fix: provisionerd: add more context to logs emitted, fix log level #6508
Conversation
This reverts commit d010b47.
| @@ -895,8 +895,9 @@ func (r *Runner) buildWorkspace(ctx context.Context, stage string, req *sdkproto | |||
| }) | |||
| case *sdkproto.Provision_Response_Complete: | |||
| if msgType.Complete.Error != "" { | |||
| r.logger.Info(context.Background(), "provision failed; updating state", | |||
| r.logger.Warn(context.Background(), "provision failed; updating state", | |||
There was a problem hiding this comment.
note: logging at level ERROR breaks existing unit tests, and I don't want to set slogtest.IgnoreErrors.
There was a problem hiding this comment.
Understandable. It might be the "correct" approach though, if we expect errors to be logged, then setting IgnoreErrors seems like the right course of action.
There was a problem hiding this comment.
@mafredri Would you agree that anything "ERROR" and above should be mainly for events useful to a system administrator?
There was a problem hiding this comment.
Could be relevant for e.g. template authors, but generally I doubt coder/provisioner logs (via logger) would be surfaced to anyone but systems admins.
| case sdkproto.LogLevel_ERROR: | ||
| r.logger.Error(ctx, msg, fields...) | ||
| default: // should never happen, but we should not explode either. | ||
| r.logger.Info(ctx, msg, fields...) |
There was a problem hiding this comment.
In theory, it could be possible that a customer misconfigures the Coder deployment, INFO will be selected as default logging level, and it can blow up the cluster if there are thousands of logs, but that's just theory.
There was a problem hiding this comment.
INFO is the status quo, and the only way for someone to select a different level is to set CODER_VERBOSE or --verbose, both of which need to be set explicitly.
ERRORWARN and including theerrorfield in the message.