report retry progress

internal/playback/supervisor.go

@@ -27,6 +27,7 @@ func runWithRetry(
 	attempt attemptFunc,
 	shouldRetry retryDecider,
 	wait waitFunc,
+	observer retryObserver,
 ) error {
 	failedAttempts := 0
 
@@ -40,11 +41,29 @@ func runWithRetry(
 		}
 
 		failedAttempts++
-		if !shouldRetry(err) || !policy.canRetry(failedAttempts) {
+		willRetry := shouldRetry(err) && policy.canRetry(failedAttempts)
+		if !willRetry {
+			if observer != nil {
+				observer(retryEvent{
+					FailedAttempts: failedAttempts,
+					Err:            err,
+					WillRetry:      false,
+				})
+			}
 			return err
 		}
 
-		if err := wait(ctx, policy.retryDelay(failedAttempts)); err != nil {
+		delay := policy.retryDelay(failedAttempts)
+		if observer != nil {
+			observer(retryEvent{
+				FailedAttempts: failedAttempts,
+				Err:            err,
+				RetryIn:        delay,
+				WillRetry:      true,
+			})
+		}
+
+		if err := wait(ctx, delay); err != nil {
 			if ctx.Err() != nil {
 				return ctx.Err()
 			}
@@ -52,3 +71,12 @@ func runWithRetry(
 		}
 	}
 }
+
+type retryEvent struct {
+	FailedAttempts int
+	Err            error
+	RetryIn        time.Duration
+	WillRetry      bool
+}
+
+type retryObserver func(retryEvent)

internal/playback/supervisor_test.go

@@ -33,6 +33,7 @@ func TestRunWithRetryFirstAttemptSucceeds(t *testing.T) {
 			t.Fatal("wait called after successful attempt")
 			return nil
 		},
+		nil,
 	)
 
 	if err != nil {
@@ -63,6 +64,7 @@ func TestRunWithRetryFailuresThenSuccess(t *testing.T) {
 			delays = append(delays, delay)
 			return nil
 		},
+		nil,
 	)
 
 	if err != nil {
@@ -94,6 +96,7 @@ func TestRunWithRetryFiniteAttemptsExhausted(t *testing.T) {
 			waits++
 			return nil
 		},
+		nil,
 	)
 
 	if !errors.Is(err, attemptErr) {
@@ -123,6 +126,7 @@ func TestRunWithRetryUnlimitedEventuallySucceeds(t *testing.T) {
 		},
 		func(error) bool { return true },
 		func(context.Context, time.Duration) error { return nil },
+		nil,
 	)
 
 	if err != nil {
@@ -149,6 +153,7 @@ func TestRunWithRetryStopsWhenErrorIsNotRetryable(t *testing.T) {
 			t.Fatal("wait called for non-retryable error")
 			return nil
 		},
+		nil,
 	)
 
 	if !errors.Is(err, attemptErr) {
@@ -175,6 +180,7 @@ func TestRunWithRetryReturnsCancellationFromAttempt(t *testing.T) {
 			t.Fatal("wait called after cancellation")
 			return nil
 		},
+		nil,
 	)
 
 	if !errors.Is(err, context.Canceled) {
@@ -195,6 +201,7 @@ func TestRunWithRetryReturnsCancellationDuringBackoff(t *testing.T) {
 			cancel()
 			return ctx.Err()
 		},
+		nil,
 	)
 
 	if !errors.Is(err, context.Canceled) {
@@ -212,6 +219,7 @@ func TestRunWithRetryReturnsWaitError(t *testing.T) {
 		func(context.Context) error { return attemptErr },
 		func(error) bool { return true },
 		func(context.Context, time.Duration) error { return waitErr },
+		nil,
 	)
 
