CI Update main lock-file with ccache-related work-around - #34763
Conversation
We were relying on it being part of the docker image but it is no longer part of it.
`ccache` is now installed in the conda environment so we don't need to modiFY PATH anymore.
ccache to the docs build dependencies
|
cc @AnneBeyer - let's see if this solves it |
|
From the log you can see that I think it's worth leaving the |
|
Oh well, it looks like OpenML is not having a good day, see HTTP 503 errors in the build log and then the build goes over the 1 hour limit? Let's trigger it again and see what happens. |
|
I thought we cache the datasets from OpenML? Does the cache work like "let me check if we have the file locally, yes we do, no need to make any network requests" or "let me get information over the network, then check for the file locally, ok we have it"? (Maybe talking about caching is a side quest for a new issue?) |
|
Maybe cloning https://github.com/lesteve/scikit_learn_data in We do have cache but it's not shared with Right now, it looks like it doesn't exist for this PR maybe because there was no successful CI run yet? |
|
Also, if the OpenML problem persists, checking what is broken exactly and report it to OpenML would be nice |
I had assumed that a PR can read the cache from I've pondered using a GitHub repo (or S3 bucket or ...) to host datasets. But, do we want to get ourselves into the business of hosting them? Because if it is too easy to use and works too well, the rest of the internet will start using it as well :-/ And suddenly we have to maintain something that was just an optimisation. |
The repo uses git-lfs and there is a limit to how much LFS "bandwidth" we can use
cf05ca2 to
95c0d6d
Compare
|
Just to be clear, this was meant as a temporary thing, not a long-term "solution". I pushed an empty commit with If it does, then I guess we can revert the temporary |
|
FWIW message from the OpenML Slack today at 14:47 Paris time:
|
|
It doesn't seem like the |
|
The full doc build worked fine in previous commit 0ba91b6, I removed the temporary work-around of fetching the cache (which didn't seem to be working). Also from a Slack message, OpenML said that it should be back to normal more or less so I guess we may be alright. In my last commit f001475, I only kept the relevant fixes i.e.:
If the full doc build works, I think we can merge this. |
ccache to the docs build dependencies|
Another hang after 6 hours in https://github.com/scikit-learn/scikit-learn/actions/runs/32278077227/job/96204883572?pr=34763. I restarted the build. Last output is a apt-get update 🤷 (similar to the other one I bumped into IIRC) |
|
The full doc build passed, let's merge this one, thanks for having a look at this @betatim! |
…n#34763) Co-authored-by: Lock file bot <[email protected]> Co-authored-by: Loïc Estève <[email protected]>
…n#34763) Co-authored-by: Lock file bot <[email protected]> Co-authored-by: Loïc Estève <[email protected]>
…n#34763) Co-authored-by: Lock file bot <[email protected]> Co-authored-by: Loïc Estève <[email protected]>
…n#34763) Co-authored-by: Lock file bot <[email protected]> Co-authored-by: Loïc Estève <[email protected]>
Reference Issues/PRs
Fixes #34761
What does this implement/fix? Explain your changes.
It looks like we were relying onccachebeing part of the docker image we get from Circle CI. We now explicitly install it.Edit: it use meson built-in detection of ccache to avoid a weird issue that
CCandCXXare not defined, maybe a miniconda/conda bug, that I (@lesteve) is not interested in investigating right now.