Some groups' error rate is suspiciously flat
Extracted from #1066 (comment 573317432)
From the logs, there are hundreds 5xx requests returned to the users in the last 7 days:
That's not a big numbers comparing to the total of 12M requests, but the metric should show something. From the metrics in Thanos, it looks like the data is flat. I suspect that something is wrong here:
Solution
The solution is simple: move labkit middleware to in front of the Gitlab::Metrics::RequestsRackMiddleware
, and use current context in that middleware instead of the rack headers.
Edited by Quang-Minh Nguyen