* feat(persistence): expand thread incarnation storage Add nullable thread and MCP task incarnation columns while preserving mixed-version writes. New thread records receive stable incarnation IDs, and new task rows copy the matching owned or shared thread incarnation without changing any read, claim, session, or deletion behavior. * test(persistence): pin incarnation rollback compatibility * test(api): pin internal thread response boundary * fix(persistence): rebase incarnation rollout after projects --------- Co-authored-by: CorgiBoyG <CorgiBoyG@users.noreply.github.com>
17 KiB
Schema Migrations (packages/harness/deerflow/persistence/migrations/)
DeerFlow's application tables (runs, threads_meta, feedback, users, run_events, plus the four channel_* tables) are owned by alembic via a hybrid bootstrap strategy. LangGraph's checkpointer tables (checkpoints, checkpoint_blobs, checkpoint_writes, checkpoint_migrations) live in the same database but are owned by LangGraph and excluded from alembic's view via migrations/_env_filters.py::include_object.
Convention: every ORM model change (new column, new table, new index) MUST ship as an alembic revision under migrations/versions/. The Gateway runs alembic upgrade head automatically on startup; routine production upgrades do not require manual Alembic commands. The audited offline recovery below is an exception for the out-of-tree incarnation revision.
Hybrid bootstrap (persistence/bootstrap.py::bootstrap_schema, invoked from persistence/engine.py::init_engine):
| DB state | Action |
|---|---|
| empty (no DeerFlow tables) | create_all + alembic stamp head |
legacy (DeerFlow tables, no alembic_version) |
create_all (baseline tables only, backfill) + alembic stamp 0001_baseline + upgrade head |
versioned (one locally known alembic_version row) |
alembic upgrade head |
0019_thread_incarnations with all current ORM tables/columns |
warn and skip migration |
0019_thread_incarnations missing local tables/columns |
refuse startup; offline recovery required |
| unknown revision, empty version table, or multiple version rows | fail closed and refuse to start |
The legacy branch handles pre-alembic databases that already have at least one DeerFlow-owned table. create_all runs first because stamping at 0001_baseline makes alembic skip the baseline's own create_table DDL on the subsequent upgrade — so any baseline table introduced into Base.metadata after the user's DB was first provisioned (e.g. the channel_* tables from PR #1930 for users upgrading across multiple releases) would otherwise never be created, and the first request hitting that table would 500 with no such table. The backfill is restricted to _BASELINE_TABLE_NAMES so it does not also create tables that future revisions introduce — those revisions' own op.create_table would otherwise fail with relation already exists. A guard test pins _BASELINE_TABLE_NAMES against 0001_baseline.upgrade()'s actual output, so editing 0001 to add or remove a table forces a matching update to the constant. Column-level shape (pre-#3658 vs post-#3658 vs manual-ALTER for token_usage_by_model) is answered by each versions/*.py revision via the idempotent helpers in migrations/_helpers.py (safe_add_column / safe_drop_column) which no-op when the change is already present and logger.warning on shape drift. Adding a new ORM column / table only requires a new revision file — no edit to bootstrap.py is needed unless the new revision adds a new baseline table (rare; only happens when a new model is part of the baseline rather than introduced by its own revision).
The empty-DB path keeps using create_all because Base.metadata is the only authoritative schema source — create_all renders both SQLite (JSON, type affinity) and Postgres (JSONB, partial indexes) correctly without anyone having to keep a hand-written baseline in lockstep. 0001_baseline.upgrade() is therefore almost never executed in practice; it exists as a stamp target + chain root.
Rolling forward compatibility: the local chain is
0018_oauth_identity_pg_partial → 0019_projects →
0020_threads_meta_project_id → 0021_batch_acceptance →
0019_thread_incarnations. The final revision deliberately retains the exact
id audited by the rollback-floor binary; Alembic orders revisions by
down_revision, not by the numeric prefix.
