We don't generally log client errors to reduce noise and have the general policy of one channel per service in logstash, so this would be quite an undertaking.

If you don't feel strongly about this, I'm inclined to decline the task.

I don't feel strongly about it since reading lists just proxy the requests so it's not any harder to do the logging on the action API side. In general though I think providing information about client errors will be important when there is an API which is complex enough that clients can easily get it wrong (ie. not read-only) and the API owner is working closely with another team doing client development.

As a middle ground, we can log them with DEBUG level and we have a feature to probabilistically elevate the logged level per channel, so we can start logging 5% of them to logstash - that will allow having some idea about most common client errors and avoid unnecessary noise in the logs. How does that sound?

@Pchelolo sounds like a good compromise to me.

PR here https://github.com/wikimedia/restbase/pull/990

Change 430087 had a related patch set uploaded (by Ppchelko; owner: Ppchelko):
[mediawiki/services/restbase/deploy@master] Add 1% sampling for no-index and action api 400 error logs.

https://gerrit.wikimedia.org/r/430087

Change 430087 merged by Mobrovac:
[mediawiki/services/restbase/deploy@master] Add 1% sampling for no-index and action api 400 error logs.

https://gerrit.wikimedia.org/r/430087