feat(#29): schema-cache reload via LISTEN/NOTIFY#55
Merged
Conversation
…TEN/NOTIFY Design doc and TDD implementation plan (superpowers spec/plan format) to close the db-channel gap: single-key Bier.SchemaCache snapshot, config-gated SchemaCacheListener with internal backoff, db_channel/db_channel_enabled options with CLI parity, and Bier.reload_schema_cache/1. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0117wYTPtSBQPTpNiR8a1jEj
Reviewed against the real toolchain and a running fixture DB: - The full suite's baseline is NOT zero failures: geojson cases 1616/1617/1618 fail locally AND on every CI matrix leg (PostGIS-dependent test.shops fixture is deliberately commented out; CI gates on BASELINE_FAILURES, red only if failures increase). Corrected every 'mix test -> 0 failures' expectation and replaced the unpassable 'mix precommit' final gate with its individual steps. - Hardened the Task 6 listener tests against a NOTIFY-vs-LISTEN race: the subscription is established in a handle_continue after boot, so tests now await it (:sys.get_state poll) before notifying; adjusted the expected red-phase output accordingly. - Documented that the persistent_term snapshot entry is never erased on instance stop (pre-existing per-key behavior, now one key). Every other claim verified exact: all ten read-site line refs, the existing telemetry span shape (relation_count included), Registry/postgrex_opts/CLI APIs, schema/config insertion points, test helpers, telemetry_test in the locked dep, and RouterBuilder ignoring its relations argument. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
load!/3 (boot path) and reload/1 had different :stop guarantees: reload/1 put the new snapshot before the telemetry span closed, while load!/3 put it after. Unify on load!/3 doing the swap inside its span so every [:bier, :schema_cache, :load, :stop] observer sees the new snapshot, and have reload/1 delegate to it instead of duplicating the span body. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
db_channel/db_channel_enabled pushed Bier.Config to 33 fields, tripping Credo.Check.Warning.StructFieldAmount's default max of 31. Config is a flat options struct that mirrors Bier.schema/0 and grows by one field per new Bier.start_link/1 option, so raise the cap instead of restructuring it.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Reject null-byte db-channel values at boot (Postgrex.Notifications.listen/3
would otherwise raise and crash-loop the listener), reload on a first
connect that only succeeds after a retry (NOTIFYs fired during the failed
attempts were previously lost), and replace the hard match on listen/3's
result with a case that retries instead of crashing on {:eventually, ref}
or a connection that died in the window before the LISTEN call.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
milmazz
changed the title
milmazz
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #29. Reload the schema cache without restarting, PostgREST-compatibly.
Bier.SchemaCache— introspection snapshot in one:persistent_termentry; reload is a single atomic swap. Swap happens inside the[:bier, :schema_cache, :load, *]span, so:stop⇒ new snapshot visible.Bier.SchemaCacheListener— dedicated LISTEN connection;NOTIFY <db_channel>, 'reload schema'coalesces and reloads;'reload config'is a logged no-op. Internal backoff (500 ms → 30 s) — DB outages never crash-loop the supervisor; reconnects reload unconditionally.db_channel("pgrst") /db_channel_enabled(true), validated at boot, withPGRST_DB_CHANNEL[_ENABLED]CLI parity.Bier.reload_schema_cache/1— programmatic reload; failure keeps the old snapshot serving.db_channel_enabled: falseto opt out (required behind transaction-mode PgBouncer).Design/plan docs in the first two commits. 19 new tests; full suite at the pre-existing 3-failure geojson/PostGIS baseline; all precommit gates clean.
🤖 Generated with Claude Code