mirror of
https://github.com/cocoindex-io/cocoindex-code.git
synced 2026-09-14 16:39:38 +08:00
Jiangzhou
90d8ff9f08
The previous `type=gha` cache was scoped per Git ref, so each release tag created a new cache scope and never reused a previous release's layers. Add to that GHA's 10 GB per-repo limit (which we already hit at ~10 GB before the v0.2.26 release), and the cache effectively never helped consecutive releases. Move to `type=registry` cache stored as separate `buildcache-<variant>` tags on the same GHCR package. Branch/tag-independent, no LRU eviction, unlimited size for our purposes. Standard pattern for multi-arch buildx + multi-registry publishes. Per-variant cache tag (`buildcache-slim`, `buildcache-full`) so the two variants don't evict each other's layers. `mode=max` exports all intermediate layers, not just the final ones — needed for the per-RUN-layer reuse strategy in the Dockerfile. Also cleaned up 23 orphaned GHA cache entries on the deleted g/docker-layer-cache-v2 branch (3.4 GB freed).