[iris] Include per-job autoscaler hint in GetJobStatus

The hint was dropped from GetJobStatus because rebuilding the full
routing table per call was 35% of wall time. #4848 now caches the
per-job hint dict per evaluate() cycle, so the lookup is a single
dict get — attach this job's hint without serializing the routing
decision.

Author: rjpower

lib/iris/src/iris/cluster/controller/service.py

@@ -1225,17 +1225,18 @@ def get_job_status(
         )
 
         # Get scheduling diagnostics for pending jobs from cache
-        # (populated each scheduling cycle by the controller).
-        #
-        # The autoscaler pending-hint used to be appended here, but
-        # ``_get_autoscaler_pending_hints`` rebuilds + serializes the full
-        # autoscaler routing table on every call (35%+ of wall-time in a
-        # live CPU profile). Skip it for now; use ListJobs for the richer
-        # pending explanation while we work out a cached hint path.
+        # (populated each scheduling cycle by the controller). The autoscaler
+        # hint dict is cached per evaluate() cycle (#4848), so the lookup here
+        # is a single dict get — we only attach this job's hint, never the
+        # full routing decision.
         pending_reason = ""
         if job.state == job_pb2.JOB_STATE_PENDING:
             sched_reason = self._controller.get_job_scheduling_diagnostics(job.job_id.to_wire())
             pending_reason = sched_reason or "Pending scheduler feedback"
+            hint = self._get_autoscaler_pending_hints().get(job.job_id.to_wire())
+            if hint is not None:
+                scaling_prefix = "(scaling up) " if hint.is_scaling_up else ""
+                pending_reason = f"Scheduler: {pending_reason}\n\nAutoscaler: {scaling_prefix}{hint.message}"
 
         resources = _resource_spec_from_job_row(job)
 

lib/iris/tests/cluster/controller/test_dashboard.py

@@ -652,15 +652,14 @@ def test_pending_reason_uses_autoscaler_hint_for_scale_up(
         )
     }
 
-    # GetJobStatus intentionally does not append the autoscaler hint — it
-    # was the dominant hot path in a live CPU profile (35% of wall time
-    # spent rebuilding / serializing the routing table per RPC). ListJobs
-    # still includes the hint since it's only computed once per page.
+    # GetJobStatus appends this job's autoscaler hint via the per-cycle hint
+    # cache (#4848) — a single dict lookup, no routing-table serialization.
     job_resp = rpc_post(
         client_with_autoscaler, "GetJobStatus", {"jobId": JobName.root("test-user", "pending-scale").to_wire()}
     )
     pending_reason = job_resp.get("job", {}).get("pendingReason", "")
-    assert "Waiting for worker scale-up in scale group 'tpu_v5e_32'" not in pending_reason
+    assert "Waiting for worker scale-up in scale group 'tpu_v5e_32'" in pending_reason
+    assert "(scaling up)" in pending_reason
 
     jobs_resp = rpc_post(client_with_autoscaler, "ListJobs")
     listed = [
@@ -702,15 +701,12 @@ def test_pending_reason_uses_passive_autoscaler_hint_over_scheduler(
         )
     }
 
-    # GetJobStatus no longer appends the autoscaler hint (see
-    # test_pending_reason_uses_autoscaler_hint_for_scale_up for rationale).
-    # It still surfaces the scheduler diagnostic.
+    # GetJobStatus appends this job's autoscaler passive-wait hint.
     job_resp = rpc_post(
         client_with_autoscaler, "GetJobStatus", {"jobId": JobName.root("test-user", "diag-constraint").to_wire()}
     )
     pending_reason = job_resp.get("job", {}).get("pendingReason", "")
-    assert pending_reason
-    assert "Waiting for workers in scale group 'tpu_v5e_32' to become ready" not in pending_reason
+    assert "Waiting for workers in scale group 'tpu_v5e_32' to become ready" in pending_reason
 
 
 def test_list_jobs_shows_passive_autoscaler_wait_hint(