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
As it turned out (see issue numbers down below), we can't really take
hardlinks for granted, so `flufl.lock` is not a panacea for all
filesystems. Considering that the vast majority of filesystems that our
users use support `zc.lockfile`(flock-based) and it has benefits like
more reliable mechanism, auto-delete when process exits, more sturdy
implementation, etc, it makes more sense to bring it back and use by
default again. For filesystems that don't support `flock()`, users will
be able to manually enable `flufl.lock` use through the config option.
It would be ideal if we could auto-detect that flock is not supported,
but in the real world, it turned out to be non-trivial, as it might hang
forever in a kernel context, which makes the implementation way too
complex for our purposes. So what we're doing instead is showing a
message before locking with `zc.lockfile` that, under normal
circumstances will disappear once the lock is taken or failed, otherwise
it will point users to the related documentation where they can learn
about how to opt-in for `flufl.lock`.
Fixes#2831Fixes#2897
Related #2860
Things changed:
- both no output and no git file result into PathMissingError now
- fail on missing cache or cache download error loudly
- removed sqlite bypass, since we use nolock now anyway
Refactoring:
- streamlined logic and made it similar to Repo.open() one
- moved exclusive exception classes to dvc.repo.get
Ruslan Kuprieiev
Casper da Costa-Luis
pawel
Alexander Schepanovski
Mayur Kularni
Anton Abrosimov
