LangBot/tests/integration/persistence/fixtures/release_4_10_11_audit.md

v4.10.11 database upgrade audit

Baseline: v4.10.11, commit 273b8839b91cb74dfd083644efea18849a3fcee1. Audited target: 48c6a1f536bab421cbaab98a1e880fe767034463.

Changes covered

• Agent definitions, run ledger, interactions/delivery results, event logs and transcripts.
• Bot legacy pipeline routing to event bindings, followed by legacy-column removal; plugin processors.
• Workspace tenancy/ownership repairs and migration branch merges.
• Unified runner state, assistant conversations and pipeline migration snapshots.
• Passkey and TOTP credentials.
• Knowledge-base initialization/drafts and engine document identity.
• Operation traceability, monitoring links/tool calls, execution host leases and metadata text storage.

The existing migration chain covers the audited ORM changes. No additional runtime migration was required by the exercised upgrade paths.

Added regression

release_4_10_11_schema.json contains immutable dialect-specific DDL exported from the tagged release ORM, rather than current metadata. The test applies historical migrations through 0023_bot_scoped_sessions to include migration-owned objects such as pgvector, seeds representative legacy records, then runs the current persistence schema initialization twice.

Assertions cover legacy bot/pipeline preservation, converted bot event bindings, default plugin processors, initialized knowledge bases, SQLite foreign-key consistency, and missing tables/columns/indexes/constraints or column type/nullability differences against current metadata. Equivalent existing index/constraint column sets are accepted; extra migration-owned objects and removed-object diffs are not treated as missing schema. This is representative upgrade coverage, not an exhaustive test of every possible historical data value.

Executed verification

Disposable SQLite files and an isolated PostgreSQL/pgvector container were used; no business database was changed.

• New tagged-release upgrade test: 2 passed (SQLite and PostgreSQL), including repeated initialization.
• Existing PostgreSQL migration / release-migration / RAG identity suite: 85 passed, 4 failed on the first completed run. The four failures were release configuration validation because the test URL omitted the required port.
• Release-migration suite rerun with explicit :5432: 5 passed.
• Ruff check and format: passed for the new regression test.

Reproduce the new regression with TEST_POSTGRES_URL pointing to a disposable PostgreSQL/pgvector instance whose test role can create/drop databases:

pytest tests/integration/persistence/test_release_41011_upgrade.py -q

Without TEST_POSTGRES_URL, PostgreSQL is explicitly skipped rather than reported as tested.