Associate events in db before queueing them to send (#2833)

Fixes a race condition between sending federation events and having them
fully associated in the database.

federationapi/queue/destinationqueue.go

@@ -76,6 +76,9 @@ func (oq *destinationQueue) sendEvent(event *gomatrixserverlib.HeaderedEvent, re
 		return
 	}
 
+	// Check if the destination is blacklisted. If it isn't then wake
+	// up the queue.
+	if !oq.statistics.Blacklisted() {
 		// If there's room in memory to hold the event then add it to the
 		// list.
 		oq.pendingMutex.Lock()
@@ -93,6 +96,7 @@ func (oq *destinationQueue) sendEvent(event *gomatrixserverlib.HeaderedEvent, re
 			oq.wakeQueueAndNotify()
 		}
 	}
+}
 
 // sendEDU adds the EDU event to the pending queue for the destination.
 // If the queue is empty then it starts a background goroutine to
@@ -103,6 +107,9 @@ func (oq *destinationQueue) sendEDU(event *gomatrixserverlib.EDU, receipt *share
 		return
 	}
 
+	// Check if the destination is blacklisted. If it isn't then wake
+	// up the queue.
+	if !oq.statistics.Blacklisted() {
 		// If there's room in memory to hold the event then add it to the
 		// list.
 		oq.pendingMutex.Lock()
@@ -120,6 +127,7 @@ func (oq *destinationQueue) sendEDU(event *gomatrixserverlib.EDU, receipt *share
 			oq.wakeQueueAndNotify()
 		}
 	}
+}
 
 // handleBackoffNotifier is registered as the backoff notification
 // callback with Statistics. It will wakeup and notify the queue

federationapi/queue/queue.go

@@ -247,9 +247,10 @@ func (oqs *OutgoingQueues) SendEvent(
 		return fmt.Errorf("sendevent: oqs.db.StoreJSON: %w", err)
 	}
 
+	destQueues := make([]*destinationQueue, 0, len(destmap))
 	for destination := range destmap {
-		if queue := oqs.getQueue(destination); queue != nil && !queue.statistics.Blacklisted() {
+		if queue := oqs.getQueue(destination); queue != nil {
-			queue.sendEvent(ev, nid)
+			destQueues = append(destQueues, queue)
 		} else {
 			delete(destmap, destination)
 		}
@@ -267,6 +268,14 @@ func (oqs *OutgoingQueues) SendEvent(
 		return err
 	}
 
+	// NOTE : PDUs should be associated with destinations before sending
+	// them, otherwise this is technically a race.
+	// If the send completes before they are associated then they won't
+	// get properly cleaned up in the database.
+	for _, queue := range destQueues {
+		queue.sendEvent(ev, nid)
+	}
+
 	return nil
 }
 
@@ -335,20 +344,21 @@ func (oqs *OutgoingQueues) SendEDU(
 		return fmt.Errorf("sendevent: oqs.db.StoreJSON: %w", err)
 	}
 
+	destQueues := make([]*destinationQueue, 0, len(destmap))
 	for destination := range destmap {
-		if queue := oqs.getQueue(destination); queue != nil && !queue.statistics.Blacklisted() {
+		if queue := oqs.getQueue(destination); queue != nil {
-			queue.sendEDU(e, nid)
+			destQueues = append(destQueues, queue)
 		} else {
 			delete(destmap, destination)
 		}
 	}
 
 	// Create a database entry that associates the given PDU NID with
-	// this destination queue. We'll then be able to retrieve the PDU
+	// these destination queues. We'll then be able to retrieve the PDU
 	// later.
 	if err := oqs.db.AssociateEDUWithDestinations(
 		oqs.process.Context(),
-		destmap, // the destination server name
+		destmap, // the destination server names
 		nid,     // NIDs from federationapi_queue_json table
 		e.Type,
 		nil, // this will use the default expireEDUTypes map
@@ -357,6 +367,14 @@ func (oqs *OutgoingQueues) SendEDU(
 		return err
 	}
 
+	// NOTE : EDUs should be associated with destinations before sending
+	// them, otherwise this is technically a race.
+	// If the send completes before they are associated then they won't
+	// get properly cleaned up in the database.
+	for _, queue := range destQueues {
+		queue.sendEDU(e, nid)
+	}
+
 	return nil
 }