Merge branch 'af/use-metrics'

metrics.go

@@ -20,17 +20,16 @@ import (
 	"context"
 	"net/http"
 	"runtime/debug"
-	"strconv"
 	"sync"
 	"time"
 
+	"github.com/google/uuid"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promauto"
 	"github.com/prometheus/client_golang/prometheus/promhttp"
 	"github.com/rs/zerolog"
 
 	"github.com/element-hq/mautrix-go/event"
-	"github.com/element-hq/mautrix-go/id"
 
 	"github.com/element-hq/mautrix-signal/database"
 )
@@ -49,9 +48,6 @@ type MetricsHandler struct {
 	signalMessageAge        prometheus.Histogram
 	signalMessageHandling   *prometheus.HistogramVec
 	countCollection         prometheus.Histogram
-	disconnections          *prometheus.CounterVec
-	incomingRetryReceipts   *prometheus.CounterVec
-	connectionFailures      *prometheus.CounterVec
 	puppetCount             prometheus.Gauge
 	activePuppetCount       prometheus.Gauge
 	bridgeBlocked           prometheus.Gauge
@@ -65,10 +61,10 @@ type MetricsHandler struct {
 	unencryptedPrivateCount prometheus.Gauge
 
 	connected          prometheus.Gauge
-	connectedState     map[string]bool
+	connectedState     map[uuid.UUID]bool
 	connectedStateLock sync.Mutex
 	loggedIn           prometheus.Gauge
-	loggedInState      map[string]bool
+	loggedInState      map[uuid.UUID]bool
 	loggedInStateLock  sync.Mutex
 }
 
@@ -101,18 +97,6 @@ func NewMetricsHandler(address string, log zerolog.Logger, db *database.Database
 			Name: "bridge_count_collection",
 			Help: "Time spent collecting the bridge_*_total metrics",
 		}),
-		disconnections: promauto.NewCounterVec(prometheus.CounterOpts{
-			Name: "bridge_disconnections",
-			Help: "Number of times a Matrix user has been disconnected from Signal",
-		}, []string{"user_id"}),
-		connectionFailures: promauto.NewCounterVec(prometheus.CounterOpts{
-			Name: "bridge_connection_failures",
-			Help: "Number of times a connection has failed to Signal",
-		}, []string{"reason"}),
-		incomingRetryReceipts: promauto.NewCounterVec(prometheus.CounterOpts{
-			Name: "bridge_incoming_retry_receipts",
-			Help: "Number of times a remote Signal user has requested a retry from the bridge. retry_count = 5 is usually the last attempt (and very likely means a failed message)",
-		}, []string{"retry_count", "message_found"}),
 		puppetCount: promauto.NewGauge(prometheus.GaugeOpts{
 			Name: "bridge_puppets_total",
 			Help: "Number of Signal users bridged into Matrix",
@@ -147,12 +131,12 @@ func NewMetricsHandler(address string, log zerolog.Logger, db *database.Database
 			Name: "bridge_logged_in",
 			Help: "Bridge users logged into Signal",
 		}),
-		loggedInState: make(map[string]bool),
+		loggedInState: make(map[uuid.UUID]bool),
 		connected: promauto.NewGauge(prometheus.GaugeOpts{
 			Name: "bridge_connected",
 			Help: "Bridge users connected to Signal",
 		}),
-		connectedState: make(map[string]bool),
+		connectedState: make(map[uuid.UUID]bool),
 	}
 }
 
@@ -186,31 +170,7 @@ func (mh *MetricsHandler) TrackSignalMessage(timestamp time.Time, messageType st
 	}
 }
 
