You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The main difficulty is that the Gibbs tests take approx 90 mins which is a bit of an outlier compared to the rest. Do you think we should we split up test/mcmc/gibbs.jl into smaller segments, or should we just bite the bullet and let CI run for 90 mins (in which case we could maybe consolidate some of the others and spawn fewer runners)? I'm kind of torn, but I think I'd lean towards the latter.
Also leaning towards the latter, because it's not obvious to me what would be a natural split of the gibbs.jl tests, except by AD backend, which I guess would eventually be taken care of by #2307.
Might it make sense to reorder the tests within gibbs.jl so that all the slow ones are at the end? This way you can, if you check the running CI job, more quickly see if one of the fast ones is failing.
Following on from #2433, some of the entries that were separately split off now run in quite short amounts of time. See e.g. CI runs on #2419
We should regroup them as necessary to make sure that we aren't spawning more runners than necessary (as that slows down other CI runs https://docs.github.com/en/actions/administering-github-actions/usage-limits-billing-and-administration#usage-limits)
The text was updated successfully, but these errors were encountered: