Skip to content

caching.md

Latest commit

 

History

History
37 lines (28 loc) · 1.23 KB

caching.md

File metadata and controls

37 lines (28 loc) · 1.23 KB
Raw

Schema cache & CI/CD

Introspection needs broad pg_catalog access and adds cold-start latency. The schema cache removes both:

$ pdbq schema dump -o schema.cache        # at build/deploy time
$ pdbq serve --schema.cache_path schema.cache   # at runtime

The cache is gzipped JSON of the introspected catalog with a format version and a content hash; serve refuses mismatched versions and corrupted files. Booting from cache never touches pg_catalog, so the runtime role only needs privileges on the actual data.

CI drift gate

$ pdbq schema check --cache schema.cache   # exit 1 when live DB drifted

Typical pipeline:

  1. Migrations run.
  2. pdbq schema check — fails the build if the committed cache no longer matches the migrated database.
  3. On failure, regenerate (pdbq schema dump), review the diff of pdbq schema print output, commit both.

Interactions

  • watch.enabled and schema.cache_path are mutually exclusive (validated at startup): watch mode exists to chase a changing schema, the cache exists to freeze one.
  • The cache stores the catalog, not the GraphQL schema — plugins and naming config still apply at boot, so changing plugin config does not require a new dump.