Closed vmarkovtsev closed 5 years ago
Can you provide context on what exactly you want us to do here? Which are the current problems that should be solved? I assume you're talking about the deployment in the staging cluster via this helm chart through travis
The major one is with the usage of sleep
to ensure the startup sequence which I successfully fixed with port/postgres polling.
Do we have an example of a failed deploy?
My talk on MLonCode meetup in front of tens of people online, around 50th minute
My talk on MLonCode meetup in front of tens of people online, around 50th minute
Any available logs?
@rporres Here https://youtu.be/ixxYwcLdkfI?t=3071
The analyzer took more than 5 seconds to start due to Zoom CPU overload and the whole thing broke.
@vmarkovtsev if we understand it correctly you want the Helm chart to work the same as the docker compose file? However we have been unable to find the compose file.
This looks like a docker-compose issue rather than a kubernetes one. If the database is not ready for any reason and style analyzer crashes retries will be handled by kubernetes.
There have been recently downtimes in staging style analyzer, but they were related to changes in the args parsing. See https://github.com/src-d/charts/pull/160
Feel free to open this if you see any issue in kubernetes deployment via Helm
There is a better docker compose template in https://github.com/src-d/lookout-sdk-ml/tree/master/lookout/core/container - we should copy swagger from it and the files nearby.