-func (mh *MetricsHandler) TrackDisconnection(userID id.UserID) {
-	if !mh.running {
-		return
-	}
-	mh.disconnections.With(prometheus.Labels{"user_id": string(userID)}).Inc()
-}
-
-func (mh *MetricsHandler) TrackConnectionFailure(reason string) {
-	if !mh.running {
-		return
-	}
-	mh.connectionFailures.With(prometheus.Labels{"reason": reason}).Inc()
-}
-
-func (mh *MetricsHandler) TrackRetryReceipt(count int, found bool) {
-	if !mh.running {
-		return
-	}
-	mh.incomingRetryReceipts.With(prometheus.Labels{
-		"retry_count":   strconv.Itoa(count),
-		"message_found": strconv.FormatBool(found),
-	}).Inc()
-}
-
-func (mh *MetricsHandler) TrackLoginState(signalID string, loggedIn bool) {
+func (mh *MetricsHandler) TrackLoginState(signalID uuid.UUID, loggedIn bool) {
 	if !mh.running {
 		return
 	}
@@ -221,13 +181,13 @@ func (mh *MetricsHandler) TrackLoginState(signalID string, loggedIn bool) {
 		mh.loggedInState[signalID] = loggedIn
 		if loggedIn {
 			mh.loggedIn.Inc()
-		} else {
+		} else if ok {
 			mh.loggedIn.Dec()
 		}
 	}
 }
 
-func (mh *MetricsHandler) TrackConnectionState(signalID string, connected bool) {
+func (mh *MetricsHandler) TrackConnectionState(signalID uuid.UUID, connected bool) {
 	if !mh.running {
 		return
 	}
@@ -238,7 +198,7 @@ func (mh *MetricsHandler) TrackConnectionState(signalID string, connected bool)
 		mh.connectedState[signalID] = connected
 		if connected {
 			mh.connected.Inc()
-		} else {
+		} else if ok {
 			mh.connected.Dec()
 		}
 	}

user.go

@@ -424,6 +424,8 @@ func (user *User) startupTryConnect(retryCount int) {
 			case signalmeow.SignalConnectionEventConnected:
 				user.log.Debug().Msg("Sending Connected BridgeState")
 				user.BridgeState.Send(status.BridgeState{StateEvent: status.StateConnected})
+				user.bridge.Metrics.TrackConnectionState(user.SignalID, true)
+				user.bridge.Metrics.TrackLoginState(user.SignalID, true)
 
 			case signalmeow.SignalConnectionEventDisconnected:
 				user.log.Debug().Msg("Received SignalConnectionEventDisconnected")
@@ -473,6 +475,7 @@ func (user *User) startupTryConnect(retryCount int) {
 						} else {
 							user.BridgeState.Send(status.BridgeState{StateEvent: status.StateTransientDisconnect, Error: "unknown-websocket-error", Message: err.Error()})
 						}
+						user.bridge.Metrics.TrackConnectionState(user.SignalID, false)
 					}
 				}
 
@@ -483,6 +486,7 @@ func (user *User) startupTryConnect(retryCount int) {
 				} else {
 					user.BridgeState.Send(status.BridgeState{StateEvent: status.StateBadCredentials, Message: err.Error()})
 				}
+				user.bridge.Metrics.TrackConnectionState(user.SignalID, false)
 				err = user.Logout(ctx)
 				if err != nil {
 					user.log.Err(err).Msg("Error logging out user in response to remote logout")
@@ -497,6 +501,7 @@ func (user *User) startupTryConnect(retryCount int) {
 			case signalmeow.SignalConnectionEventError:
 				user.log.Debug().Msg("Sending UnknownError BridgeState")
 				user.BridgeState.Send(status.BridgeState{StateEvent: status.StateUnknownError, Error: "unknown-websocket-error", Message: err.Error()})
+				user.bridge.Metrics.TrackConnectionState(user.SignalID, false)
 
 			case signalmeow.SignalConnectionCleanShutdown:
 				if user.Client == nil {
@@ -507,6 +512,7 @@ func (user *User) startupTryConnect(retryCount int) {
 					user.log.Debug().Msg("Clean Shutdown, but logged out - Sending BadCredentials BridgeState")
 					user.BridgeState.Send(status.BridgeState{StateEvent: status.StateBadCredentials, Message: "You have been logged out of Signal, please reconnect"})
 				}
+				user.bridge.Metrics.TrackConnectionState(user.SignalID, false)
 			}
 		}
 	}()
@@ -865,6 +871,7 @@ func (user *User) handleLoggedOut(ctx context.Context) {
 }
 
 func (user *User) handleLoggedOutNoLock(ctx context.Context) {
+	user.bridge.Metrics.TrackLoginState(user.SignalID, false)
 	id := user.SignalID
 	user.SignalID = uuid.Nil
 	user.SignalUsername = ""