Repository navigation
Unregister the RMW subscription callback before finalizing the handle - #3370
Draft
Quchaosheng wants to merge 2 commits into
Draft
Quchaosheng wants to merge 2 commits into
Quchaosheng wants to merge 2 commits into
Conversation
EventsExecutor registers an on-new-message callback with the middleware and only unregisters it when it notices the entity is gone. The subscription handle could therefore be finalized while the callback was still reachable. Unregister the callback from ~SubscriptionBase(), before the shared rcl_subscription_t deleter runs rcl_subscription_fini(). Destruction must not throw, so failures are logged and the handle is still finalized. Refs ros2#2437 Signed-off-by: Chaosheng Qu <chaosheng.qu.linux@gmail.com>
Assert that rcl_subscription_set_on_new_message_callback() has been called with a null callback by the time rcl_subscription_fini() is entered. Signed-off-by: Chaosheng Qu <chaosheng.qu.linux@gmail.com>
Quchaosheng
force-pushed
the
fix/2437-subscription-callback-lifetime
branch
from
October 21, 2026 07:12
f981f6b to
e53f501
Compare
Quchaosheng
marked this pull request as draft
October 21, 2026 07:20
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description
~SubscriptionBase()never unregisters the middleware on-new-message callback. The callback'sstd::functionstorage is destroyed by the destructor, while Fast DDS can still be insideRMWSubscriptionEvent::update_data_available()on its own thread and call the trampoline with a pointer to that storage.This unregisters the callback at the start of the destructor, before the
rcl_subscription_tshared-pointer deleter runsrcl_subscription_fini(). Destruction must not throw, so a failure to unregister is logged and the handle is still finalized.Refs #2437
Status
Draft, not ready for merge.
I have not reproduced #2437 in a standalone test. My reduced two-node/
EventsExecutorstress harness only produced an unrelated stack-buffer-overflow inside the harness, and the corrected version did not crash. So this change addresses a destructor ordering gap, not a confirmed root cause of the original report.The new test asserts that
rcl_subscription_set_on_new_message_callback()is called with a null callback beforercl_subscription_fini()is entered. That is a narrower property than the one the issue actually needs, which is that teardown synchronizes with an in-flight middleware callback. Please treat this PR as a question about whether that unregister belongs in~SubscriptionBase()at all.I could not run the full
test_subscriptiontarget locally: the available Rolling dependency prefix predates this tree and lacksrcl_publisher_event_type_is_supported, and the installed Jazzy prefix is too old for the rest of the build. CI should provide the first full build.Is this user-facing behavior change?
No. Only the destruction path changes: the middleware callback is cleared earlier, and a teardown error is logged instead of being ignored.
Did you use Generative AI?
Yes. OpenAI Codex assisted with source analysis and test preparation.