You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Consumer just hangs as if everything is ok, except no new messages are received. No exceptions reach the caller code on the consumer side.
If the server comes back online:
docker start broadcaster_redis_test
The subscriber code stays in the same state regardless.
Even if we restart the publisher code, the subscriber side never does anything else, unable to receive new messages but still not raising any exceptions.
Expected behaviour
I'd expect both sides of the communication to receive an exception when a connection error happens. Or any other error raised by the Redis library not explicitly handled by the backend.
That way if the library can't handle the network issue (trying to reconnect, etc) at least the caller code could try to do that, or handle it in some way depending on business logic.
Comments
I think the issue is caused by two similar failing points:
The tasks created with asyncio.create_task(...) are never observed in case they stop or fail.
Using add_done_callback in both places I was able to pass the exception from the backend to the Broadcaster class, and then re-raise it from there.
I've got half a fix ready, but it's tricky to test it.
If everything above makes sense I can try to get it ready for a PR.
When I subscribe to events using the Redis backend, any errors raised by the Redis library seem to be swallowed by the subscriber side of
broadcaster
.Steps to reproduce
Launch a Redis server (for example using Docker):
Launch a message consumer:
Launch a message publisher:
Stop the Redis server:
Observed behaviour
Publisher stops with a
ConnectionError
:Consumer just hangs as if everything is ok, except no new messages are received. No exceptions reach the caller code on the consumer side.
If the server comes back online:
The subscriber code stays in the same state regardless.
Even if we restart the publisher code, the subscriber side never does anything else, unable to receive new messages but still not raising any exceptions.
Expected behaviour
I'd expect both sides of the communication to receive an exception when a connection error happens. Or any other error raised by the Redis library not explicitly handled by the backend.
That way if the library can't handle the network issue (trying to reconnect, etc) at least the caller code could try to do that, or handle it in some way depending on business logic.
Comments
I think the issue is caused by two similar failing points:
broadcaster/broadcaster/_base.py
Line 71 in 69cf29a
broadcaster/broadcaster/_backends/redis.py
Line 21 in 69cf29a
The tasks created with
asyncio.create_task(...)
are never observed in case they stop or fail.Using
add_done_callback
in both places I was able to pass the exception from the backend to the Broadcaster class, and then re-raise it from there.I've got half a fix ready, but it's tricky to test it.
If everything above makes sense I can try to get it ready for a PR.
Versions used
The text was updated successfully, but these errors were encountered: