A report the store cannot take yet is handed back to the broker, not acknowledged and lost (novox/hq issue 082)

internal/link/serve.go

@@ -52,8 +52,20 @@ type Server struct {
 	log      *log.Logger
 	upgrader Upgrader
 	replayer Replayer
+	// again is how long a report the store could not take waits before it is handed back to
+	// the broker; zero means TryAgainAfter.
+	again time.Duration
 }
 
+// ErrTryAgain marks a listener's failure as "not now": what it was given is worth keeping and
+// asking again, as when the store is restarting (novox/hq issue 082).
+var ErrTryAgain = errors.New("not now, try again")
+
+// TryAgainAfter is the pause before a report the store could not take goes back to the broker.
+// The consumer takes one message at a time, so without it a restarting store would be asked in
+// a tight loop; a store comes back in seconds, and a report a few seconds late is still current.
+const TryAgainAfter = 2 * time.Second
+
 // Records tells the server where to keep build results.
 //
 // Set after Connect rather than passed to it, because a control plane that only publishes — the
@@ -311,6 +323,19 @@ func (s *Server) handleReport(delivery amqp.Delivery) {
 	}
 	if s.listener != nil {
 		if err := s.listener.Heard(context.Background(), report); err != nil {
+			if errors.Is(err, ErrTryAgain) {
+				// Kept, not acknowledged. The node reports an apply once, and a report lost here
+				// is a node the mesh never hears from again: the store restarting under the
+				// adoption that node just applied lost exactly that (novox/hq issue 082).
+				again := s.again
+				if again == 0 {
+					again = TryAgainAfter
+				}
+				s.log.Printf("could not record %s's report yet, and will again in %s: %v", report.Node, again, err)
+				time.Sleep(again)
+				_ = delivery.Nack(false, true)
+				return
+			}
 			// Said rather than swallowed. A report the mesh heard and failed to write down is a
 			// node whose recovery copy is silently older than it looks.
 			s.log.Printf("could not record %s's report: %v", report.Node, err)