Add context reasons to notifications failed counter

notify/notify.go

@@ -790,6 +790,11 @@
 		case <-ctx.Done():
 			if iErr == nil {
 				iErr = ctx.Err()
+				if errors.Is(iErr, context.Canceled) {
+					iErr = NewErrorWithReason(ContextCanceledReason, iErr)
+				} else if errors.Is(iErr, context.DeadlineExceeded) {
+					iErr = NewErrorWithReason(ContextDeadlineExceededReason, iErr)
+				}
 			}
 
 			return ctx, nil, errors.Wrapf(iErr, "%s/%s: notify retry canceled after %d attempts", r.groupName, r.integration.String(), i)
@@ -808,14 +813,15 @@
 				if !retry {
 					return ctx, alerts, errors.Wrapf(err, "%s/%s: notify retry canceled due to unrecoverable error after %d attempts", r.groupName, r.integration.String(), i)
 				}
-				if ctx.Err() == nil && (iErr == nil || err.Error() != iErr.Error()) {
-					// Log the error if the context isn't done and the error isn't the same as before.
-					level.Warn(l).Log("msg", "Notify attempt failed, will retry later", "attempts", i, "err", err)
+				if ctx.Err() == nil {
+					if iErr == nil || err.Error() != iErr.Error() {
+						// Log the error if the context isn't done and the error isn't the same as before.
+						level.Warn(l).Log("msg", "Notify attempt failed, will retry later", "attempts", i, "err", err)
+					}
+					// Save this error to be able to return the last seen error by an
+					// integration upon context timeout.
+					iErr = err
 				}
-
-				// Save this error to be able to return the last seen error by an
-				// integration upon context timeout.
-				iErr = err
 			} else {
 				lvl := level.Info(l)
 				if i <= 1 {
@@ -827,7 +833,12 @@
 			}
 		case <-ctx.Done():
 			if iErr == nil {
-				iErr = ctx.Err()
+				iErr := ctx.Err()
+				if errors.Is(iErr, context.Canceled) {
+					iErr = NewErrorWithReason(ContextCanceledReason, iErr)
+				} else if errors.Is(iErr, context.DeadlineExceeded) {
+					iErr = NewErrorWithReason(ContextDeadlineExceededReason, iErr)
+				}
 			}
 
 			return ctx, nil, errors.Wrapf(iErr, "%s/%s: notify retry canceled after %d attempts", r.groupName, r.integration.String(), i)

notify/notify_test.go

@@ -469,6 +469,39 @@ func TestRetryStageWithErrorCode(t *testing.T) {
 	}
 }
 
+func TestRetryStageWithContextCanceled(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+
+	i := Integration{
+		name: "test",
+		notifier: notifierFunc(func(ctx context.Context, alerts ...*types.Alert) (bool, error) {
+			cancel()
+			return true, errors.New("request failed: context canceled")
+		}),
+		rs: sendResolved(false),
+	}
+	r := NewRetryStage(i, "", NewMetrics(prometheus.NewRegistry(), featurecontrol.NoopFlags{}))
+
+	alerts := []*types.Alert{
+		{
+			Alert: model.Alert{
+				EndsAt: time.Now().Add(time.Hour),
+			},
+		},
+	}
+
+	ctx = WithFiringAlerts(ctx, []uint64{0})
+
+	// Notify with a non-recoverable error.
+	resctx, _, err := r.Exec(ctx, log.NewNopLogger(), alerts...)
+	counter := r.metrics.numTotalFailedNotifications
+
+	require.Equal(t, 1, int(prom_testutil.ToFloat64(counter.WithLabelValues(r.integration.Name(), ContextCanceledReason.String()))))
+
+	require.NotNil(t, err)
+	require.NotNil(t, resctx)
+}
+
 func TestRetryStageNoResolved(t *testing.T) {
 	sent := []*types.Alert{}
 	i := Integration{

notify/util.go

@@ -270,6 +270,8 @@ const (
 	DefaultReason Reason = iota
 	ClientErrorReason
 	ServerErrorReason
+	ContextCanceledReason
+	ContextDeadlineExceededReason
 )
 
 func (s Reason) String() string {
@@ -280,13 +282,17 @@ func (s Reason) String() string {
 		return "clientError"
 	case ServerErrorReason:
 		return "serverError"
+	case ContextCanceledReason:
+		return "contextCanceled"
+	case ContextDeadlineExceededReason:
+		return "contextDeadlineExceeded"
 	default:
 		panic(fmt.Sprintf("unknown Reason: %d", s))
 	}
 }
 
 // possibleFailureReasonCategory is a list of possible failure reason.
-var possibleFailureReasonCategory = []string{DefaultReason.String(), ClientErrorReason.String(), ServerErrorReason.String()}
+var possibleFailureReasonCategory = []string{DefaultReason.String(), ClientErrorReason.String(), ServerErrorReason.String(), ContextCanceledReason.String(), ContextDeadlineExceededReason.String()}
 
 // GetFailureReasonFromStatusCode returns the reason for the failure based on the status code provided.
 func GetFailureReasonFromStatusCode(statusCode int) Reason {