Ignore errors when attempting to restore messages if the message was acked

[soceanainn]

I am seeing a lot of errors from SQS in my logs coming from here (through the restore_unacked_once path) with the message:

Message does not exist or is not available for visibility timeout change.

My assumption is that while we are closing the channel, we call restore_unacked_once, but some task invocations are still completing while we are restoring unacked messages. This results in a race condition where messages may be acked after the '_flush' call earlier in this function, and the loop iteration where we attempt to restore that particular message.

If we encounter an exception when we attempt to restore a message but then find that the message has already been acked by another thread - then we should ignore the error.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.75%. Comparing base (0b293c7) to head (6fd1cc6).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2552   +/-   ##
=======================================
  Coverage   82.75%   82.75%           
=======================================
  Files          79       79           
  Lines       10261    10262    +1     
  Branches     1174     1175    +1     
=======================================
+ Hits         8491     8492    +1     
  Misses       1569     1569           
  Partials      201      201           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

[Copilot]

Pull request overview

This PR targets a shutdown/concurrency race in kombu.transport.virtual.base.QoS.restore_unacked() where a message can be acked by another thread after _flush() but before/during a restore attempt (notably surfacing as noisy SQS errors). The change aims to suppress restore errors when the message is already known to have been acked.

Changes:

• Track the delivery-tag key when popping from _delivered so restore failures can be correlated to _dirty (acked/rejected) tags.
• Suppress restore exceptions for messages whose delivery tag is now present in _dirty.
• Add a unit test asserting errors are ignored when restore raises and the message becomes acked/dirty.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
kombu/transport/virtual/base.py	Ignores restore exceptions when the delivery tag is marked dirty (acked) during restore_unacked processing.
t/unit/transport/virtual/test_base.py	Adds a regression test covering “restore raises but message is acked” behavior.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[auvipy]

is there any open or relevant issue this PR is handling?

[soceanainn]

is there any open or relevant issue this PR is handling?

I can't see any associated issues in Github for this - but it is a real issue I am experiencing in my production environment