Cache miss even though has cache before with exact restore key

[DawnNguyenAhiho]

I have already run cache action on base branch before, it saved with restore key macOS-ruby-<my Gemfile.lock has>. But after 4 day not using that cache, today I run an action and when to the step cache, it say cache not found for inputs key: the key with the exact key that saved before.

I'm using cache v3.

Is this an expected behavior of cache action, how can I expand the timeout of cache file. I don't want after some days off, I have to bundle install again.

Thank you

[keithslater]

They are transitions their backend from v1 to v2. This could be related to an issue I saw where sometimes it will use v1 and sometimes it will use v2.

[DawnNguyenAhiho]

@keithslater oh does that mean I can't do anything but accept the situation that sometime my cache lost and I have to install all again

[fe-ax]

This is breaking our pipelines too

[AlexanderRichert-NOAA]

I've rerun my pipeline any number of times and it's still not working, please fix/revert whatever broke this.

[chohanbin]

I've been having lots of cache miss with actions/cache@v3 too. For me, using actions/cache@v4 fixed the issue for now.

[keithslater]

One thing you can try is creating a repository variable called ACTIONS_CACHE_SERVICE_V2 and set it to true. It sounds like this should force GitHub Actions to always use v2 backend and not switch back and forth.

[fe-ax]

I feel like this maybe has something to do with runner types. One of my runners is arm64 and one is amd64. I guess one of them talks to v1 and one to v2. I write to the cache from both arches and then read it only from amd64.

I'm excited to try this env variable tomorrow.

For now I'm pushing to artifact store and downloading it in the next job while immediately deleting the artifact again after. This is reliable.

[AlexanderRichert-NOAA]

No luck here with ACTIONS_CACHE_SERVICE_V2=true as repo var, but thanks @keithslater for the suggestion.

[DawnNguyenAhiho]

I'm using cache@v4 now and it work for me. Even though the runs are a week apart, I still get a cache hit. I think you guys can upgrade to cache v4