Send E2EE related errors to sentry (#2784)

Only sends errors if we're not retrying them in NATS.
Not sure if those should be scoped/tagged with something like "E2EE".

4 files changed, 31 insertions, 7 deletions

diff --git a/federationapi/consumers/keychange.go b/federationapi/consumers/keychange.go
index f3314bc9..67dfdc1d 100644
--- a/federationapi/consumers/keychange.go
+++ b/federationapi/consumers/keychange.go
@@ -18,6 +18,11 @@ import (
 	"context"
 	"encoding/json"
 
+	"github.com/getsentry/sentry-go"
+	"github.com/matrix-org/gomatrixserverlib"
+	"github.com/nats-io/nats.go"
+	"github.com/sirupsen/logrus"
+
 	"github.com/matrix-org/dendrite/federationapi/queue"
 	"github.com/matrix-org/dendrite/federationapi/storage"
 	"github.com/matrix-org/dendrite/federationapi/types"
@@ -26,9 +31,6 @@ import (
 	"github.com/matrix-org/dendrite/setup/config"
 	"github.com/matrix-org/dendrite/setup/jetstream"
 	"github.com/matrix-org/dendrite/setup/process"
-	"github.com/matrix-org/gomatrixserverlib"
-	"github.com/nats-io/nats.go"
-	"github.com/sirupsen/logrus"
 )
 
 // KeyChangeConsumer consumes events that originate in key server.
@@ -78,6 +80,7 @@ func (t *KeyChangeConsumer) onMessage(ctx context.Context, msgs []*nats.Msg) boo
 	msg := msgs[0] // Guaranteed to exist if onMessage is called
 	var m api.DeviceMessage
 	if err := json.Unmarshal(msg.Data, &m); err != nil {
+		sentry.CaptureException(err)
 		logrus.WithError(err).Errorf("failed to read device message from key change topic")
 		return true
 	}
@@ -105,6 +108,7 @@ func (t *KeyChangeConsumer) onDeviceKeyMessage(m api.DeviceMessage) bool {
 	// only send key change events which originated from us
 	_, originServerName, err := gomatrixserverlib.SplitID('@', m.UserID)
 	if err != nil {
+		sentry.CaptureException(err)
 		logger.WithError(err).Error("Failed to extract domain from key change event")
 		return true
 	}
@@ -118,6 +122,7 @@ func (t *KeyChangeConsumer) onDeviceKeyMessage(m api.DeviceMessage) bool {
 		WantMembership: "join",
 	}, &queryRes)
 	if err != nil {
+		sentry.CaptureException(err)
 		logger.WithError(err).Error("failed to calculate joined rooms for user")
 		return true
 	}
@@ -125,6 +130,7 @@ func (t *KeyChangeConsumer) onDeviceKeyMessage(m api.DeviceMessage) bool {
 	// send this key change to all servers who share rooms with this user.
 	destinations, err := t.db.GetJoinedHostsForRooms(t.ctx, queryRes.RoomIDs, true)
 	if err != nil {
+		sentry.CaptureException(err)
 		logger.WithError(err).Error("failed to calculate joined hosts for rooms user is in")
 		return true
 	}
@@ -147,6 +153,7 @@ func (t *KeyChangeConsumer) onDeviceKeyMessage(m api.DeviceMessage) bool {
 		Keys:              m.KeyJSON,
 	}
 	if edu.Content, err = json.Marshal(event); err != nil {
+		sentry.CaptureException(err)
 		logger.WithError(err).Error("failed to marshal EDU JSON")
 		return true
 	}
@@ -160,6 +167,7 @@ func (t *KeyChangeConsumer) onCrossSigningMessage(m api.DeviceMessage) bool {
 	output := m.CrossSigningKeyUpdate
 	_, host, err := gomatrixserverlib.SplitID('@', output.UserID)
 	if err != nil {
+		sentry.CaptureException(err)
 		logrus.WithError(err).Errorf("fedsender key change consumer: user ID parse failure")
 		return true
 	}
@@ -176,12 +184,14 @@ func (t *KeyChangeConsumer) onCrossSigningMessage(m api.DeviceMessage) bool {
 		WantMembership: "join",
 	}, &queryRes)
 	if err != nil {
+		sentry.CaptureException(err)
 		logger.WithError(err).Error("fedsender key change consumer: failed to calculate joined rooms for user")
 		return true
 	}
 	// send this key change to all servers who share rooms with this user.
 	destinations, err := t.db.GetJoinedHostsForRooms(t.ctx, queryRes.RoomIDs, true)
 	if err != nil {
+		sentry.CaptureException(err)
 		logger.WithError(err).Error("fedsender key change consumer: failed to calculate joined hosts for rooms user is in")
 		return true
 	}
@@ -196,6 +206,7 @@ func (t *KeyChangeConsumer) onCrossSigningMessage(m api.DeviceMessage) bool {
 		Origin: string(t.serverName),
 	}
 	if edu.Content, err = json.Marshal(output); err != nil {
+		sentry.CaptureException(err)
 		logger.WithError(err).Error("fedsender key change consumer: failed to marshal output, dropping")
 		return true
 	}
diff --git a/federationapi/consumers/sendtodevice.go b/federationapi/consumers/sendtodevice.go
index ffc1d889..9aec22a3 100644
--- a/federationapi/consumers/sendtodevice.go
+++ b/federationapi/consumers/sendtodevice.go
@@ -18,16 +18,18 @@ import (
 	"context"
 	"encoding/json"
 
