v1.0.5

Release Notes - v1.0.5

Highlights

• Fixed leaked advisory lock-holder sessions that could block subsequent CI runs.
• Hardened stale-lock recovery for create with safer targeting and retry behavior.
• Added stronger regression coverage, including real grant-permission verification for app users.

What Changed

• Lock reliability in scripts/preview-branch.sh:

  • Track and terminate the exact PostgreSQL backend PID used by the lock-holder session.
  • Keep stale-lock cleanup scoped to application_name=postgres-branching-lock-holder.
  • Preserve timeout recovery flow (cleanup + retry) for lock acquisition.

• Regression tests for lock behavior:

  • tests/test_create_action_integration.sh
    • run create twice and assert no lingering lock-holder session
    • inject synthetic stale lock-holder and verify auto-cleanup + retry path
    • strengthen cleanup for lock-holder backends
  • tests/test_create_action.sh
    • make mock psql counter concurrency-safe with file locking
    • align negative-path env setup consistency

• Grant-permission coverage improvements:

  • tests/test_reset_action.sh
    • add grant SQL coverage for reset flow with APP_DB_USER
  • tests/test_create_action_integration.sh
    • validate effective read/write privileges for APP_DB_USER
    • validate default privileges on new tables/sequences created after grant
  • tests/test_reset_action_integration.sh
    • same effective-permission and default-privilege validation after reset

Validation

Executed locally and in integration Docker PostgreSQL 18:

• bash tests/test_create_action.sh
• bash tests/test_reset_action.sh
• bash tests/test_delete_action.sh
• bash tests/test_create_action_integration.sh
• bash tests/test_reset_action_integration.sh
• bash tests/test_delete_action_integration.sh

Compatibility

• No breaking input/output changes.
• Existing workflows remain compatible.

What's Changed

• fix: prevent leaked advisory lock holders and add regression tests by @Eiley2 in #2

Full Changelog: v1.0.4...v1.0.5