The deployed 0020_threads_meta_project_id rollback-floor binary knows neither
0021_batch_acceptance nor 0019_thread_incarnations. It treats only the final
incarnation revision as forward-compatible, after reflection confirms every
table and column in its own ORM schema. The intervening acceptance columns and
the incarnation columns are nullable and have no server default, so old
repositories may omit them. Tests must prove old reads and writes across both
additive revisions; do not model the rollback binary with 0021 in its local
revision set.
The same incarnation revision id existed briefly as an out-of-tree child of
0018. Current and future binaries that know the reused
0019_thread_incarnations id validate a fixed table/column snapshot of the
canonical in-tree 0019 schema whenever they see that stamp. The original
0018-plus-incarnation shape is rejected because it lacks Projects and
batch-acceptance schema; matching the revision string alone is not proof of the
new ancestry. The fixed floor is deliberately not derived from Base.metadata:
a future binary may add mapped columns after 0019 and must validate this floor
before Alembic adds them. A binary that knows 0019 upgrades normally after the
check.
Tests that model the rollback binary by removing 0021_batch_acceptance and
0019_thread_incarnations from the mocked local revision set exercise the
unknown-revision branch of the current implementation. They therefore use the
new fixed canonical-0019 floor; they do not execute the ORM-derived check from
the published 0020 binary. Reflection checks presence only; each migration
preflight owns type, nullability, and default compatibility.
Any other unknown revision, an empty version table, or multiple version rows
fails closed. Do not broaden the allowlist without proving that the rollback
repositories can read, insert, and update through the newer schema. The
exception is reviewed only for the additive 0021 JSON columns plus nullable
VARCHAR(32) threads_meta.incarnation and mcp_tasks.thread_incarnation
columns, all without non-NULL defaults, constraints, or data backfills. The
incarnation migration and tests cross-pin its revision id and DDL shape;
changing either requires a fresh old-repository compatibility audit. SQLite's
stale-upgrade recovery remains available only to a rollback binary whose
migration tree owns neither post-0020 revision. A current binary owns both,
so migration failures remain fatal.
For an existing database with the original 0018-plus-incarnation shape, use the audited offline recovery procedure. Bootstrap never re-stamps an unknown revision automatically. After stopping all writers, backing up, and verifying the exact additive schema, the operator may purge-stamp the known 0018 parent and apply the Projects, acceptance, and incarnation migrations through the current head; the extra nullable columns and their data remain intact. A regression exercises that procedure from the original schema and verifies repository reads/inserts and preservation of incarnation data.
Concurrency safety: Postgres uses pg_advisory_lock to serialise concurrent Gateway instances. SQLite uses a per-engine asyncio.Lock for same-process startup and is best-effort across processes via SQLite's file-level write lock + PRAGMA busy_timeout; multi-instance deployments should use Postgres. Column revisions in versions/ additionally use idempotent helpers (_helpers.py::safe_add_column, safe_drop_column) so repeated post-baseline changes and retries are no-ops when the change is already present.
Authoring a new revision:
cd backend && make migrate-rev MSG="add foo column to runs"
This invokes alembic revision --autogenerate against the live ORM models. Review the generated file under migrations/versions/ and switch raw op.add_column / op.drop_column calls to the idempotent helpers from _helpers.py before committing. There is no make migrate / make migrate-stamp target on purpose — routine upgrades execute at Gateway startup; the documented offline recovery is reserved for the audited out-of-tree schema.
