remove(pidfile)
             dvc.close()
             if old_cwd:
                 os.chdir(old_cwd)
@@ -386,9 +473,32 @@ class BaseExecutor(ABC):
             kwargs = {}
         return args, kwargs
 
+    @staticmethod
+    def _auto_push(
+        dvc: "Repo",
+        scm: "Git",
+        git_remote: Optional[str],
+        push_cache=True,
+        run_cache=True,
+    ):
+        branch = scm.get_ref(EXEC_BRANCH, follow=False)
+        try:
+            dvc.experiments.push(
+                git_remote,
+                branch,
+                push_cache=push_cache,
+                run_cache=run_cache,
+            )
+        except BaseException as exc:
+            logger.warning(
+                "Something went wrong while auto pushing experiment "
+                f"to the remote '{git_remote}': {exc}"
+            )
+
     @classmethod
     def checkpoint_callback(
         cls,
+        dvc: "Repo",
         scm: "Git",
         name: Optional[str],
         force: bool,
@@ -400,6 +510,10 @@ class BaseExecutor(ABC):
             exp_rev = cls.commit(
                 scm, exp_hash, exp_name=name, force=force, checkpoint=True
             )
+
+            if env2bool(DVC_EXP_AUTO_PUSH):
+                git_remote = os.getenv(DVC_EXP_GIT_REMOTE)
+                cls._auto_push(dvc, scm, git_remote)
             logger.info("Checkpoint experiment iteration '%s'.", exp_rev[:7])
         except UnchangedExperimentError:
             pass

dvc/repo/experiments/executor/local.py

@@ -45,6 +45,7 @@ class TempDirExecutor(BaseLocalExecutor):
     # suggestions) that are not applicable outside of workspace runs
     WARN_UNTRACKED = True
     QUIET = True
+    DEFAULT_LOCATION: Optional[str] = "temp"
 
     def __init__(
         self,

dvc/repo/experiments/push.py

@@ -2,6 +2,7 @@ import logging
 
 from dvc.exceptions import DvcException, InvalidArgumentError
 from dvc.repo import locked
+from dvc.repo.experiments.base import ExpRefInfo
 from dvc.repo.scm_context import scm_context
 
 from .utils import exp_commits, exp_refs_by_name
@@ -12,7 +13,13 @@ logger = logging.getLogger(__name__)
 @locked
 @scm_context
 def push(
-    repo, git_remote, exp_name, *args, force=False, push_cache=False, **kwargs
+    repo,
+    git_remote,
+    exp_name: str,
+    *args,
+    force=False,
+    push_cache=False,
+    **kwargs,
 ):
     exp_ref = _get_exp_ref(repo, exp_name)
 
@@ -35,9 +42,9 @@ def push(
         _push_cache(repo, exp_ref, **kwargs)
 
 
-def _get_exp_ref(repo, exp_name):
+def _get_exp_ref(repo, exp_name: str) -> ExpRefInfo:
     if exp_name.startswith("refs/"):
-        return exp_name
+        return ExpRefInfo.from_ref(exp_name)
 
     exp_refs = list(exp_refs_by_name(repo.scm, exp_name))
     if not exp_refs:

dvc/repo/experiments/remove.py

@@ -1,8 +1,10 @@
 import logging
+from typing import List, Optional
 
 from dvc.exceptions import InvalidArgumentError
 from dvc.repo import locked
 from dvc.repo.scm_context import scm_context
+from dvc.scm.base import RevError
 
 from .base import EXPS_NAMESPACE, ExpRefInfo
 from .utils import exp_refs_by_name, remove_exp_refs
@@ -21,32 +23,43 @@ def remove(repo, exp_names=None, queue=False, **kwargs):
         removed += len(repo.experiments.stash)
         repo.experiments.stash.clear()
     if exp_names:
-        ref_infos = list(_get_exp_refs(repo, exp_names))
-        remove_exp_refs(repo.scm, ref_infos)
-        removed += len(ref_infos)
+        remained = _remove_commited_exps(repo, exp_names)
+        remained = _remove_queued_exps(repo, remained)
+        if remained:
+            raise InvalidArgumentError(
+                "'{}' is not a valid experiment".format(";".join(remained))
+            )
+        removed += len(exp_names) - len(remained)
     return removed
 
 
-def _get_exp_refs(repo, exp_names):
+def _get_exp_stash_index(repo, ref_or_rev: str) -> Optional[int]:
+    stash_revs = repo.experiments.stash_revs
+    for _, ref_info in stash_revs.items():
+        if ref_info.name == ref_or_rev:
+            return ref_info.index
+    try:
+        rev = repo.scm.resolve_rev(ref_or_rev)
+        if rev in stash_revs:
+            return stash_revs.get(rev).index
+    except RevError:
+        pass
+    return None
+
+
+def _get_exp_ref(repo, exp_name: str) -> Optional[ExpRefInfo]:
     cur_rev = repo.scm.get_rev()
-    for name in exp_names:
-        if name.startswith(EXPS_NAMESPACE):
-            if not repo.scm.get_ref(name):
-                raise InvalidArgumentError(
-                    f"'{name}' is not a valid experiment name"
-                )
-            yield ExpRefInfo.from_ref(name)
-        else:
-
-            exp_refs = list(exp_refs_by_name(repo.scm, name))
-            if not exp_refs:
-                raise InvalidArgumentError(
-                    f"'{name}' is not a valid experiment name"
-                )
-            yield _get_ref(exp_refs, name, cur_rev)
+    if exp_name.startswith(EXPS_NAMESPACE):
+        if repo.scm.get_ref(exp_name):
+            return ExpRefInfo.from_ref(exp_name)
+    else:
+        exp_refs = list(exp_refs_by_name(repo.scm, exp_name))
+        if exp_refs:
+            return _get_ref(exp_refs, exp_name, cur_rev)
+    return None
 
 
-def _get_ref(ref_infos, name, cur_rev):
+def _get_ref(ref_infos, name, cur_rev) -> Optional[ExpRefInfo]:
     if len(ref_infos) > 1:
         for info in ref_infos:
             if info.baseline_sha == cur_rev:
@@ -61,3 +74,28 @@ def _get_ref(ref_infos, name, cur_rev):
         msg.extend([f"\t{info}" for info in ref_infos])
         raise InvalidArgumentError("\n".join(msg))
     return ref_infos[0]
+
+
+def _remove_commited_exps(repo, refs: List[str]) -> List[str]:
+    remain_list = []
+    remove_list = []
+    for ref in refs:
+        ref_info = _get_exp_ref(repo, ref)
+        if ref_info:
+            remove_list.append(ref_info)
+        else:
+            remain_list.append(ref)
+    if remove_list:
+        remove_exp_refs(repo.scm, remove_list)
+    return remain_list
+
+
+def _remove_queued_exps(repo, refs_or_revs: List[str]) -> List[str]:
+    remain_list = []
+    for ref_or_rev in refs_or_revs:
+        stash_index = _get_exp_stash_index(repo, ref_or_rev)
+        if stash_index is None:
+            remain_list.append(ref_or_rev)
+        else:
+            repo.experiments.stash.drop(stash_index)
+    return remain_list