Repository navigation
Retry failed benchmark result pushes - #488
Conversation
There was a problem hiding this comment.
Two notes, maybe as a follow-up?
- What do you think about just adding a new command to the bench runner? Say:
python -m bench_runner push_results. This seems to be in line withLine 25 in 78c5aae
- This seems to touch only
push, not touchingpull.
There was a problem hiding this comment.
I just realized that the latest EndBug/add-and-commit has push_attempts:
The problem is that it's a bit forceful:
We've got a backoff, and they do not sleep etc. Thinking about submitting a PR there
On the other hand:
with:
pull: --rebase --autostash origin main
push_attempts: 5
looks so nice.
|
ooo, I do like using their built-in retry better. It's probably fine to go that route. Would be nice if they had sleep but it might be enough just with raw retries? |
|
@savannahostrowski My hunch it's fine til we start processing a significant number of cpython PRs (then we definitely run into more concurrency-related issues here; the whole thing is hard to follow), and it's much simpler! |
…retries # Conflicts: # bench_runner/templates/_benchmark.src.yml
|
Thanks! I’ve updated to use the action’s built-in retries. I included --autostash and looked into the existing -X ours. #323 already tracks problems with that pull’s conflict handling; an attempted fix was reverted because it broke on Windows. |
|
@savannahostrowski This is great. Thank you for checking my review so carefully. |
Another cherry-pick from my fork, with a few adjustments for upstream :)
A few months back, I encountered failed result pushes when running benchmarks across multiple machines, causing jobs to fail after the benchmarks had completed.
This PR adds up to five push attempts using EndBug/add-and-commit’s built-in push_attempts, with --rebase --autostash pulls between failed attempts. It covers benchmark results, pystats, derived results, and failure reports.