Perf notes repo is growing to an unwieldly size

The perf notes repo has grown to ~800mb in size, it takes several minutes to clone.

I have long wondered why we are pushing the results of perf notes from MRs. This can also lead to confusing results if you are updating a MR by adding commits.

Would it make more sense to just permanently record the perf notes from the stable branches?

added needs triage label

Would it make more sense to just permanently record the perf notes from the stable branches?

Yes, this sounds like a reasonable measure given the extent to which it has grown although this means that people would lose the ability to use the perf notes to analyze the performance characteristics of their patches in MR (although I am not sure anyone currently takes advantage of this).

Other approaches include:

• ensuring that the perf notes clones are cached
• using git clone --shallow
• pushing MR results to a separate branch

I'm not close enough to the detail here, but I'd like to make sure that any perf failures reported by CI related to this MR only, and not to *the cumulative effect of all MRs since some earlier point.

(Not tracking the cumulative behaviour is a bit of a problem; but it's even more of a problem if MRs are rejected because of sins committed in unrelated MRs, that just happen to get tipped over the edge by a new MR.)

Just checking!

using git clone --shallow

git recently added partial clones as well, which imo supersede shallow clones, and are a pretty nice way of speeding up clones for CI etc. See: https://github.blog/open-source/git/get-up-to-speed-with-partial-clone-and-shallow-clone/

Gitlab CI doesn't have native support for partial clones yet (https://gitlab.com/gitlab-org/gitlab-runner/-/issues/26631), but it's pretty OK to set up manually by just providing the correct command line

removed needs triage label

added Ttask infrastructure perf tests labels

We decided that the benefit of pushing results from anything but stable branches and @marge-bot batches is probably not worth the size. We should update ci.sh to only push in those cases.

mentioned in commit 7e4b1461

mentioned in merge request !13335 (closed)

mentioned in commit 97f1c516

closed with commit 50490075

mentioned in commit 50490075

reopened

mentioned in commit 7b07c101

mentioned in commit 655c1b9f

mentioned in commit 3574768a