Skip to content

concurrent.futures.ProcessPoolExecutor can deadlock when tcmalloc is used #90622

Closed
Closed
concurrent.futures.ProcessPoolExecutor can deadlock when tcmalloc is used#90622
#92497 (+3)
Assignees
gpshead
Labels
3.10 (EOL)end of life3.11only security fixes3.9 (EOL)end of lifeperformancePerformance or resource usagestdlibStandard Library Python modules in the Lib/ directory

Description

@yilei
yilei
mannequin
opened on Jan 21, 2022
BPO 46464
Nosy @gpshead, @pitrou, @methane, @yilei
PRs
  • gh-90622: Prevent concurrent.futures.process deadlock due to fork() after spawning a thread. #30847
    		•

    Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.

    Show more details

    GitHub fields:

    assignee = 'https://github.com/gpshead'
closed_at = None
created_at = <Date 2022-01-21.23:32:53.148>
labels = ['3.11', 'library', '3.9', '3.10', 'performance']
title = 'concurrent.futures.ProcessPoolExecutor can deadlock when tcmalloc is used'
updated_at = <Date 2022-01-25.11:06:46.406>
user = 'https://github.com/yilei'

    bugs.python.org fields:

    activity = <Date 2022-01-25.11:06:46.406>
actor = 'pitrou'
assignee = 'gregory.p.smith'
closed = False
closed_date = None
closer = None
components = ['Library (Lib)']
creation = <Date 2022-01-21.23:32:53.148>
creator = 'yilei'
dependencies = []
files = []
hgrepos = []
issue_num = 46464
keywords = ['patch', '3.9regression']
message_count = 7.0
messages = ['411208', '411213', '411451', '411543', '411544', '411545', '411574']
nosy_count = 4.0
nosy_names = ['gregory.p.smith', 'pitrou', 'methane', 'yilei']
pr_nums = ['30847']
priority = 'normal'
resolution = None
stage = 'patch review'
status = 'open'
superseder = None
type = 'resource usage'
url = 'https://bugs.python.org/issue46464'
versions = ['Python 3.9', 'Python 3.10', 'Python 3.11']

    Activity

    1. yilei commented on Jan 21, 2022

      @yilei
      yileimannequin
      on Jan 21, 2022
      MannequinAuthor

      When Python is built and linked with tcmalloc, using ProcessPoolExecutor may deadlock. Here is a reproducible example:

      $ cat t.py
from concurrent import futures
import sys
def work(iteration, item):
  sys.stdout.write(f'working: iteration={iteration}, item={item}\n')
  sys.stdout.flush()
for i in range(0, 10000):
    with futures.ProcessPoolExecutor(max_workers=2) as executor:
        executor.submit(work, i, 1)
        executor.submit(work, i, 2)

$ python t.py
working: iteration=0, item=1
working: iteration=0, item=2
working: iteration=1, item=1
working: iteration=1, item=2
...
working: iteration=3631, item=1
working: iteration=3631, item=2
<hang here>

      The child process fails to finish. It's more likely to reproduce when the system is busy.

      With some bisect search internally, this commit 1ac6e37 "triggered" the deadlock threshold with tcmalloc.

    2. yilei
      added
      3.9 (EOL)end of life
      3.10 (EOL)end of life
      3.11only security fixes
      stdlibStandard Library Python modules in the Lib/ directory
      on Jan 21, 2022

    3. gpshead commented on Jan 21, 2022

      @gpshead
      gpshead
      on Jan 21, 2022
      Member

      For context, the fundamental problem is that os.fork() is unsafe to use in any process with threads. concurrent/futures/process.py violates this. So long as its design involves a thread, it can never be guaranteed not to deadlock.

      POSIX forbids execution of most code after fork() has been called. Returning to the CPython interpreter after os.fork() in the child process is never safe unless the parent process had no threads at fork time.

      The change that triggered the issue moved from doing all of the os.fork() calls for the workers up front *before the thread was spawned* to doing them on demand after the thread was running.

      The bugfix for this is simply a rollback to revert the bpo-39207 change. It was a performance enhancement, but it causes a new deadlock in code that didn't already have one when thread tuned malloc implementations are used.

      The only way to safely launch worker processes on demand is to spawn a worker launcher process spawned prior to any thread creation that remains idle, with a sole job of spawn new worker processes for us. That sounds complicated. That'd be a feature. Lets go with the bugfix first.

    4. gpshead
      self-assigned this
      on Jan 21, 2022
    5. gpshead
      added
      performancePerformance or resource usage
      on Jan 21, 2022
    6. gpshead
      self-assigned this
      on Jan 21, 2022
    7. gpshead
      added
      performancePerformance or resource usage
      on Jan 21, 2022

    8. gpshead commented on Jan 24, 2022

      @gpshead
      gpshead
      on Jan 24, 2022
      Member

      https://bugs.python.org/issue44733 for 3.11 attempts to build upon the dynamic spawning ability and will need reverting unless a non-thread+fork implementation is provided.

    9. 12 remaining items

    10. gpshead
      closed this as completedin #91598on May 8, 2022
    11. miss-islington
      added a commit that references this issue on May 8, 2022

      pythongh-90622: Do not spawn ProcessPool workers on demand via fork m…

      6f08a59
    12. gpshead
      added 3 commits that reference this issue on May 8, 2022

      gh-90622: Do not spawn ProcessPool workers on demand via fork method. (…

      ebb37fc

      [3.10] pythongh-90622: Do not spawn ProcessPool workers on demand via…

      5595472

      [3.10] gh-90622: Do not spawn ProcessPool workers on demand via fork …

      b795376
    13. miss-islington
      added a commit that references this issue on May 8, 2022

      [3.10] pythongh-90622: Do not spawn ProcessPool workers on demand via…

      ecfa914
    14. gpshead
      added a commit that references this issue on May 8, 2022

      gh-90622: Do not spawn ProcessPool workers on demand via fork method. (…

      4270b79
    15. gpshead
      linked a pull request that will close this issue[3.9] [3.10] gh-90622: Do not spawn ProcessPool workers on demand via fork method. (GH-91598) (GH-92497) #92499on May 8, 2022
    16. gpshead
      reopened this on May 8, 2022
    17. gpshead
      added a commit that references this issue on May 8, 2022

      [3.10] gh-90622: Do not spawn ProcessPool workers on demand via fork …

      8e523c1
    18. gpshead
      closed this as completedon May 8, 2022
    19. gpshead
      linked a pull request that will close this issue[3.11] gh-90622: Do not spawn ProcessPool workers on demand via fork method. (GH-91598) #92495on May 8, 2022
    20. gpshead
      linked a pull request that will close this issue[3.10] gh-90622: Do not spawn ProcessPool workers on demand via the "fork" spawn method (GH-91598) #92497on May 8, 2022

    21. gpshead commented on May 8, 2022

      @gpshead
      gpshead
      on May 8, 2022
      Member

      Note that there is still one deadlock possibility that I left as a comment in the code. If the "fork" mp_context spawn method is used and one of the worker processes dies, a new one will be forked after multiprocessing has already started its executor manager thread. mp_context "forkserver" and "spawn" methods avoid that.

    22. hello-adam
      added a commit that references this issue on Jun 2, 2022

      [3.10] pythongh-90622: Do not spawn ProcessPool workers on demand via…

      78e11e4
    Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

    Metadata

    Metadata

    Assignees

    Labels

    3.10 (EOL)end of life3.11only security fixes3.9 (EOL)end of lifeperformancePerformance or resource usagestdlibStandard Library Python modules in the Lib/ directory

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions