handle repeated update-safety digest changes

internal/ai/findings_update_safety.go

@@ -105,13 +105,13 @@ func (w *UpdateSafetyWatcher) Observe(hosts []models.DockerHost, now time.Time)
 					continue
 				}
 				// Digest changed -- transition to state B.
-				snap.priorDigest         = snap.digest
-				snap.changeDigest        = c.ImageDigest
-				snap.baseRestarts        = snap.restartCount
+				snap.priorDigest = snap.digest
+				snap.changeDigest = c.ImageDigest
+				snap.baseRestarts = snap.restartCount
 				snap.lastEmittedRestarts = snap.restartCount
-				snap.detectedAt          = now
-				snap.digest              = c.ImageDigest
-				snap.restartCount        = c.RestartCount
+				snap.detectedAt = now
+				snap.digest = c.ImageDigest
+				snap.restartCount = c.RestartCount
 
 				severity := FindingSeverityInfo
 				if c.RestartCount > snap.baseRestarts {
@@ -123,7 +123,28 @@ func (w *UpdateSafetyWatcher) Observe(hosts []models.DockerHost, now time.Time)
 			}
 
 			// State B: change already detected, verifying stability.
-			snap.digest       = c.ImageDigest
+			if c.ImageDigest != snap.digest {
+				// Another image change landed before the prior verification
+				// window completed. Treat it as a fresh update and restart the
+				// window instead of resolving against stale evidence.
+				snap.priorDigest = snap.digest
+				snap.changeDigest = c.ImageDigest
+				snap.baseRestarts = snap.restartCount
+				snap.lastEmittedRestarts = snap.restartCount
+				snap.detectedAt = now
+				snap.digest = c.ImageDigest
+				snap.restartCount = c.RestartCount
+
+				severity := FindingSeverityInfo
+				if c.RestartCount > snap.baseRestarts {
+					severity = FindingSeverityWarning
+					snap.lastEmittedRestarts = c.RestartCount
+				}
+				emit = append(emit, buildUpdateSafetyFinding(key, host, c, snap, severity, now, now))
+				continue
+			}
+
+			snap.digest = c.ImageDigest
 			snap.restartCount = c.RestartCount
 			restartsAfterChange := c.RestartCount - snap.baseRestarts
 
@@ -138,10 +159,10 @@ func (w *UpdateSafetyWatcher) Observe(hosts []models.DockerHost, now time.Time)
 				// Stable for the full window -- emit resolve sentinel and reset.
 				dedupKey := UpdateSafetyFindingPrefix + ":" + key
 				resolve = append(resolve, resolveSentinel{DedupKey: dedupKey, Reason: updateSafetyResolveReason})
-				snap.detectedAt          = time.Time{}
-				snap.priorDigest         = ""
-				snap.changeDigest        = ""
-				snap.baseRestarts        = 0
+				snap.detectedAt = time.Time{}
+				snap.priorDigest = ""
+				snap.changeDigest = ""
+				snap.baseRestarts = 0
 				snap.lastEmittedRestarts = 0
 			}
 			// Otherwise: still in window, no new restarts -- do nothing.

internal/ai/findings_update_safety_test.go

@@ -138,6 +138,43 @@ func TestUpdateSafety_StableWindowEmitsResolveSentinel(t *testing.T) {
 	}
 }
 
+// TestUpdateSafety_SecondDigestChangeResetsVerificationWindow verifies that a
+// second image update during the verification window restarts the window and
+// updates the existing finding evidence rather than resolving the old update.
+func TestUpdateSafety_SecondDigestChangeResetsVerificationWindow(t *testing.T) {
+	w := newUpdateSafetyWatcher()
+	t0 := time.Now()
+
+	w.Observe([]models.DockerHost{makeHost("h1", "c1", "sha256:aaa", 0)}, t0)
+
+	emit1, resolve1 := w.Observe([]models.DockerHost{makeHost("h1", "c1", "sha256:bbb", 0)}, t0.Add(5*time.Second))
+	if len(emit1) != 1 || len(resolve1) != 0 {
+		t.Fatalf("first update: want 1 emit and 0 resolves, got emit=%d resolve=%d", len(emit1), len(resolve1))
+	}
+
+	secondUpdateAt := t0.Add(20 * time.Second)
+	emit2, resolve2 := w.Observe([]models.DockerHost{makeHost("h1", "c1", "sha256:ccc", 0)}, secondUpdateAt)
+	if len(emit2) != 1 || len(resolve2) != 0 {
+		t.Fatalf("second update: want 1 emit and 0 resolves, got emit=%d resolve=%d", len(emit2), len(resolve2))
+	}
+	if emit2[0].Evidence != "prior_digest=sha256:bbb new_digest=sha256:ccc restart_count=0" {
+		t.Fatalf("second update evidence = %q", emit2[0].Evidence)
+	}
+
+	// The first update's window has elapsed, but the second update's window has
+	// not. Resolving now would close against stale evidence.
+	oldWindowElapsed := t0.Add(5*time.Second + updateSafetyVerifyWindow + time.Second)
+	emit3, resolve3 := w.Observe([]models.DockerHost{makeHost("h1", "c1", "sha256:ccc", 0)}, oldWindowElapsed)
+	if len(emit3) != 0 || len(resolve3) != 0 {
+		t.Fatalf("old window elapsed: want silent, got emit=%d resolve=%d", len(emit3), len(resolve3))
+	}
+
+	emit4, resolve4 := w.Observe([]models.DockerHost{makeHost("h1", "c1", "sha256:ccc", 0)}, secondUpdateAt.Add(updateSafetyVerifyWindow+time.Second))
+	if len(emit4) != 0 || len(resolve4) != 1 {
+		t.Fatalf("second window elapsed: want 0 emit and 1 resolve, got emit=%d resolve=%d", len(emit4), len(resolve4))
+	}
+}
+
 // TestUpdateSafety_EmptyDigestEmitsNothing verifies that containers with an
 // empty ImageDigest (agent not yet reporting one) are silently skipped.
 func TestUpdateSafety_EmptyDigestEmitsNothing(t *testing.T) {