 	if !errors.Is(err, waitErr) {
@@ -228,3 +236,132 @@ func TestWaitForRetryReturnsCancellation(t *testing.T) {
 		t.Fatalf("waitForRetry() error = %v, want context.Canceled", err)
 	}
 }
+
+func TestRetryObserverReportsFailuresBeforeSuccess(t *testing.T) {
+	attemptErr := errors.New("attempt failed")
+	attempts := 0
+	var events []retryEvent
+
+	err := runWithRetry(
+		context.Background(),
+		testRetryPolicy(3),
+		func(context.Context) error {
+			attempts++
+			if attempts < 3 {
+				return attemptErr
+			}
+			return nil
+		},
+		func(error) bool { return true },
+		func(context.Context, time.Duration) error { return nil },
+		func(event retryEvent) {
+			events = append(events, event)
+		},
+	)
+
+	if err != nil {
+		t.Fatalf("runWithRetry() error = %v, want nil", err)
+	}
+	if len(events) != 2 {
+		t.Fatalf("event count = %d, want 2", len(events))
+	}
+
+	wantDelays := []time.Duration{500 * time.Millisecond, time.Second}
+	for i, event := range events {
+		wantAttempts := i + 1
+		if event.FailedAttempts != wantAttempts {
+			t.Errorf("event %d failed attempts = %d, want %d", i, event.FailedAttempts, wantAttempts)
+		}
+		if !errors.Is(event.Err, attemptErr) {
+			t.Errorf("event %d error = %v, want %v", i, event.Err, attemptErr)
+		}
+		if event.RetryIn != wantDelays[i] {
+			t.Errorf("event %d retry delay = %s, want %s", i, event.RetryIn, wantDelays[i])
+		}
+		if !event.WillRetry {
+			t.Errorf("event %d WillRetry = false, want true", i)
+		}
+	}
+}
+
+func TestRetryObserverReportsExhaustion(t *testing.T) {
+	attemptErr := errors.New("attempt failed")
+	var events []retryEvent
+
+	err := runWithRetry(
+		context.Background(),
+		testRetryPolicy(2),
+		func(context.Context) error { return attemptErr },
+		func(error) bool { return true },
+		func(context.Context, time.Duration) error { return nil },
+		func(event retryEvent) {
+			events = append(events, event)
+		},
+	)
+
+	if !errors.Is(err, attemptErr) {
+		t.Fatalf("runWithRetry() error = %v, want %v", err, attemptErr)
+	}
+	if len(events) != 2 {
+		t.Fatalf("event count = %d, want 2", len(events))
+	}
+	if !events[0].WillRetry || events[0].RetryIn != 500*time.Millisecond {
+		t.Errorf("first event = %+v, want retry after 500ms", events[0])
+	}
+	final := events[1]
+	if final.FailedAttempts != 2 {
+		t.Errorf("final failed attempts = %d, want 2", final.FailedAttempts)
+	}
+	if final.WillRetry {
+		t.Error("final WillRetry = true, want false")
+	}
+	if final.RetryIn != 0 {
+		t.Errorf("final retry delay = %s, want 0", final.RetryIn)
+	}
+	if !errors.Is(final.Err, attemptErr) {
+		t.Errorf("final error = %v, want %v", final.Err, attemptErr)
+	}
+}
+
+func TestRetryObserverNotCalledOnImmediateSuccess(t *testing.T) {
+	observerCalls := 0
+	err := runWithRetry(
+		context.Background(),
+		testRetryPolicy(3),
+		func(context.Context) error { return nil },
+		func(error) bool { return true },
+		func(context.Context, time.Duration) error { return nil },
+		func(retryEvent) { observerCalls++ },
+	)
+
+	if err != nil {
+		t.Fatalf("runWithRetry() error = %v, want nil", err)
+	}
+	if observerCalls != 0 {
+		t.Fatalf("observer call count = %d, want 0", observerCalls)
+	}
+}
+
+func TestRetryObserverNotCalledWhenAttemptCancelsContext(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	observerCalls := 0
+
+	err := runWithRetry(
+		ctx,
+		testRetryPolicy(0),
+		func(context.Context) error {
+			cancel()
+			return errors.New("attempt interrupted")
+		},
+		func(error) bool { return true },
+		func(context.Context, time.Duration) error { return nil },
+		func(retryEvent) { observerCalls++ },
+	)
+
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("runWithRetry() error = %v, want context.Canceled", err)
+	}
+	if observerCalls != 0 {
+		t.Fatalf("observer call count = %d, want 0", observerCalls)
+	}
+}