Extension-owned tables. An extension that persists data owns its schema
end to end and must not register models against deerflow.persistence.base.Base
— doing so makes the host's empty-DB create_all create the extension's tables
on installs that never enabled it. The convention is:
-
one
MetaDatainstance private to the extension; -
every table sharing one prefix, declared via the
plugins:record'sExtensionSpec.table_prefixfield, soalembic revision --autogenerateignores them instead of reflecting them, finding them absent fromBase.metadata, and proposingdrop_table. Registration happens in two places on purpose, because two different processes read the filter:extensions/loader.py::load_extensionscovers the Gateway, andregister_configured_extension_table_prefixes()— called frommigrations/env.py— covers the alembic process, which never starts a Gateway and would otherwise see an empty prefix set exactly whereinclude_objectconsumes it. The alembic side reads the declaration out ofconfig.yamland never imports extension code: a migration process must not execute third-party code. The Gateway side registers unconditionally, even for a disabled or later-failing spec, because the tables it names may already exist in the database from a previous run. Because those two readers cannot both be right about an empty prefix — the Gateway's truthiness test reads it as "no prefix", a literal reader as one matching every table —ExtensionSpec.table_prefixcarriesmin_length=1, and the alembic-side reader (which parses raw YAML, so pydantic never runs there) skips anything that model would reject rather than raising: an operator does not expect to hear about a malformedconfig.yamlfrom alembic, and Gateway startup runs that same module throughbootstrap_schema;Scope, because it is narrower than it first appears:
make migrate-revis already safe without this.scripts/_autogen_revision.pybuilds a throwaway SQLite from the migration chain and diffs against that, so no extension table — and no LangGraph table — is ever reflected. The exposed path is runningalembic revision --autogeneratedirectly from the migrations directory, wherealembic.inipointssqlalchemy.urlat a real./data/deerflow.db. That is the same pathLANGGRAPH_OWNED_TABLEScovers, which is why that exclusion exists even though the throwaway-DB script landed in the same commit; -
an independent alembic chain with its own
version_table="<prefix>alembic_version", run fromExtensionService.start()againstExtensionRuntimeDeps.session_factory's bind — which is sequenced after the host's own bootstrap by construction, since services start once persistence is ready; -
a Postgres advisory lock around that upgrade, mirroring
bootstrap_schema, so concurrent Gateway instances serialise.
Where things live:
migrations/env.py— alembic env, delegates filter to_env_filters.py, setsrender_as_batch=Truefor SQLite ALTER supportmigrations/_env_filters.py::include_object— drops LangGraph checkpointer tables and any registered extension-owned tables (EXTENSION_TABLE_PREFIXES) from alembic's viewmigrations/_env_filters.py::register_configured_extension_table_prefixes— populates that set inside the alembic process, readingplugins[*].table_prefixfromconfig.yamland never importing extension code; called at import frommigrations/env.py, becauseload_extensions()only ever runs in the Gatewaymigrations/_helpers.py—safe_add_column/safe_drop_columnmigrations/versions/0001_baseline.py— chain root, matches the schemacreate_allproduces fromBase.metadatamigrations/versions/0002_runs_token_usage.py— fixes issue #3682migrations/versions/0004_run_ownership.py—runsmulti-worker ownership + theuq_runs_thread_activepartial unique index, with a_dedupe_active_runs_per_thread()pre-step soCREATE UNIQUE INDEXcannot fail on a field DB that already has duplicate active rows per threadmigrations/versions/0007_scheduled_run_active_index.py— theuq_scheduled_task_run_activepartial unique index (at most one queued/runningscheduled_task_runsrow pertask_id), with a_dedupe_active_scheduled_runs_per_task()pre-step (keeps the newest active row per task, supersedes the rest tointerruptedwith an explanatoryerror+finished_at) mirroring 0004; chains after0006_agentsmigrations/versions/0008_thread_operation_kind.py— addsruns.operation_kindfor durable non-run thread reservations; chains after0007_scheduled_run_active_indexmigrations/versions/0010_run_cancel_request.py— adds