This context describes the cache data managed by the server throughout its lifecycle.
Cache Entry: A cache item available for matching and restoration by a workflow.
Storage Budget: The maximum amount of cache-server-managed, finalized cache payloads that may occupy storage before capacity-based eviction reclaims space. An explicit byte-based maximum may define it for any storage backend; otherwise it is a configurable percentage of the filesystem capacity, defaulting to 90%. Object-storage backends have no budget unless an explicit maximum is configured. Avoid: Storage limit, disk limit
Filesystem Capacity: The total capacity and occupancy of the mounted volume that contains filesystem storage, including data outside the cache directory. Avoid: Cache directory size
Capacity-based Eviction: Removal of finalized cache entries after an upload completes to bring cache storage within its Storage Budget, ordered by Cache Recency. Each eviction pass reclaims space until use is 90% of the budget. Avoid: Cache rotation, cleanup at X%
Storage Reconciliation: A resumable, one-time startup measurement that records byte usage for stored cache data predating size tracking. Avoid: Size backfill, bucket scan
Cache Access: Authorization to retrieve a cache payload, either by starting a proxied download or issuing a direct-download URL. It determines a cache payload's recency even when completion cannot be observed. Avoid: Successful download
Cache Recency: The ordering value for Capacity-based Eviction: a cache payload's most recent Cache Access, falling back to the time its cache entry was last saved or replaced. Avoid: Cache age
Upload: Cache data that is being received but has not yet become a Cache Entry.
Storage Location: The stored data belonging to a Cache Entry.
Part: A segment of cache data stored before its merged representation has been created.
Merge: The creation of a cache's consolidated stored representation from its Parts.
Merge Lease: A time-bound, fenced claim granting one worker authority to complete a Merge.
Storage Reader Lease: A time-bound claim that a download is actively reading data from a Storage Location.
Part Reader Lease: A Storage Reader Lease held by a download that is reading a Storage Location's Parts.
Orphaned Storage: Stored cache data that, after a safety grace period, belongs to neither an Upload nor a Storage Location. Avoid: Orphan blob, orphaned storage location
Dangling Cache Entry: A Cache Entry whose Storage Location references storage that no longer physically exists, caused by external mutation of storage the server owns (bucket wipe, external lifecycle expiry, out-of-sync database restore). The mirror of Orphaned Storage. Avoid: stale entry, missing cache
Cache Hit: A download-URL request that matched an existing Cache Entry — by exact key or by a restore-key prefix — and passed dangling-entry validation.
Cache Miss: A download-URL request that found no usable Cache Entry. Includes the case where the only match was a Dangling Cache Entry that was purged with no fallback match.
Results Passthrough: The transparent forwarding of a Results request that the cache server does not handle, and its response, without interpreting either.
Default Results Origin:
The upstream GitHub Actions Results service that receives Results Passthrough requests, configured by DEFAULT_ACTIONS_RESULTS_URL.
Avoid: Artifact server, fallback server