+	"github.com/getsentry/sentry-go"
+	"github.com/matrix-org/gomatrixserverlib"
+	"github.com/matrix-org/util"
+	"github.com/nats-io/nats.go"
+	log "github.com/sirupsen/logrus"
+
 	"github.com/matrix-org/dendrite/federationapi/queue"
 	"github.com/matrix-org/dendrite/federationapi/storage"
 	"github.com/matrix-org/dendrite/setup/config"
 	"github.com/matrix-org/dendrite/setup/jetstream"
 	"github.com/matrix-org/dendrite/setup/process"
 	syncTypes "github.com/matrix-org/dendrite/syncapi/types"
-	"github.com/matrix-org/gomatrixserverlib"
-	"github.com/matrix-org/util"
-	"github.com/nats-io/nats.go"
-	log "github.com/sirupsen/logrus"
 )
 
 // OutputSendToDeviceConsumer consumes events that originate in the clientapi.
@@ -76,6 +78,7 @@ func (t *OutputSendToDeviceConsumer) onMessage(ctx context.Context, msgs []*nats
 	sender := msg.Header.Get("sender")
 	_, originServerName, err := gomatrixserverlib.SplitID('@', sender)
 	if err != nil {
+		sentry.CaptureException(err)
 		log.WithError(err).WithField("user_id", sender).Error("Failed to extract domain from send-to-device sender")
 		return true
 	}
@@ -85,12 +88,14 @@ func (t *OutputSendToDeviceConsumer) onMessage(ctx context.Context, msgs []*nats
 	// Extract the send-to-device event from msg.
 	var ote syncTypes.OutputSendToDeviceEvent
 	if err = json.Unmarshal(msg.Data, &ote); err != nil {
+		sentry.CaptureException(err)
 		log.WithError(err).Errorf("output log: message parse failed (expected send-to-device)")
 		return true
 	}
 
 	_, destServerName, err := gomatrixserverlib.SplitID('@', ote.UserID)
 	if err != nil {
+		sentry.CaptureException(err)
 		log.WithError(err).WithField("user_id", ote.UserID).Error("Failed to extract domain from send-to-device destination")
 		return true
 	}
@@ -116,6 +121,7 @@ func (t *OutputSendToDeviceConsumer) onMessage(ctx context.Context, msgs []*nats
 		},
 	}
 	if edu.Content, err = json.Marshal(tdm); err != nil {
+		sentry.CaptureException(err)
 		log.WithError(err).Error("failed to marshal EDU JSON")
 		return true
 	}
diff --git a/federationapi/queue/queue.go b/federationapi/queue/queue.go
index 88664fcf..8245aa5b 100644
--- a/federationapi/queue/queue.go
+++ b/federationapi/queue/queue.go
@@ -21,6 +21,7 @@ import (
 	"sync"
 	"time"
 
+	"github.com/getsentry/sentry-go"
 	"github.com/matrix-org/gomatrixserverlib"
 	"github.com/prometheus/client_golang/prometheus"
 	log "github.com/sirupsen/logrus"
@@ -307,11 +308,13 @@ func (oqs *OutgoingQueues) SendEDU(
 
 	ephemeralJSON, err := json.Marshal(e)
 	if err != nil {
+		sentry.CaptureException(err)
 		return fmt.Errorf("json.Marshal: %w", err)
 	}
 
 	nid, err := oqs.db.StoreJSON(oqs.process.Context(), string(ephemeralJSON))
 	if err != nil {
+		sentry.CaptureException(err)
 		return fmt.Errorf("sendevent: oqs.db.StoreJSON: %w", err)
 	}
 
diff --git a/federationapi/routing/send.go b/federationapi/routing/send.go
index 060af676..b3bbaa39 100644
--- a/federationapi/routing/send.go
+++ b/federationapi/routing/send.go
@@ -22,6 +22,7 @@ import (
 	"sync"
 	"time"
 
+	"github.com/getsentry/sentry-go"
 	"github.com/matrix-org/gomatrixserverlib"
 	"github.com/matrix-org/util"
 	"github.com/prometheus/client_golang/prometheus"
@@ -350,6 +351,7 @@ func (t *txnReq) processEDUs(ctx context.Context) {
 				for deviceID, message := range byUser {
 					// TODO: check that the user and the device actually exist here
 					if err := t.producer.SendToDevice(ctx, directPayload.Sender, userID, deviceID, directPayload.Type, message); err != nil {
+						sentry.CaptureException(err)
 						util.GetLogger(ctx).WithError(err).WithFields(logrus.Fields{
 							"sender":    directPayload.Sender,
 							"user_id":   userID,
@@ -360,6 +362,7 @@ func (t *txnReq) processEDUs(ctx context.Context) {
 			}
 		case gomatrixserverlib.MDeviceListUpdate:
 			if err := t.producer.SendDeviceListUpdate(ctx, e.Content, t.Origin); err != nil {
+				sentry.CaptureException(err)
 				util.GetLogger(ctx).WithError(err).Error("failed to InputDeviceListUpdate")
 			}
 		case gomatrixserverlib.MReceipt:
@@ -395,6 +398,7 @@ func (t *txnReq) processEDUs(ctx context.Context) {
 			}
 		case types.MSigningKeyUpdate:
 			if err := t.producer.SendSigningKeyUpdate(ctx, e.Content, t.Origin); err != nil {
+				sentry.CaptureException(err)
 				logrus.WithError(err).Errorf("Failed to process signing key update")
 			}
 		case gomatrixserverlib.MPresence: