fix(meshmapper): retry failed catalogue fetches sooner

A transient failure pushed the next attempt out by the full refresh interval. Failures now retry in five minutes, with a server Retry-After still honoured as a ceiling.

internal/meshmapper/scopes.go

@@ -26,6 +26,8 @@ const (
 	MaxScopes    = 64 // per source; at most 16 configured sources (1024 imported identities)
 	MaxBody      = 64 << 10
 	PollInterval = 15 * time.Second // one source per tick, at most four requests/minute
+
+	failureRetry = 5 * time.Minute // transient failures shouldn't cost a whole refresh interval
 )
 
 // Cache stores source provenance and freshness separately from packet evidence.
@@ -142,6 +144,7 @@ func (i *Importer) refresh(ctx context.Context, s *source, now time.Time) error
 	response, err := i.client.Do(request)
 	var entries []scopestore.Entry
 	var generated time.Time
+	var retryAfter time.Time
 	if err != nil {
 		if ctx.Err() != nil {
 			return ctx.Err()
@@ -169,10 +172,7 @@ func (i *Importer) refresh(ctx context.Context, s *source, now time.Time) error
 			if response.StatusCode == http.StatusTooManyRequests || response.StatusCode == http.StatusServiceUnavailable {
 				// The published contract uses seconds. Ignore invalid/overflowing headers.
 				if seconds, parseErr := strconv.ParseInt(response.Header.Get("Retry-After"), 10, 64); parseErr == nil && seconds > 0 && seconds <= int64((1<<63-1)/time.Second) {
-					until := now.Add(time.Duration(seconds) * time.Second)
-					if until.After(update.NextAttempt) {
-						update.NextAttempt = until
-					}
+					retryAfter = now.Add(time.Duration(seconds) * time.Second)
 				}
 			}
 		}
@@ -191,6 +191,13 @@ func (i *Importer) refresh(ctx context.Context, s *source, now time.Time) error
 			}
 		}
 	}
+	if update.LastError != "" {
+		// Transient failures shouldn't cost a whole refresh interval, but never beat a Retry-After.
+		update.NextAttempt = now.Add(min(i.interval, failureRetry))
+		if retryAfter.After(update.NextAttempt) {
+			update.NextAttempt = retryAfter
+		}
+	}
 	if update.LastError == "HTTP 429" {
 		i.retryAfter = update.NextAttempt
 	}

internal/meshmapper/scopes_test.go

@@ -278,3 +278,42 @@ func TestInvalidSavedCatalogueDoesNotBlockStartupOrRefresh(t *testing.T) {
 		t.Fatalf("normal shutdown is a task failure: %v", err)
 	}
 }
+
+func TestRetryAfterShorterThanIntervalIsHonoured(t *testing.T) {
+	ctx := context.Background()
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		w.Header().Set("Retry-After", "1800")
+		w.WriteHeader(http.StatusTooManyRequests)
+	}))
+	defer server.Close()
+	cfg := config.MeshMapperScopesConfig{Enabled: true, Sources: map[string]string{"YOW": server.URL}}
+	imp, err := New(ctx, cfg, &memoryStore{rows: map[string]Cache{}}, scopestore.New(), nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	now := time.Now().UTC()
+	if err := imp.refresh(ctx, &imp.sources[0], now); err != nil {
+		t.Fatal(err)
+	}
+	if got := imp.sources[0].cache.NextAttempt.Sub(now); got != 30*time.Minute {
+		t.Fatalf("next attempt after 429 in %v, want %v (server Retry-After must not be overridden by the 1h interval)", got, 30*time.Minute)
+	}
+}
+
+func TestFailedRefreshRetriesSoon(t *testing.T) {
+	ctx := context.Background()
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) { w.WriteHeader(500) }))
+	defer server.Close()
+	cfg := config.MeshMapperScopesConfig{Enabled: true, Sources: map[string]string{"YOW": server.URL}}
+	imp, err := New(ctx, cfg, &memoryStore{rows: map[string]Cache{}}, scopestore.New(), nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	now := time.Now().UTC()
+	if err := imp.refresh(ctx, &imp.sources[0], now); err != nil {
+		t.Fatal(err)
+	}
+	if got := imp.sources[0].cache.NextAttempt.Sub(now); got != failureRetry {
+		t.Fatalf("next attempt after failure in %v, want %v", got, failureRetry)
+	}
+}