Chatie API Server Down Accident Report

[su-chang]

Token Service Discovery Service Accident

Our wechaty puppet service discovery service has been experiencing out-of-service issues from 3 pm Feb 7.

1. 10 am Feb 7: notice the disk usage of some instances are abnormal, then clear logs file and make instance keep running right, at the same time the api.chatie.io work well
2. 3 pm Feb 7: this problem outbreak in the afternoon then we working on it, and found that the http response status code 503 of api.chatie.io
3. 2 am Feb 8: @huan show some detail info from heroku, see: 🔥🔥🔥 api.chatie.io服务异常，HTTP错误码503 #97 (comment)
4. 8 am Feb 8: confirm api.chatie.io out-of-service due receive too many requests (init token on api.chatie.io) in few seconds
5. 9 am Feb 8: find the bug in wechaty-puppet-workpro, one NodeJS Timer function init token on api.chatie.io has not been clear right, and we notice that the only way which could fix this bug temporarily is to restart all containers
6. 10 am Feb 8: confirm the operation time to restart all containers
7. 2 pm Feb 8: restart all containers
8. 2:30 pm Feb8`: the server fully restored
9. 6 pm Feb 8: create the hotfix PR to fix this problem
10. 9 pm Feb 8: PR has been merged, and ready to deploy
11. 0 pm Feb 9: start deploy for some instances

[su-chang]

TODO

We will continue to deploy the fixed version to rest instances before Feb 15

[huan]

Could you explain point 5: why wechaty-puppet-workpro needs to init token on api.chatie.io?

If I remember correctly, the wechaty service discovery is managed by Wechaty itself?

[su-chang]

wechaty-puppet-workpro is based on wechaty-grpc, but not wechaty-puppet-service.

And the logic about init token on api.chatie.io is maintained by workpro.

[huan]

I think this is a bad idea, but I hope it will work well in the future.

The protocol might be changed someday so please be prepared to follow new protocols.

[su-chang]

Thanks for your advice, we will pay attention to it.