fix(alerts): compare gate arrivals over an hour, not a day

The growth check asked whether a day's worth of jobs had arrived, so the
alert only tripped once roughly a day's backlog had accumulated behind
the gate. Compare against what the gate admits over an hour instead, and
drop the drain-horizon condition, which the growth check already implied.

Emitter changes live in windmill-ee-private (see ee-repo-ref.txt bump);
this commit carries the regression test and the query cache.

backend/.sqlx/query-11ccd3d7a4fb565d66d330586c4fbb28567d9d538cefbc4f7b984114b8be6fb7.json

@@ -1,6 +1,6 @@
 {
   "db_name": "PostgreSQL",
-  "query": "\n        SELECT concurrency_gate_id AS \"gate_id!\",\n               COUNT(*) AS \"backlog!\",\n               MIN(workspace_id) AS \"workspace_id!\",\n               COUNT(*) FILTER (WHERE created_at >= NOW() - INTERVAL '24 hours')\n                   AS \"arrivals_24h!\"\n        FROM v2_job_queue\n        WHERE running = false\n          AND concurrency_gate_id IS NOT NULL\n          AND canceled_by IS NULL\n        GROUP BY concurrency_gate_id\n        HAVING MAX(concurrency_gated_at) >= NOW() - INTERVAL '1 hour'\n        ",
+  "query": "\n        SELECT concurrency_gate_id AS \"gate_id!\",\n               COUNT(*) AS \"backlog!\",\n               MIN(workspace_id) AS \"workspace_id!\",\n               COUNT(*) FILTER (WHERE created_at >= NOW() - INTERVAL '1 hour' * $1)\n                   AS \"arrivals!\"\n        FROM v2_job_queue\n        WHERE running = false\n          AND concurrency_gate_id IS NOT NULL\n          AND canceled_by IS NULL\n        GROUP BY concurrency_gate_id\n        HAVING MAX(concurrency_gated_at) >= NOW() - INTERVAL '1 hour'\n        ",
   "describe": {
     "columns": [
       {
@@ -20,12 +20,14 @@
       },
       {
         "ordinal": 3,
-        "name": "arrivals_24h!",
+        "name": "arrivals!",
         "type_info": "Int8"
       }
     ],
     "parameters": {
-      "Left": []
+      "Left": [
+        "Float8"
+      ]
     },
     "nullable": [
       true,
@@ -34,5 +36,5 @@
       null
     ]
   },
-  "hash": "a2278097381823098e1dff7cd91ffafc0262ff9e8440b562a0b5b32180f7d926"
+  "hash": "11ccd3d7a4fb565d66d330586c4fbb28567d9d538cefbc4f7b984114b8be6fb7"
 }

backend/ee-repo-ref.txt

@@ -1 +1 @@
-a31d6ae5e573d549e00e73f08b694a64a60ca119
+f1f951359200bedfe31a75abd3c3f8962553f702

backend/windmill-common/tests/concurrency_gate_alerts.rs

@@ -94,8 +94,9 @@ mod tests {
             "a gate that cannot drain its backlog must alert, got {messages:?}"
         );
         assert!(
-            messages[0].contains("17280") && messages[0].contains("40000"),
-            "alert must name the backlog and the ceiling it exceeds: {}",
+            messages[0].contains("17280") && messages[0].contains("40000")
+                && messages[0].contains("720"),
+            "alert must name the backlog and both ceilings it exceeds: {}",
             messages[0]
         );
     }
@@ -195,4 +196,23 @@ mod tests {
         .unwrap_or(0);
         assert_eq!(unresolved, 0, "a fully drained gate must have its alert recovered");
     }
+
+    /// The window governs how soon the alert fires, so a gate taking on less than
+    /// it admits over that window is keeping up and must stay silent even though
+    /// a long backlog sits in front of it.
+    #[ignore = "requires database setup - run with --ignored flag"]
+    #[sqlx::test(migrations = "../migrations")]
+    async fn arrivals_under_the_window_ceiling_do_not_alert(db: Pool<Postgres>) {
+        free_the_lock(&db).await;
+        queue_gated(&db, 40_000, SLOW_GATE, 48).await;
+        // 700 in the last hour against a 720/hour ceiling: still keeping up.
+        queue_gated(&db, 700, SLOW_GATE, 0).await;
+
+        concurrency_gate_alerts(&db).await;
+
+        assert!(
+            alert_messages(&db).await.is_empty(),
+            "arrivals below the gate ceiling for the window must not alert"
+        );
+    }
 }