MRG: Use caching in Github Actions and Azure Pipelines#8695
MRG: Use caching in Github Actions and Azure Pipelines#8695larsoner merged 47 commits intomne-tools:masterfrom
Conversation
.github/workflows/compat_minimal.yml
Outdated
| id: cache-dataset | ||
| with: | ||
| path: /home/runner/mne_data | ||
| key: ${{ runner.os }}-dataset |
There was a problem hiding this comment.
I don't think we need to separate by OS, the data files should all be the same
|
Among the first batch of runners, no one detected the cache: and Failing skips the I'll push an empty commit to confirm that they at least restore the cache successfully. |
|
|
|
I see no problem at restoration 👌 :
|
|
Do you want to try dealing with testing dataset |
|
You'll probably be more efficient than me 😅 . Go ahead if you have the time. |
|
Will do now, could really use something in my wheelhouse to jump-start things over here... |
|
@GuillaumeFavelier if you want to look into it, we should do the same for Azure |
|
@GuillaumeFavelier you can use something like the following in a one-liner When Dan's #8679 is in progress (probably after my PR then yours) then this line can probably be adjusted fairly easily |
|
Checking the key based on the testing version is a success!
|
|
The Perl extension is not available on MacOS: |
|
Also, the cache is different depending on the OS: On MacOS: For the others (on Linux): |
|
Because there is a And no way to make their tip work since it's not a fixed string but it comes from a variable: I'll |
|
I don't know what to say 🤦 |
|
The jobs on Windows detected the key: And restored the cache correctly But the data are downloaded anyway because the path is different on Windows. |
|
Hah! 44 sec to download the data on And 1m 11 sec to get it from cache: So we actually get a slowdown from this on Azure at least (!) but at least it should be more reliable. |
|
Hm... But for some reason it downloads more no? Restoring the cache: Compared to |
That's probably the uncompressed size, vs the download which is compressed. |
|
Unrelated failure on DetailsMaybe one last empty commit for testing but everything is ok for me. Is there something missing @larsoner ? |
I don't think we need the empty commit, I'll just restart the segfaulted GitHub actions build, and it looks like everything else should be okay. Will merge once CIs are happy |
|
Thanks @GuillaumeFavelier ! |
|
Something interesting about caching:
|
* upstream/master: MAINT: Actually use all caches (mne-tools#8702) MRG: Prototype of notebook viz (ipyvtk) (mne-tools#8503) MRG: Use caching in Github Actions and Azure Pipelines (mne-tools#8695)
* First test [skip azp][skip circle] * Download anyway [skip azp][skip circle] * Unify cache key [skip azp][skip circle] * Test new key * Test linux_pip [skip azp][skip circle] * Try again [skip azp][skip circle] * Deploy to all jobs * Try another syntax * Test new key format [skip azp][skip circle] * Use home env variable * Fix key [skip azp][skip circle] * Try naive version on Azure * Try again [skip github][skip circle] * Try again [skip github][skip circle] * Try another syntax [skip github][skip circle] * Try again [skip github] * Try again [skip github] * Try again [skip github] * Try again [skip github] * Try again [skip github] * Debug variable [skip github] * Debug variable [skip github] * Debug variable [skip github] * Debug variable [skip github] * Debug variable [skip github] * Debug variable [skip github] * Debug variable [skip github] * Debug variable [skip github] * Debug variable [skip github] * Debug variable [skip github] * Test again? [skip github] * Debug variable [skip github] * Debug variable [skip github] * Debug variable [skip github] * Debug variable [skip github] * Test new key format [skip github] * Test new key format [skip github] * Test new key format [skip github] * Try again [skip github] * Try again [skip github] * Try again [skip github] * Deploy to all jobs * Try another syntax [skip github][skip circle] * Try another syntax [skip github][skip circle] * Try another syntax [skip github][skip circle] * Test again

This PR follows #8685 (comment) and attempts to solve the recent failures related to testing data download.
ToDo:
$HOMEforpathvariable