Webhook and sensor log to stderr

[l15k4]

Describe the bug
Webhook and sensor log to stderr which polutes stackdriver with errors.

Screenshots

Environment (please complete the following information):

• Kubernetes: [e.g. v1.16]
• Argo: [e.g. v2.10]
• Argo Events: [e.g. v1.0.0]

Reason
This is probably related to logrus package. Its default output stream is stderr.

Solution
Good precaution could be using severity instead of level in the json object, it would at least make this common problem disappear on stackdriver.

[whynowy]

It should addressed by making zap log configurable with a configmap.

[l15k4]

The Argo workflow sidecar container wait also logs everything to stderr ... it is really hard to setup monitoring for anything running on Argo right now.

[whynowy]

Curious about how you concluded it logs to stderr? I thought it was just the keyword level, however it seems not. Can you explain more?

[l15k4]

We are using stackdriver where you can see what goes to stdout and stderr

[whynowy]

I see. This is an easy fix. By default zap log use stderr as default output, we can change it to stdout. Question about the keyword level, is it reserved in stackdriver? or what is the problem?

[l15k4]

I meant, if the keyword was severity instead of level, then stackdriver would use it to determine the severity ... ie. even if it was put on stderr and there was severity=info then stackdriver would classify it as info ... otherwise stackdriver decides by type of output stderr => ERROR and stdout => INFO if the field severity is missing in the json

[whynowy]

I see. Then updating info output to stdout should resolve the issue.

[l15k4]

Correct, I'm from the JVM world where it is very strict and one usually finds only problems on stderr ... but I've noticed that there are infos on stderr very casually in the golang world :-)