the nullableruns.cancel_action/cancel_requested_athandoff used by non-owning workers; chains after0009_webhook_dedupemigrations/versions/0011_mcp_tasks.py— creates the durable long-running MCP task table and its user/server/remote uniqueness constraintmigrations/versions/0012_mcp_task_results.py— adds bounded result preview/truncation/artifact fields for ordinary task driversmigrations/versions/0013_mcp_task_notifications.py— adds durable Agent-run notification snapshots, delivery leases, idempotency fields, and the separate bounded-retry attempt countermigrations/versions/0014_managed_subagents.py— creates the deployment-level managed Subagent catalog tablemigrations/versions/0015_scheduled_task_enqueue.py— interrupts legacy transient queued rows, adds durable scheduled-run launch leases and attempt counts, expands the one-active-occurrence index toqueued/launching/running, and migrates the overlap policy fromskiptoenqueue; chains after0014_managed_subagentsmigrations/versions/0016_subagent_batches.py— creates durable native-subagent batch and item tables, including owner/submission idempotency, item identity, lease/recovery state, and result fieldsmigrations/versions/0017_personal_access_tokens.py— creates the personal access token table for programmatic API accessmigrations/versions/0018_oauth_identity_pg_partial.py— convertsidx_users_oauth_identityto a partial index on Postgres (postgresql_where), matching whatUserRow.__table_args__already builds viacreate_all;0001_baselinenever applied the predicate on Postgres, so everyalembic upgrade head-provisioned deployment carried a full index until this revision. Postgres-only, idempotent (checkspg_index.indpreddirectly), no-op on SQLite (already partial viasqlite_where) and on a DB where the index doesn't exist yet. Originally generated as 0017 and renumbered to 0018 after 0017_personal_access_tokens merged first and kept that slotmigrations/versions/0019_projects.py— creates theprojectstable (id/user_id/name/instructions/presentation/status + timestamps) for the Projects Phase-1 organization feature; chains after0018_oauth_identity_pg_partialmigrations/versions/0020_threads_meta_project_id.py— adds nullablethreads_meta.project_idplusix_threads_meta_project_id(no FK by design: project delete clears membership first, and the reserveddeerflow_project_idmetadata key stays in sync); chains after0019_projectsmigrations/versions/0021_batch_acceptance.py— adds nullable per-item acceptance criteria and verdict JSON columns after0020_threads_meta_project_id; legacy rows remain uncheckedmigrations/versions/0019_thread_incarnations.py— current head; chains after0021_batch_acceptancewhile retaining the exact revision id audited by the rollback-floor binary. Adds nullablethreads_meta.incarnation/mcp_tasks.thread_incarnationcolumns. New thread rows get a random 32-character incarnation. Memory mutations serialize per thread; an overwrite inherits the existing incarnation, while a delete/recreate gets a new one. SQLite MCP task INSERTs copy the owner-or-shared incarnation with a scalar subquery in the same statement. PostgreSQL task creation holdsFOR SHARE, which conflicts with both currentFOR UPDATEmutations and an older writer's plain owner update (FOR NO KEY UPDATE). Missing or differently owned threads store NULL, old writers may omit both columns, and current API/task serialization hides them. The migration preflights both tables before DDL and its SQLite downgrade cleans only safe remnants from its own interrupted batch-copy attemptpersistence/bootstrap.py—bootstrap_schema(engine, backend=...), the three-branch provisioning decision, locked revision validation, and the narrow 0019 forward-compatibility exceptionextensions/loader.py::load_extensions— registers each spec'stable_prefixwithregister_extension_table_prefix()- Tests:
tests/test_persistence_bootstrap.py(branches),tests/test_persistence_bootstrap_concurrency.py(concurrency),tests/test_persistence_bootstrap_regression.py(issue #3682),tests/test_persistence_migrations_env.py(filter, including extension-owned tables),tests/test_extension_loader.py::TestTablePrefixRegistration(spec-to-filter wiring),tests/blocking_io/test_persistence_bootstrap.py(asyncio.to_thread anchor),tests/test_migration_0004_run_ownership_dedupe.py+tests/test_migration_0007_scheduled_run_active_dedupe.py(dedupe-before-unique-index pre-steps)