dvc: exp: Too many openfiles error in `exp` related commands.

Bug Report

Sometimes on MacOS, we will meet the “too many open files error”. This can happen in exp run, exp show or inside celery work.
And the time it happening is also varied, setup, running, or result collection, I met this error in all stages. ulimit to a larger number ( for example 1024 instead of default 256 ) can prevent this.

Traceback (most recent call last):
  File "/Users/gao/Code/dvc/dvc/repo/experiments/queue/tasks.py", line 66, in collect_exp
    BaseStashQueue.collect_executor(
  File "/Users/gao/Code/dvc/dvc/repo/experiments/queue/base.py", line 643, in collect_executor
    results = cls.collect_git(exp, executor, exec_result)
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/funcy/decorators.py", line 45, in wrapper
    return deco(call, *dargs, **dkwargs)
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/funcy/flow.py", line 127, in retry
    return call()
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/funcy/decorators.py", line 66, in __call__
    return self._func(*self._args, **self._kwargs)
  File "/Users/gao/Code/dvc/dvc/repo/experiments/utils.py", line 40, in wrapper
    return f(exp, *args, **kwargs)
  File "/Users/gao/Code/dvc/dvc/repo/experiments/queue/base.py", line 623, in collect_git
    for ref in executor.fetch_exps(
  File "/Users/gao/Code/dvc/dvc/repo/experiments/executor/base.py", line 367, in fetch_exps
    dest_scm.fetch_refspecs(
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/scmrepo/git/__init__.py", line 289, in _backend_func
    result = func(*args, **kwargs)
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/scmrepo/git/backend/dulwich/__init__.py", line 634, in fetch_refspecs
    fetch_result = client.fetch(
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/dulwich/client.py", line 1502, in fetch
    refs = r.fetch(
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/dulwich/repo.py", line 427, in fetch
    count, pack_data = self.fetch_pack_data(
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/dulwich/repo.py", line 460, in fetch_pack_data
    objects = self.fetch_objects(
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/dulwich/repo.py", line 494, in fetch_objects
    obj = self.object_store[sha]
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/dulwich/object_store.py", line 144, in __getitem__
    type_num, uncomp = self.get_raw(sha)
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/dulwich/object_store.py", line 581, in get_raw
    ret = self._get_loose_object(hexsha)
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/dulwich/object_store.py", line 745, in _get_loose_object
    return ShaFile.from_path(path)
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/dulwich/objects.py", line 420, in from_path
    with GitFile(path, "rb") as f:
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/dulwich/file.py", line 94, in GitFile
    return io.open(filename, mode, bufsize)
OSError: [Errno 24] Too many open files: '/Users/gao/test/vscode-dvc/demo/.dvc/tmp/exps/tmp9x5praul/.git/objects/55/4aa474348369ca4eec2945226686b9e4f11666'
[2022-10-25 16:25:09,522: ERROR/MainProcess] Task dvc.repo.experiments.queue.tasks.run_exp[cf2fd3f4-a0c2-4c42-80b9-9554e03d9a55] raised unexpected: OSError(24, 'Too many open files')
Traceback (most recent call last):
  File "/opt/homebrew/Cellar/python@3.10/3.10.6_1/Frameworks/Python.framework/Versions/3.10/lib/python3.10/shutil.py", line 628, in _rmtree_safe_fd
    with os.scandir(topfd) as scandir_it:
OSError: [Errno 24] Too many open files: '/Users/gao/test/vscode-dvc/demo/.dvc/tmp/exps/tmp9x5praul/demo'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/celery/app/trace.py", line 451, in trace_task
    R = retval = fun(*args, **kwargs)
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/celery/app/trace.py", line 734, in __protected_call__
    return self.run(*args, **kwargs)
  File "/Users/gao/Code/dvc/dvc/repo/experiments/queue/tasks.py", line 111, in run_exp
    cleanup_exp.s(executor, infofile)()
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/celery/canvas.py", line 168, in __call__
    return self.type(*args, **kwargs)
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/celery/app/trace.py", line 735, in __protected_call__
    return orig(self, *args, **kwargs)
  File "/Users/gao/test/vscode-dvc/demo/.venv/lib/python3.10/site-packages/celery/app/task.py", line 392, in __call__
    return self.run(*args, **kwargs)
  File "/Users/gao/Code/dvc/dvc/repo/experiments/queue/tasks.py", line 87, in cleanup_exp
    executor.cleanup(infofile)
  File "/Users/gao/Code/dvc/dvc/repo/experiments/executor/local.py", line 128, in cleanup
    remove(self.root_dir)
  File "/Users/gao/Code/dvc/dvc/utils/fs.py", line 69, in remove
    shutil.rmtree(path, onerror=_chmod)
  File "/opt/homebrew/Cellar/python@3.10/3.10.6_1/Frameworks/Python.framework/Versions/3.10/lib/python3.10/shutil.py", line 724, in rmtree
    _rmtree_safe_fd(fd, path, onerror)
  File "/opt/homebrew/Cellar/python@3.10/3.10.6_1/Frameworks/Python.framework/Versions/3.10/lib/python3.10/shutil.py", line 657, in _rmtree_safe_fd
    _rmtree_safe_fd(dirfd, fullname, onerror)
  File "/opt/homebrew/Cellar/python@3.10/3.10.6_1/Frameworks/Python.framework/Versions/3.10/lib/python3.10/shutil.py", line 632, in _rmtree_safe_fd
    onerror(os.scandir, path, sys.exc_info())
  File "/Users/gao/Code/dvc/dvc/utils/fs.py", line 54, in _chmod
    func(p)
OSError: [Errno 24] Too many open files: '/Users/gao/test/vscode-dvc/demo/.dvc/tmp/exps/tmp9x5praul/demo'

Description

Reproduce

Expected

Environment information

Output of dvc doctor:

$ dvc doctor

Additional Information (if any):

About this issue

  • Original URL
  • State: closed
  • Created 2 years ago
  • Comments: 18 (18 by maintainers)

Most upvoted comments

@karajan1001 I’m curious if anyone else on the vscode team has run into this same problem? Also, could you run git count-objects -vH in your clone of the vscode repo? (In your main clone of the repo, not in a temp exp workspace)

What I am wondering is that if you aren’t making regular changes in the vscode repo (and are only ever using dvc exp ... for testing purposes), you also aren’t using CLI git in that clone regularly, so it may not be gc’d at all, which could also lead to dulwich hitting the file limit when we try to collect exps. There could be too many loose objects on the receiving end of the fetch (in the main repo) and not necessarily too many objects in the source end (in an exp temp workspace). A typical user would not hit this scenario, because they would presumably be using regular CLI git themselves in conjunction with DVC on a regular enough basis that git would perform a gc at some point.

+2
pmrowla on Oct 28, 2022

I think we can close this for now given that it is unlikely this will occur in normal DVC use (where users are at least semi-regularly also using CLI git commands)

+1
pmrowla on Nov 1, 2022

Yeah, so your repo hasn’t been garbage collected. If you just manually run git gc in that repo it may resolve the too many open files issue.

Yes, I never saw this error after I gced the repo.

+1
karajan1001 on Nov 1, 2022
Read more comments on GitHub
← dvc: `exp run`: `--set-params` forces YAML 1.2 (breaks code that uses 1.1)
dvc: Failing to install due to the lack of GCC, required for psutil →