Partition dataflow workers by tenant

src/govoplan_dataflow/backend/manifest.py

@@ -266,7 +266,9 @@ DOCUMENTATION = (
             "Publication to a governed Datasource uses forward recovery: an unknown provider outcome is reconciled "
             "before retry so output is not duplicated. Staging and production promotion is explicit and does not "
             "rewrite a revision. Cancellation is best effort once external work has started; the final evidence "
-            "states whether work stopped, completed, failed, or requires operator reconciliation."
+            "states whether work stopped, completed, failed, or requires operator reconciliation. Scheduled, event, "
+            "and queued execution is partitioned by tenant module entitlement before a run is claimed. Disabling "
+            "Dataflow stops new admission and leaves accepted runs available for an explicit operator decision."
         ),
         layer="available",
         documentation_types=("admin", "user"),

src/govoplan_dataflow/backend/run_worker.py

@@ -52,6 +52,7 @@ def dispatch_pending_runs(
     session: Session,
     *,
     registry: object | None,
+    tenant_id: str | None = None,
     now: datetime | None = None,
     limit: int = 10,
     worker_id: str | None = None,
@@ -65,6 +66,7 @@ def dispatch_pending_runs(
     recovered, recovered_outcome_unknown = _recover_expired_leases(
         session,
         now=current,
+        tenant_id=tenant_id,
     )
     session.commit()
     summary: dict[str, object] = {
@@ -82,6 +84,7 @@ def dispatch_pending_runs(
             session,
             now=current,
             worker_id=resolved_worker_id,
+            tenant_id=tenant_id,
         )
         if run_id is None:
             session.rollback()
@@ -105,21 +108,23 @@ def dispatch_pending_runs(
 def purge_expired_runs(
     session: Session,
     *,
+    tenant_id: str | None = None,
     now: datetime | None = None,
     limit: int = 500,
 ) -> dict[str, object]:
     current = _as_utc(now or utcnow())
+    clauses = [
+        DataflowRun.status.in_(("succeeded", "failed", "cancelled")),
+        DataflowRun.retention_until.is_not(None),
+        DataflowRun.retention_until <= current,
+        DataflowRun.purged_at.is_(None),
+    ]
+    if tenant_id:
+        clauses.append(DataflowRun.tenant_id == tenant_id)
     runs = list(
         session.scalars(
             select(DataflowRun)
-            .where(
+            .where(*clauses)
-                DataflowRun.status.in_(
-                    ("succeeded", "failed", "cancelled")
-                ),
-                DataflowRun.retention_until.is_not(None),
-                DataflowRun.retention_until <= current,
-                DataflowRun.purged_at.is_(None),
-            )
             .order_by(DataflowRun.retention_until, DataflowRun.id)
             .limit(max(1, min(int(limit), 5_000)))
             .with_for_update(skip_locked=True)
@@ -204,15 +209,19 @@ def _recover_expired_leases(
     session: Session,
     *,
     now: datetime,
+    tenant_id: str | None = None,
 ) -> tuple[int, int]:
+    clauses = [
+        DataflowRun.status == "running",
+        DataflowRun.lease_expires_at.is_not(None),
+        DataflowRun.lease_expires_at < now,
+    ]
+    if tenant_id:
+        clauses.append(DataflowRun.tenant_id == tenant_id)
     runs = list(
         session.scalars(
             select(DataflowRun)
-            .where(
+            .where(*clauses)
-                DataflowRun.status == "running",
-                DataflowRun.lease_expires_at.is_not(None),
-                DataflowRun.lease_expires_at < now,
-            )
             .with_for_update(skip_locked=True)
         )
     )
@@ -276,17 +285,21 @@ def _claim_next_run(
     *,
     now: datetime,
     worker_id: str,
+    tenant_id: str | None = None,
 ) -> str | None:
+    clauses = [
+        DataflowRun.status.in_(("queued", "retrying")),
+        or_(
+            DataflowRun.available_at.is_(None),
+            DataflowRun.available_at <= now,
+        ),
+    ]
+    if tenant_id:
+        clauses.append(DataflowRun.tenant_id == tenant_id)
     candidates = list(
         session.scalars(
             select(DataflowRun)
-            .where(
+            .where(*clauses)
-                DataflowRun.status.in_(("queued", "retrying")),
-                or_(
-                    DataflowRun.available_at.is_(None),
-                    DataflowRun.available_at <= now,
-                ),
-            )
             .order_by(DataflowRun.available_at, DataflowRun.created_at)
             .limit(40)
             .with_for_update(skip_locked=True)
@@ -723,6 +736,7 @@ class SqlDataflowRunWorker:
         self,
         session: object,
         *,
+        tenant_id: str | None = None,
         now: datetime | None = None,
         limit: int = 10,
         worker_id: str | None = None,
@@ -732,6 +746,7 @@ class SqlDataflowRunWorker:
         return dispatch_pending_runs(
             session,
             registry=self._registry,
+            tenant_id=tenant_id,
             now=now,
             limit=limit,
             worker_id=worker_id,
@@ -741,12 +756,18 @@ class SqlDataflowRunWorker:
         self,
         session: object,
         *,
+        tenant_id: str | None = None,
         now: datetime | None = None,
         limit: int = 500,
     ) -> Mapping[str, object]:
         if not isinstance(session, Session):
             raise TypeError("Dataflow retention requires a Session.")
-        return purge_expired_runs(session, now=now, limit=limit)
+        return purge_expired_runs(
+            session,
+            tenant_id=tenant_id,
+            now=now,
+            limit=limit,
+        )
 
 
 __all__ = [

src/govoplan_dataflow/backend/triggers.py

@@ -966,6 +966,7 @@ class SqlDataflowTriggerDispatcher:
         self,
         session: object,
         *,
+        tenant_id: str | None = None,
         now: datetime | None = None,
         limit: int = 50,
     ) -> Mapping[str, object]:
@@ -974,6 +975,7 @@ class SqlDataflowTriggerDispatcher:
         return dispatch_due_triggers(
             session,
             registry=self._registry,
+            tenant_id=tenant_id,
             now=now,
             limit=limit,
         ).model_dump(mode="json")