Currently only one instance of dvc could run in a particular repo
because of the `.dvc/lock`. That gets furstrating when you are running
some long-running command and blocked from doing anything else.
This PR solved that issue for `dvc run` by releasing the repo lock
while running the command. To protect against conflicts, we now use
rwlock, which keeps the list of readers and writers for particular
paths. The process looks like this:
1) aquire read locks for deps;
2) aquire write locks for outs;
3) release repo lock;
4) run cmd;
5) aquire repo lock;
6) release write locks for outs;
7) release read locks for outs;
We also take rwlocks for operations such as `import`, `add` and `checkout`
to ensure that we won't hit the same files that are being used by
another `dvc run` in the background.
We don't aquire rwlocks for the stage file itself, as we don't really
care about anyone overwriting a stage file, as there is a very little
difference here between parallel runs and sequential runs that overwrite
the same stage file.
Next steps might be to release repo lock on `import`(when dowloading
stuff) and when computing checksums or copying files.
Fixes#755
Imagine the scenario of shared cache. To set it up, user would do:
```
dvc config cache.dir /path
dvc config cache.shared group
```
Now, when he runs any dvc operation that involves the Repo instance,
RemoteLOCAL will also be initialized and it will currently auto-create
the directory with a regular `os.mkdir`, which is not using group
permissions. To solve that, we can't simply use `self.makdirs` instead,
as it will cause same issues if you run any dvc command before
`dvc config cache.shared group`. So what we should do is create
directories on demand, instead of creating it in advance, same as we do
in other remotes.
The change itself is pretty trivial and involved removing that `mkdir`
line, adjusting tests to not rely on cache dir existence and also
adjusting `dvc version` to not rely on cache directory existance.
Fixes#2563
Current test is testing for signal handler restoration in an obscure
way. As a result, the test wasn't actually testing anything, because
KeyboardInterrupt is raised by default_int_handler, and so raising it
in a mock won't be caught by signal handlers, because they are dealing
with real singnals.
Fix#2406
Signed-off-by: Ruslan Kuprieiev <[email protected]>
Replaced Path* dataclasses with PathInfo/URLInfo/CloudURLInfo, which
know how parse, serialize, stringify and join themselves. They also
provide useful accessors like part of an url or pathlib.Path like
.parent or .name.
Along the way two families of bugs where fixed and should never occur in
the future:
- paths not matched as strings anymore, e.g. using str.startswith()
- Output.path is eliminated as ambiguous
Both families might had caused undefined surprising behavior when local
and remote paths clash.
* test: clean up test dirs
Closes#1770. Partially does #1888.
* dvc: close anything opened
* test: finalize git repo explicitly to remove test dir later
This was previously an issue on Windows.
* test: remove never used copy of load_stage_file()
* dvc: save comments and custom formatting in stage files
* Add a couple of TODOs
* test: test repro doesn't clear comments and formatting in .dvc
As part of this I started writing pytest-like tests and added commonly
used fixtures.
* test: simplify and prettify dvc formatting retained test
Also test that line comments after md5 fields are retained.
* test: fix tests
* dvc: fix stage and metrics opens not closed
* dvc: always use utf-8 in stage files
It used to encode in system default encoding. Not sure if PyYAML was
handling this silently or we were just lucky all the way.
* test: unit test apply_diff()
* test: remove TODOs and ASK
* dvc: polish dvc-formatting PR
* dvc: move stage load/dump utils to a new stage utils file
* dvc: clean up apply_diff()
* test: test stage md5 ignores comments
* test: remove pytests starting marker
Paweł Redzyński
Ruslan Kuprieiev
Mr. Outis
Ruslan Kuprieiev
Alexander Schepanovski
Kirill Vasin