Remove "experimental" banner for OTel Metrics

[ferruzzi]

OpenTelemetry metrics have been live for a while now and there have been no major bugs except the ongoing discussions around timers reporting in seconds or milliseconds (example). I propose we drop the "Experimental" banner for OTel Metrics and leave the new Experimental for OTel Traces.

@howardyoo
@potiuk
@romanzdk
@o-nikolas

closes: #40218

[eladkal]

If we make it stable we need also newsfragmant to change log

[potiuk]

Yep. Agree with Elad on the newsfragment adding. Also would be worth checking if there are no other places where otel is marked as experimental (if not done already).

[hussein-awala]

Usually, we start lazy voting on the dev mailing list to merge similar PRs.

[ferruzzi]

I'll get the voting thread going and add the news fragment, thanks folks.

[ferruzzi]

@hussein-awala@potiuk - No comments on the lazy consensus email after 7 days. I think we're good?

[potiuk]

Absolutely. I also think we should remove it from traces - it will have the same problem :)

[ferruzzi]

I thought we discussed that and decided to do Traces separately... but now I don't see it in this thread so maybe I made that up. Unfortunately, my lazy consensus email explicitly stated it was for OTel Metrics, so I'll need a new consensus post which would mean leaving this open for another week. Up to you, I'm not aware of any particular time-sensitive reason to rush this so either way works for me.

[potiuk]

Oh yes. It could / should be separate PR.

[ferruzzi]

I've got a bunch of meetings today but I'll squeeze in time to get the LC email out for the Traces and I'll own that as well, once that clears the waiting period.

[DjVinnii]

@utkarsharma2 I thought this PR was going to be cherry-picked for Airflow 2.9.3 (See last comment of @potiuk in the discussion #40218). Is there a specific reason this is now moved to 2.10? We would like to migrate to OTel metrics, however this banner is still blocking us as we don't want to show that in our production environments.

[potiuk]

Yeah. I think it would be great to include that one. Really the banner was a mistake - it is enough to describe feature as experimental, showing a banner is a bad idea, because if someone turns it on, they do it deliberately and do not want to see the banner. So we can easily treat that one as "bugfix".

[utkarsharma2]

@potiuk@DjVinnii There was a conflict that was difficult to resolve and to do so would require us to include a PR which is part of AIP-49 and have a milestone of 2.10.

I can take a second look at it, but the conflict resolution won't be clean if we don't include the earlier PR. Please let me know if we still want to include it.

[potiuk]

I think the change could be applied manually - this is a very simple change - we've done that in the past.

[utkarsharma2]

I think the change could be applied manually - this is a very simple change - we've done that in the past.

Sure, I'll manually apply them.

[raphaelauv]

@potiuk@ferruzzi

#34405 , is that not a concern ?

airflow that stop working cause otel-collector is down is not similar to what happen with statsd

[potiuk]

#34405 , is that not a concern ?

airflow that stop working cause otel-collector is down is not similar to what happen with statsd

It's independent from being "experimental" - it's just regular - potential - bug fix that @feruzzi was supposed to follow up on :D

It does require discussion. I think the current behaviour (failure when otel collector is down) is correct. Mostly because if you are using OTEL collector, it becomes your way to monitor some issues and silently skipping over metrics generation is not likely a good idea (if you have permanent issue with your otel collector - you can still disable it in configuration until it is solved, with generally the same result as if we skipped it).

But I think it requires a short discussion and asking what others think, but it's not blocking the experimental status changing (also removal of the banner is not equal with removing experimental status - it's just that the banner was a bad idea in the first place).