Flusher task is stopping due to InvalidStateError

[debbyglance]

Observed behavior

See #606
The flusher() task is stopping because InvalidStateError is raised when attempting to call .set_result() on the future

Expected behavior

The flusher task should never stop

Server and client version

client: nats.py 2.6.0
server: 2.10.7

Host environment

NATS server is running on docker image nats:2.10.7-alpine3.18
The client is a python quart application running behing hypercorn.

Steps to reproduce

I don't have steps to reproduce, but this happens sporadically in production, and I think could be easy to fix. This is what I think is happening.

1. During a reconnect attempt, _attempt_reconnect() calls _flush_pending()
2. The _flush_pending() task creates a Future and adds it to the _flush_queue
3. _flush_pending waits on the future
4. The reconnect fails and _attempt_reconnect() is cancelled.
5. This cancels the _flush_pending task which cancels the Future that was created in step 2 (python cancels a future that is being awaited by a task when the task is cancelled).
6. Now there is a Future in the _flush_queue that is in cancelled state
7. The new reconnect attempt starts a new _flusher() task
8. The _flusher() task pulls the cancelled Future out of the queue
9. When the _flusher() task calls set_result() on the future, it results in an InvalidStateError exception because the future is already done
10. The flusher task aborts

Possible fixes would be:

• _flusher() could ignore any cancelled futures in the _flush_queue
• clear the _flush_queue on reconnect (or create a new queue)

[shoriminimoe]

Hi NATS team! 👋

Is there a plan to merge #636? My project is affected by this issue. 😦

[swelborn]

looks like this was merged @shoriminimoe.