Repository navigation
Broken Alembic migration chain: KeyError '20260216_002' — fresh-DB upgrade impossible #9759
Description
Summary
The backend Alembic migration chain is broken: alembic upgrade head aborts with KeyError: '20260216_002' — a migration's down_revision references revision 20260216_002, which no file in autobot-backend/migrations/versions/ defines. The history has a gap, so migrations cannot be applied to a fresh database at all.
How found
Wiring RBAC-by-default for docker compose (umbrella #9720) requires bootstrapping the Postgres schema via alembic upgrade head. Against a fresh autobot_users DB it fails:
File ".../alembic/script/revision.py", line 245, in _revision_map
down_revision = map_[downrev]
KeyError: '20260216_002'
Also found (fixed in #9720 branch)
migrations/env.py:get_url() read config.database_url / config.db_host off the Alembic Config object (which has no such attributes) → AttributeError, making migrations unrunnable even before the chain gap. Fixed there to read AUTOBOT_DATABASE_URL / AUTOBOT_POSTGRES_HOST from env. A second latent mismatch: env.py builds an async engine but get_url() falls back to postgres_sync_url (postgresql:// → psycopg2, not installed); async callers need postgresql+asyncpg://.
Impact
- Any Postgres-backed deployment (
single_company/multi_company/provider) cannot initialize its schema from a clean DB. - Blocks RBAC-by-default in docker compose (Docker compose: make the stack run all-healthy out of the box (umbrella for #9712–#9719) #9720).
Fix needed
Locate the missing 20260216_002 revision (deleted/squashed?), repair the chain (restore the revision or re-point down_revisions), and verify alembic upgrade head applies cleanly end-to-end on a fresh Postgres. Then wire a one-shot alembic upgrade head step into the docker startup (init service or entrypoint) before the backend serves.
Activity
mrveiss commentedon Jun 8, 2026
Update — root cause is deeper than a chain gap (full investigation)
Attempting
alembic upgrade headagainst a fresh Postgres surfaced four compounding faults. Fixed the first three in the #9720 branch; the fourth is architectural:get_url()readconfig.database_url/config.db_hostoff the AlembicConfigobject (no such attrs) →AttributeError. Fixed to readAUTOBOT_DATABASE_URL/AUTOBOT_POSTGRES_HOST.20260216_003.down_revision = "20260216_002", but the secrets migration's id is the short"002"→KeyError. Re-pointed003to"002"(safe: down_revisions aren't stored in DBs).20260524_036b,20260527_046,20260604_051) — added merge migration20260608_052_merge_heads.002runsALTER TABLE secrets ADD COLUMN org_id, but no migration evercreate_table('secrets').secretsis a SQLAlchemy model (models/secret.py), created viametadata.create_all, not migrations. Lifespan onlycreate_allsSkillsBase.metadata(Skills tables) — never user_managementBase,LLCBase, or thesecrets/secrets-owning metadata. So a fresh Postgres can be bootstrapped by neitheralembic upgrade head(assumes pre-existingcreate_alltables) nor the currentcreate_all(covers only Skills). Existing single-host DBs work only because they accreted tables over time.Real fix (proposed)
Fresh-DB bootstrap should be:
create_allover the full combined metadata (user_managementBase+LLCBase+ model Bases incl.secrets) thenalembic stamp head(mark migrations applied without replaying history), wired as a one-shot docker startup step gated onpostgres_enabled. Requires auditing every model Base for create_all coverage and confirming the create_all output matcheshead.Status
RBAC-by-default (#9720) is blocked on this. All other out-of-box fixes are complete and verified healthy.
fix(docker): make compose stack run all-healthy out of the box (#9720) (
mrveiss commentedon Jun 10, 2026
Bug sweep 2026-06-10 → TOO-BIG. Cross-cutting schema/migration repair (locate the missing
20260216_002revision or re-pointdown_revisions, then provealembic upgrade headapplies clean on a fresh DB, plus wire a one-shot upgrade into docker startup). Blocks RBAC-by-default (#9720). Requires migration-chain design judgment, not a local fix — out of scope for the sweep.fix(backend): repair Alembic chain + compose migration bootstrap — fr…
github-actions commentedon Jun 11, 2026
PR #9988 (merged to
Dev_new_gui) references this issue with a close keyword.If this issue is fully resolved, close it manually. If work remains, no action is needed.
mrveiss commentedon Jun 11, 2026
Fixed by PR #9988 (merged to Dev_new_gui, squash
afd58590d).alembic upgrade head→20260608_053(68 tables); idempotent re-run; legacy create_all-secrets DB also upgrades cleansecretstable; 018 altered 5 more create_all-only tables; 022/027/030/031/032/036 re-emitted CREATE TYPE (generic sa.Enum ignores create_type=False); 047 duplicate index; env.py async/sync driver mismatchalembic upgrade head(1 retry) for Postgres-backed user modes before uvicorn —AUTOBOT_USER_MODE=single_companyis now a single-var flipfix(migrations): baseline adoption for never-migrated DBs + strict An…
consolidation(migrations): shared alembic guard helpers — has_table/h…
fix(llc): sa.Enum columns emit member VALUES not NAMES via values_cal…
deploy(backend): single_company mode — surfaces Company OS + enables …