From a61ef55d5622e0ec38edd4296ce232aa45078287 Mon Sep 17 00:00:00 2001
From: "pulse-triage[bot]"
 <249995291+pulse-triage[bot]@users.noreply.github.com>
Date: Sun, 6 Sep 2026 00:13:19 +0100
Subject: [PATCH] Protect recurring alert delivery after resolution and restart

Resolution must retire obsolete queue rows without permanently muting a resource condition. Extend the restart/retry regression to require delivery of a new incident sharing the resolved alert ID, alongside recovery and surviving grouped alerts.

Change-source: pulse-maintainer
---
 .../notifications/queue_resolution_retry_test.go   | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/internal/notifications/queue_resolution_retry_test.go b/internal/notifications/queue_resolution_retry_test.go
index 96cedecee..4c2cfaf7d 100644
--- a/internal/notifications/queue_resolution_retry_test.go
+++ b/internal/notifications/queue_resolution_retry_test.go
@@ -78,6 +78,15 @@ func TestResolvedTerminalFiringIsNotReplayedAfterRestart(t *testing.T) {
 			if retried != 2 {
 				t.Errorf("retried %d rows, want only surviving group and recovery", retried)
 			}
+			// Alert IDs identify a resource/condition and can recur. Resolution
+			// must suppress the old queue row, not permanently mute that ID.
+			if err := q.Enqueue(&QueuedNotification{
+				ID: "new-incident", Type: "webhook", Status: QueueStatusPending,
+				Config: []byte("{}"), MaxAttempts: 3,
+				Alerts: []*alerts.Alert{{ID: "healthy"}},
+			}); err != nil {
+				t.Fatal(err)
+			}
 			delivered := map[string][]string{}
 			var mu sync.Mutex
 			q.SetProcessor(func(n *QueuedNotification) error {
@@ -104,10 +113,11 @@ func TestResolvedTerminalFiringIsNotReplayedAfterRestart(t *testing.T) {
 				time.Sleep(time.Millisecond)
 			}
 			mu.Lock()
-			if len(delivered) != 2 || len(delivered["group"]) != 1 ||
+			if len(delivered) != 3 || len(delivered["new-incident"]) != 1 ||
+				delivered["new-incident"][0] != "healthy" || len(delivered["group"]) != 1 ||
 				delivered["group"][0] != "still-firing" ||
 				len(delivered["recovery"]) != 1 || delivered["recovery"][0] != "healthy" {
-				t.Errorf("replayed payloads = %v, want only still-firing and genuine recovery", delivered)
+				t.Errorf("replayed payloads = %v, want still-firing, genuine recovery and the new incident", delivered)
 			}
 			mu.Unlock()
 			var failures int