Possible Message Loss

[sschepens]

Expected behavior

Pulsar should never loose messages.

Actual behavior

It seems to me that now, when this condition returns false we have message loss.
Consumer receives a message and discards it, Broker considers it belongs to that Consumer, it will never redeliver it unless it disconnects.
Shouldn't Consumer ask for redelivery of the message instead of discarding it?

[merlimat]

I wouldn't characterize this as "message loss", since even if the particular message could be stuck, it would not be deleted until a consumer acknowledges it. So, reconnecting the consumer would unblock it.

I'm also not convinced that the condition can actually happen. Based on the code there would need to be these conditions :

1. Use receiveAsync()
2. Disable consumer prefetch queue
3. We receive a message but we don't have a callback waiting for it

I'm not sure how if this can happen given that:

1. If application used receiveAsync() then we would have a callback waiting
2. reciveAsync() request cannot be cancelled
3. When prefetch queue is 0, the consumer would have requested 1 single message from the broker, and timeout would not be allowed

To conclude, we could trigger the re-delivery of the ignored message, though it would be good to understand under which circumstances that condition can happen.

[ivankelly]

@sschepens has this happened again?

[ivankelly]

Closing due to lack of activity.