Producer block may block forever

[jonyhy96]

Expected behavior

If producer.Close are called,it will close within configured OperationTimeout

Actual behavior

producer.Close may block forever and won't time out after configured OperationTimeout

Steps to reproduce

Just run the perf program for a while and send SIGINT or ctrl+c to it. This situation may be triggered.

This is because the callback of producer.SendAsyc contains a send behavior on a channel which under the same scope with stopCh. When we trigger producer.Close, if case <-stop: happens before case latency := <-ch:, it will block forever because it may be block at here and RequestOnCnx has no time out guarantee.

System configuration

Pulsar version: any

[jonyhy96]

/cc @freeznet @wolfstudy
I would like to fix this bug by add time out guarantee on RequestOnCnx
BTW, Can i join the slack of this project to get in touch with the mainainers?

[freeznet]

@jonyhy96 thanks, PR are always welcome ;)

Also, the slack is free to join:

The Slack instance is at https://apache-pulsar.slack.com/
You can self-register at https://apache-pulsar.herokuapp.com/

And you can join #dev-go channel to get in touch with go devs and users.

[jonyhy96]

https://apache-pulsar.slack.com/ shows me hy352144278@gmail.com doesn’t have an account on this workspace. and https://apache-pulsar.herokuapp.com/ seems doesn't have #dev-go channel 😅

[freeznet]

@jonyhy96 You can browse all available channels from the plus icon next to Channels dropdown item on your Slack. Also I have sent an invitation to your email hy352144278@gmail.com as well.