Umm, maybe @beorn7 or @zenador can help take a look?

I agree that we should not issue the info-level warning for an unnamed metric, mostly because it is possible to write a query where an unnamed metric can still be seen as a counter.

However, in your case, i.e. rate(sum by (series) http_requests_total[5m:1m]), I would recommend to write the expression as sum by (series) (rate(http_requests_total[5m:1m])) to make sure counter resets are treated properly. ("First the rate, then aggregate.") In fact, in most cases, I would assume an unnamed metric should not be considered a counter anymore. But as said, because there are cases where that's fine, we should not issue the warning.

Note that there is already #12945 for another case where we issue this info-level warning falsely.

However, in your case, i.e. rate(sum by (series) http_requests_total[5m:1m]), I would recommend to write the expression as sum by (series) (rate(http_requests_total[5m:1m])) to make sure counter resets are treated properly.

Thanks for the reply. I agree with you. The query might not be a good example. Just want to show that this warning might be wrong for those unamed metrics.

Yes, definitely. We should not create the annotation if rate is applied to an unnamed metric.

So we had a bunch of people submitting fixes for this. A fix for the issue addressed by this PR has already been merged (without tracking who submitted what first, please don't be mad at me because of that). Therefore, I'll close this. Thanks for your contribution anyway.