Repository navigation
Warn when the pre-test init time becomes much slower than a worker restart - #183
Conversation
82d0b68 to
3bc5c39
Compare
|
Haven't looked at the code, I appreciate the idea, but that message so long feels a bit too much, probably a very brief explanation and a pointer to the docs would be better |
I agree I'll fix it up in the near future |
3bc5c39 to
896ff0a
Compare
|
@giordano This should be ready for final review. I improved the job selection docs so that the warning message has something to refer to, and I got rid of the pointless |
45e928a to
2fd0193
Compare
giordano
left a comment
There was a problem hiding this comment.
Looks good, apart from making default_njobs private
| ```@docs | ||
| ParsedArgs | ||
| WorkerTestSet | ||
| default_njobs |
There was a problem hiding this comment.
Uhm, default_njobs used to be non-internal, and I'm already using it in the wild 🥲 it's useful to query what would be the default number of jobs
There was a problem hiding this comment.
I’ll revert that
There was a problem hiding this comment.
I fixed this up. I somehow missed that this had been purposely made public so I also reverted the docstring changes I made in #142
e1eec80 to
c102a49
Compare
|
I'm going to clean up the commits and merge with rebase since I snuck in a few unrelated changes (like getting rid of the |
…start Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
c102a49 to
2388463
Compare
|
Uhm, I'm seeing the warning at https://buildkite.com/clima/oceananigans/builds/33771#01a0c0e5-d574-45c1-b8fd-2abf0eb504b8/L719, but I don't think the machine is memory-constrained, GC time is good, following jobs have ~stable init time. |
|
Maybe I should increase the multiplier threshold to 3x? |
|
Or maybe same multipler but also a minimum threshold of 5 or 10 seconds or so |
|
Yeah, another warning at https://buildkite.com/clima/oceananigans/builds/33771#01a0c0e5-d573-476c-bf5f-64e2bd93153e/L713, but there are only 11 concurrent workers out of 32 available cores at the moment, and memory usage is expected and well below critical values.
For the numbers I've seen so far 3x should suffice in these jobs, I'm now getting worried that any factor could have counterexamples 😕 |
A threshold may help |
|
Once I’m at my computer I’ll try threshold I have a feeling this is only a problem for fast init time testsuites. But that’s probably most. |
Close #182
Screenshots:
