GitHub actions runners resource usage

[larsoner]

I was just waiting half an hour for a GHA job to start and it wouldn't. GHA runners (20 of them) are shared for the mne-tools organization, so I looked at what's taking up all the resources (here if you have access), and all 20 were in some form of these:

Any ideas for how to possibly reduce the usage here?

[larsoner]

... without having looked if you do similar, in MNE-Python and other repos we make use of concurrency with PRs etc to limit usage at least for PRs

https://github.com/mne-tools/mne-python/blob/c08a2c7664410d20998e4e74db081bad8279ae07/.github/workflows/tests.yml#L2-L4

[larsoner]

Seeing slowdowns across MNE-Python / MNE-BIDS-Pipeline / MNE-Realtime today because of this issue... any ideas here?

Maybe as a quick fix CodeQL could move to a weekly and workflow_dispatch (which you could manually trigger before release) and/or release?

I'll also open a support ticket with GitHub about increasing the limit but would be good to decrease usage here regardless if possible.

[larsoner]

Closing since hopefully concurrency commits have fixed this, will reopen if needed!

[larsoner]

FYI the action to fix the teams from mne-tools/mne-tools#21 is held up by all 20 runners being in use (16 of 20 for our org by MNE-CPP at the moment) 😓

So looks like this is maybe still an issue. Do you see any possible additional improvements at your end?

I noticed 18 of the 21 jobs are failing for recent commits:

... maybe disabling those until they're being fixed / worked on would make sense?

[larsoner]

For now I'll resort to manually killing some actions on staging (like the MaxQt actions runs that are all failing currently anyway) since it looks like there are pushes there that will continue to clog things for the org, but it would be better to come up with a more sustainable solution. If you want to see the failures, feel free to restart the builds (and hopefully my cancellations don't cause too much interruption)!