log less for failed key querys, add counters for incoming pdus/edus

federationapi/routing/send.go

@@ -30,9 +30,40 @@ import (
 	"github.com/matrix-org/dendrite/setup/config"
 	"github.com/matrix-org/gomatrixserverlib"
 	"github.com/matrix-org/util"
+	"github.com/prometheus/client_golang/prometheus"
 	"github.com/sirupsen/logrus"
 )
 
+var (
+	pduCountTotal = prometheus.NewCounter(
+		prometheus.CounterOpts{
+			Namespace: "dendrite",
+			Subsystem: "federationapi",
+			Name:      "recv_pdus",
+		},
+	)
+	eduCountTotal = prometheus.NewCounter(
+		prometheus.CounterOpts{
+			Namespace: "dendrite",
+			Subsystem: "federationapi",
+			Name:      "recv_edus",
+		},
+	)
+	pduSuccessTotal = prometheus.NewCounter(
+		prometheus.CounterOpts{
+			Namespace: "dendrite",
+			Subsystem: "federationapi",
+			Name:      "recv_pdus_success",
+		},
+	)
+)
+
+func init() {
+	prometheus.MustRegister(
+		pduCountTotal, eduCountTotal, pduSuccessTotal,
+	)
+}
+
 // Send implements /_matrix/federation/v1/send/{txnID}
 func Send(
 	httpReq *http.Request,
@@ -133,6 +164,7 @@ func (t *txnReq) processTransaction(ctx context.Context) (*gomatrixserverlib.Res
 
 	pdus := []*gomatrixserverlib.HeaderedEvent{}
 	for _, pdu := range t.PDUs {
+		pduCountTotal.Inc()
 		var header struct {
 			RoomID string `json:"room_id"`
 		}
@@ -224,6 +256,7 @@ func (t *txnReq) processTransaction(ctx context.Context) (*gomatrixserverlib.Res
 			}
 		} else {
 			results[e.EventID()] = gomatrixserverlib.PDUResult{}
+			pduSuccessTotal.Inc()
 		}
 	}
 
@@ -281,6 +314,7 @@ func (t *txnReq) haveEventIDs() map[string]bool {
 
 func (t *txnReq) processEDUs(ctx context.Context) {
 	for _, e := range t.EDUs {
+		eduCountTotal.Inc()
 		switch e.Type {
 		case gomatrixserverlib.MTyping:
 			// https://matrix.org/docs/spec/server_server/latest#typing-notifications

federationsender/consumers/eduserver.go

@@ -212,8 +212,7 @@ func (t *OutputEDUConsumer) onReceiptEvent(msg *sarama.ConsumerMessage) error {
 		return nil
 	}
 	if receiptServerName != t.ServerName {
-		log.WithField("other_server", receiptServerName).Info("Suppressing receipt notif: originated elsewhere")
-		return nil
+		return nil // don't log, very spammy as it logs for each remote receipt
 	}
 
 	joined, err := t.db.GetJoinedHosts(context.TODO(), receipt.RoomID)

keyserver/internal/device_list_update.go

@@ -330,16 +330,16 @@ func (u *DeviceListUpdater) processServer(serverName gomatrixserverlib.ServerNam
 		logger.WithError(err).Error("failed to load stale device lists")
 		return waitTime, true
 	}
-	hasFailures := false
+	failCount := 0
 	for _, userID := range userIDs {
 		if ctx.Err() != nil {
 			// we've timed out, give up and go to the back of the queue to let another server be processed.
-			hasFailures = true
+			failCount += 1
 			break
 		}
 		res, err := u.fedClient.GetUserDevices(ctx, serverName, userID)
 		if err != nil {
-			logger.WithError(err).WithField("user_id", userID).Error("failed to query device keys for user")
+			failCount += 1
 			fcerr, ok := err.(*fedsenderapi.FederationClientError)
 			if ok {
 				if fcerr.RetryAfter > 0 {
@@ -351,20 +351,22 @@ func (u *DeviceListUpdater) processServer(serverName gomatrixserverlib.ServerNam
 				waitTime = time.Hour
 				logger.WithError(err).Warn("GetUserDevices returned unknown error type")
 			}
-			hasFailures = true
 			continue
 		}
 		err = u.updateDeviceList(&res)
 		if err != nil {
 			logger.WithError(err).WithField("user_id", userID).Error("fetched device list but failed to store/emit it")
-			hasFailures = true
+			failCount += 1
 		}
 	}
+	if failCount > 0 {
+		logger.WithField("total", len(userIDs)).WithField("failed", failCount).Error("failed to query device keys for some users")
+	}
 	for _, userID := range userIDs {
 		// always clear the channel to unblock Update calls regardless of success/failure
 		u.clearChannel(userID)
 	}
-	return waitTime, hasFailures
+	return waitTime, failCount > 0
 }
 
 func (u *DeviceListUpdater) updateDeviceList(res *gomatrixserverlib.RespUserDevices) error {