fix(mt): Tighten tenant work-gating writer hooks (2.5/3) (#10268)

backend/onyx/background/celery/tasks/connector_deletion/tasks.py

@@ -166,16 +166,21 @@ def check_for_connector_deletion_task(self: Task, *, tenant_id: str) -> bool | N
 
             r.set(OnyxRedisSignals.BLOCK_VALIDATE_CONNECTOR_DELETION_FENCES, 1, ex=300)
 
-        # collect cc_pair_ids
+        # collect cc_pair_ids and note whether any are in DELETING status
         cc_pair_ids: list[int] = []
+        has_deleting_cc_pair = False
         with get_session_with_current_tenant() as db_session:
             cc_pairs = get_connector_credential_pairs(db_session)
             for cc_pair in cc_pairs:
                 cc_pair_ids.append(cc_pair.id)
+                if cc_pair.status == ConnectorCredentialPairStatus.DELETING:
+                    has_deleting_cc_pair = True
 
-        # Tenant-work-gating hook: any cc_pair means deletion could have
-        # cleanup work to do for this tenant on some cycle.
-        if cc_pair_ids:
+        # Tenant-work-gating hook: mark only when at least one cc_pair is in
+        # DELETING status. Marking on bare cc_pair existence would keep
+        # nearly every tenant in the active set since most have cc_pairs
+        # but almost none are actively being deleted on any given cycle.
+        if has_deleting_cc_pair:
             maybe_mark_tenant_active(tenant_id)
 
         # try running cleanup on the cc_pair_ids

backend/onyx/background/celery/tasks/docprocessing/tasks.py

@@ -897,11 +897,6 @@ def check_for_indexing(self: Task, *, tenant_id: str) -> int | None:
 
                 secondary_cc_pair_ids = standard_cc_pair_ids
 
-        # Tenant-work-gating hook: refresh this tenant's active-set membership
-        # whenever indexing actually has work to dispatch.
-        if primary_cc_pair_ids or secondary_cc_pair_ids:
-            maybe_mark_tenant_active(tenant_id)
-
         # Flag CC pairs in repeated error state for primary/current search settings
         with get_session_with_current_tenant() as db_session:
             for cc_pair_id in primary_cc_pair_ids:
@@ -1019,6 +1014,14 @@ def check_for_indexing(self: Task, *, tenant_id: str) -> int | None:
                     f"Skipping secondary indexing: switchover_type=INSTANT for search_settings={secondary_search_settings.id}"
                 )
 
+        # Tenant-work-gating hook: refresh membership only when indexing
+        # actually dispatched at least one docfetching task. `_kickoff_indexing_tasks`
+        # internally calls `should_index()` to decide per-cc_pair; using
+        # `tasks_created > 0` here gives us a "real work was done" signal
+        # rather than just "tenant has a cc_pair somewhere."
+        if tasks_created > 0:
+            maybe_mark_tenant_active(tenant_id)
+
         # 2/3: VALIDATE
         # Check for inconsistent index attempts - active attempts without task IDs
         # This can happen if attempt creation fails partway through

backend/onyx/background/celery/tasks/pruning/tasks.py

@@ -229,11 +229,7 @@ def check_for_pruning(self: Task, *, tenant_id: str) -> bool | None:
                 for cc_pair_entry in cc_pairs:
                     cc_pair_ids.append(cc_pair_entry.id)
 
-            # Tenant-work-gating hook: any cc_pair means pruning could have
-            # work to do for this tenant on some cycle.
-            if cc_pair_ids:
-                maybe_mark_tenant_active(tenant_id)
-
+            prune_dispatched = False
             for cc_pair_id in cc_pair_ids:
                 lock_beat.reacquire()
                 with get_session_with_current_tenant() as db_session:
@@ -256,9 +252,18 @@ def check_for_pruning(self: Task, *, tenant_id: str) -> bool | None:
                         logger.info(f"Pruning not created: {cc_pair_id}")
                         continue
 
+                    prune_dispatched = True
                     task_logger.info(
                         f"Pruning queued: cc_pair={cc_pair.id} id={payload_id}"
                     )
+
+            # Tenant-work-gating hook: mark only when at least one cc_pair
+            # was actually due for pruning AND a prune task was dispatched.
+            # Marking on bare cc_pair existence over-counts the population
+            # since most tenants have cc_pairs but almost none are due on
+            # any given cycle.
+            if prune_dispatched:
+                maybe_mark_tenant_active(tenant_id)
             r.set(OnyxRedisSignals.BLOCK_PRUNING, 1, ex=_get_pruning_block_expiration())
 
         # we want to run this less frequently than the overall task

backend/onyx/redis/redis_pool.py

@@ -126,6 +126,8 @@ class TenantRedis(redis.Redis):
             "srem",
             "scard",
             "zadd",
+            "zrange",
+            "zrevrange",
             "zrangebyscore",
             "zremrangebyscore",
             "zscore",

web/tests/e2e/chat/welcome_page.spec.ts

@@ -35,7 +35,6 @@ for (const theme of THEMES) {
 
       await expectScreenshot(page, {
         name: `welcome-${theme}-full-page`,
-        hide: ['[data-testid="onyx-logo"]'], // greeting text is random, hide to prevent size variation
       });
     });
 

web/tests/e2e/utils/visualRegression.ts

@@ -31,6 +31,7 @@ const DEFAULT_MASK_SELECTORS: string[] = [
  */
 const DEFAULT_HIDE_SELECTORS: string[] = [
   '[data-testid="toast-container"]',
+  '[data-testid="onyx-logo"] p', // greeting text is random, hide to prevent size variation
   // TODO: Remove once it loads consistently.
   '[data-testid="actions-container"]',
 ];