Fix AsyncioConnection race conditions causing EBADF errors (#614)#697
Open
dkropachev wants to merge 1 commit intomasterfrom
Open
Fix AsyncioConnection race conditions causing EBADF errors (#614)#697dkropachev wants to merge 1 commit intomasterfrom
dkropachev wants to merge 1 commit intomasterfrom
Conversation
46524ac to
7b815c8
Compare
Lorak-mmk
requested changes
Feb 16, 2026
cassandra/io/asyncioreactor.py
Outdated
Comment on lines
181
to
184
| if self.is_closed or self.is_defunct: | ||
| raise ConnectionShutdown( | ||
| "Connection to %s is already closed" % (self.endpoint,)) | ||
|
|
There was a problem hiding this comment.
Why do we need to check that in push? push only puts data in the asyncio queue, it has nothing to do with actually putting that data in the socket, and thus no way to cause the error.
Collaborator
Author
There was a problem hiding this comment.
Idea was to narrow race down, removed.
cassandra/io/asyncioreactor.py
Outdated
Comment on lines
219
to
224
| # Peer disconnected — do a clean close instead of defunct | ||
| if isinstance(err, (BrokenPipeError, ConnectionResetError)): | ||
| log.debug("Connection %s closed by peer during write: %s", | ||
| self, err) | ||
| self.close() | ||
| return |
There was a problem hiding this comment.
Could you explain how is close "cleaner" that defunct, and what is the difference between them in general? I see that defunct does call close internally, so its not obvious to me.
Fix race conditions in AsyncioConnection that cause "[Errno 9] Bad file descriptor" errors during node restarts, especially with TLS: 1. close() now waits for _close() to complete when called from outside the event loop thread, eliminating the window where is_closed=True but the socket fd is still open. 2. handle_read() sets last_error on server EOF so factory() detects dead connections instead of returning them to callers. 3. handle_write() treats peer disconnections as clean close instead of defuncting, and both I/O handlers skip defunct() if the connection is already shutting down. Peer-disconnect detection is extracted into _is_peer_disconnect() helper covering platform-specific behaviors: - Windows: ProactorEventLoop raises plain OSError with winerror 10054 (WSAECONNRESET) or 10053 (WSAECONNABORTED) instead of ConnectionResetError. Detection uses ConnectionError base class plus winerror check. - macOS: Raises OSError(57) ENOTCONN when writing to a peer-disconnected socket, which is not a ConnectionError subclass. Detection uses errno-based checks for ENOTCONN, ESHUTDOWN, ECONNRESET, and ECONNABORTED. - Windows _close(): ProactorEventLoop does not support remove_reader/remove_writer (raises NotImplementedError). These calls are wrapped so the socket is always closed regardless, and try/finally ensures connected_event is always set even if cleanup fails.
2b9d555 to
85d08e1
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
close()to wait for_close()completion from non-event-loop threads, eliminating the window whereis_closed=Truebut the socket fd is still open (root cause of EBADF)last_erroron server EOF inhandle_read()sofactory()detects dead connections instead of returning themis_closed/is_defunctguard inpush()to reject writes on closed connectionsBrokenPipeError/ConnectionResetErrorinhandle_write()as clean peer disconnections instead of defuncting, and skipdefunct()in both I/O handlers if the connection is already shutting downTest plan
tests/unit/io/test_asyncio_race_614.pycover all four race conditions using real socket pairs