[Do not merge] testing CI build cache - #4916
Conversation
|
I need to dig into this a bit more but the size of the cache seem to have dropped quite a bit with this change. The cache used in this PR are tagged with refs/pull/4916/merge when compared to another PR 4915 in the screenshot. This is helpful since we have 10 GB cache capacity and this PR used up 2.15 GB of cache. Linux-gcc-* dropped from 680 MB to 540 MB 
|
| runs-on: ubuntu-latest | ||
| permissions: | ||
| # `actions:write` permission is required to delete caches | ||
| actions: write |
There was a problem hiding this comment.
The cleanup of old cache would need Actions: write permission for the GITHUB_TOKEN. The current token has read permissions.
https://github.com/stellar/stellar-core/actions/runs/17415709423/job/49449860025?pr=4916#step:1:20
GITHUB_TOKEN Permissions
Actions: read
Attestations: read
Checks: read
Contents: read
Deployments: read
Discussions: read
Issues: read
Metadata: read
Models: read
Packages: read
Pages: read
PullRequests: read
RepositoryProjects: read
SecurityEvents: read
Statuses: read
graydon
left a comment
graydon
left a comment
There was a problem hiding this comment.
I think the added cache pressure coming from branches/PRs here will probably cause cached master state to get evicted and lead to more misses overall. at least that was the theory we were operating on: both final merges and branches/PRs are all ultimately close to master, so master is the most important thing to remain in-cache. so I'm not sure this is worth doing at this point.
(Of course if we ever get larger caches through github or 3rd party runners we should cache everything, it's just we only have 10GB to work with)
|
Closing this since we are exploring 3rd party cache options. |
2 participants
testing CI build cache