Frequent bot restart required

[jrrbelmont]

We've been running the bot for over a year here and every week or so the bot will stop responding to requests (and we see it go offline in slack). The node process is still running and there isn't anything interesting in the logs. We need to kill the current process and restart it.

Bot 0.0.13
Ubuntu 14.04

Known issue with node / slack / looker?

[wilg]

Hmm... I haven't heard of this. The bot is supposed to exit if the slack connection terminates, so I don't know why the process would stay running. Are you using slash commands or @ messages to communicate with the bot?

[jrrbelmont]

Thanks for the quick response. No slash commands. Only @ mentions or, more often, direct messages in the Looker user channel.

[wilg]

Definitely sounds like the realtime API is shutting down. It should be exiting in that case though. Looks like this needs some investigation.

[jrrbelmont]

What can I provide for debugging? Thanks.

[jtao27]

+1 a customer needing restarts

[tomdev]

Yeah, running into this a lot as well. Needed to restart lookerbot several times last week.

Please let me know if there is anything I can do to help debug. I'm not sure if there are any log files I can dig into?

[jrrbelmont]

I'd love to help debug this. This is very annoying and too frequent. What logging can I add to see if this is a problem talking to Slack or Looker?

[wilg]

I am pretty sure it’s a problem talking to Slack but unfortunately I just am not sure how to debug it off the top of my head and I haven’t had the bandwidth to do a deep dive on this yet.

[wilg]

Please reopen if this still happens under 0.0.14.

[wilg]

I'm seeing this internally – the bot continues responding to Slash commands but goes offline. Slack RTM is disconnecting somehow.

[hporter]

Experienced this today - bot was not responding to commands, offline in Slack. Restarting the docker container in Kubernetes fixed it instantly. Running latest version, deployed for the first time on Thursday.

We will try to debug and share anything we find, as everyone is pretty excited to see this working!

Even if it is tough to work out why this happening, a /status route within the app would at least allow Kube to restart the container automatically when it happens - assuming the problem could be detected from within Lookerbot. Some more logging to trawl through might help as well.

[wilg]

There is /health_check, but I think the issue is the process stays up but only the real-time websocket closes or stops listening. The bot should restart itself if it closes though. So I'm not sure what's going on.

[DylanBaker]

+1. We are seeing this a few times a week. Running in Elastic Beanstalk. Restart fixes every time.