[pujitha24]

Description

CREATE TRIGGER statements were never replicated in postgres-to-postgres CDC, while DROP TRIGGER statements were. In strict mode, dropping a trigger that was never created downstream causes pgstream to stop with an error. This fixes the root cause in the pgstream.emit_ddl() Postgres event trigger function: pg_event_trigger_ddl_commands() reports a NULL schema_name for CREATE TRIGGER (triggers are not schema-qualified catalog objects the way tables or functions are), and emit_ddl() silently returns without emitting anything when the schema name is NULL. DROP TRIGGER uses a different function, pg_event_trigger_dropped_objects(), which does report the schema, so it was unaffected.

Related Issue(s)

• Fixes #(issue number)

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• Added pgstream.get_trigger_schema(trigger_oid), a small SQL helper that resolves a trigger's schema via pg_trigger → pg_class → pg_namespace, mirroring the existing pgstream.get_table_metadata helper.
• Redefined pgstream.emit_ddl() (both the core and injector migration variants) so that wherever it reads schema_name from pg_event_trigger_ddl_commands(), it falls back to pgstream.get_trigger_schema(objid) for trigger objects. This is needed in two places: the rec_schema_name lookup (which gates whether anything is emitted at all) and the cmd subquery used to build the objects array (whose WHERE filter also excluded NULL-schema rows, so the trigger object itself was missing from affected_objects too).
• Added the new logic as migrations core/4_fix_create_trigger_schema and injector/6_fix_create_trigger_schema, following this repo's existing pattern for DDL emission fixes (e.g. core/3_fix_duplicate_ddl_emission). Regenerated migrations.go for both packages via make gen-migrations.

Testing

• Manual testing performed
• Unit tests added/updated
• Integration tests added/updated
• All existing tests pass

This bug lives entirely in a Postgres event trigger function (plpgsql), so it can only be proven against a real Postgres server, not a Go unit test. Docker/testcontainers were unavailable in the environment this change was built in, so I validated directly against a local PostgreSQL 16 server (Homebrew, wal_level=logical) instead of make integration-test:

• Applied the actual pre-fix migration SQL (core 1-3) and reproduced the reported bug exactly: creating a table, function, and trigger, then dropping the trigger, and reading the real logical replication stream (pg_create_logical_replication_slot + pg_logical_slot_get_changes) showed 0 CREATE TRIGGER messages and 1 DROP TRIGGER message.
• Applied the fix (core migrations 1-4) and reran the same scenario: CREATE TRIGGER now emits a pgstream.ddl message with the correct "schema_name": "public" and the trigger object listed with "schema": "public".
• Verified the down migration round-trips correctly: applied up 1-4, ran the down migration for 4, confirmed ALTER TABLE ... DROP COLUMN still emits exactly one message (i.e. migration 3's duplicate-emission fix is preserved after reverting 4) and that CREATE TRIGGER goes back to emitting nothing (expected, since that reverts to the pre-fix state); then re-applied the up migration and confirmed CREATE TRIGGER works again.
• go build, go vet, and golangci-lint run --config=.golangci.yml all pass cleanly against the changed packages (./migrations/..., ./internal/migrator/...).
• ./bin/license-header-checker -r ./license-header.txt . go (the exact command make license-check runs) passes: "492 licenses ok, 0 licenses replaced, 0 licenses added".
• I could not run go build ./..., make test, or make integration-test for the full repository in this environment: the sandbox's disk is nearly completely full for reasons unrelated to this change (confirmed the same "no space left on device" errors occur building unrelated large dependency subtrees, while the specific packages this change touches build and vet cleanly). Base branch CI (gh run list --branch main) is green, so this is an environment limitation, not a signal about the change itself.

Checklist

• Code follows project style guidelines
• Self-review completed
• Code is well-commented
• Documentation updated where necessary

Additional Notes

Note: make test / make integration-test could not be run locally due to a disk-space limitation in the build environment, unrelated to this change (see Testing section above for the live-Postgres validation performed instead).

Fixes #1146

[github-actions]

Coverage

Total: 63.1% (+4.7% vs main)

Coverage in packages changed by this PR:

Package	Coverage	Δ
migrations/postgres/core	0.0%	±0.0%
migrations/postgres/injector	0.0%	±0.0%

[pujitha24]

This is rebased on main and CI is green; let me know if anything would make it easier to review.