Open KristofferC opened 3 months ago
The first line from the next step is 2 min later, but not sure why
Perhaps not relevant, but latest major version of this workflow (and setup-julia) is 2, but it's worth making sure this happens on latest version (if the issue is reproducible at all)
@KristofferC can you re-run via the GitHub interface with debugging turned on.
Hopefully updating will fix it, but also good to identify the issue if it does.
Here's Kristoffer's original job, re-ran with debug logging enabled: https://github.com/JuliaIO/BufferedStreams.jl/actions/runs/10114228187/job/28018266876
Looks like post action cleanup, perhaps the cloned registry because 1.6 just takes a long time to delete.. which it shouldn't on linux at least
Perhaps not relevant, but latest major version of this workflow (and setup-julia) is 2, but it's worth making sure this happens on latest version (if the issue is reproducible at all)
The workflow will be updated to v2 in: https://github.com/JuliaIO/BufferedStreams.jl/pull/86
Looks like the cleanup task took between 1 and 5 seconds in that PR and the following commit to master
Okay, I think that is fixed then. I wonder how many are stuck at v1, it isn't like you check for updates on github action workflows very often.
Dependabot or renovate help with that by automating the updates.
I don't think it did on that repo.
If no one merges dependabot PRs for months the bot goes stale, until someone resumes it. Which is what happened when Dilum merged https://github.com/JuliaIO/BufferedStreams.jl/pull/83.
Looks like this is the issue https://github.com/nodejs/node/issues/47228
Nice find
Is it the same? The slowdown was with version 1 of this workflow, which wasn't using node20 (as far as I understand), while there's apparently no slowdown with version 2 of this workflow which uses node20, and in the linked issue they say that's specific to node20
Isn't gha force updating old node to 20 now?
The exact 2 min hang that was reported in that issue seems suspiciously similar for it to be something else entirely.
But in v2 of this workflow the issue isn't observed, so I'm missing why the issue was reopened. And based on https://github.com/actions/cache/pull/1284#issuecomment-1885094213 the issue you found (which referred to v3 of actions/cache, not v4) was already worked around upstream.
In https://github.com/JuliaIO/BufferedStreams.jl/actions/runs/10114228187/job/27972388361, the cache step seems to take about 2 minutes.
It is however unclear where the time is actually spent since looking at the timestamps:
it is not possible to see where the 2 minutes are actually spent.