feat: Add metrics for colibri-ws closing and receiving an error. (#2128)

jitsi · Apr 29, 2024 · 748a626 · 748a626
1 parent 462b7f6
commit 748a626
Show file tree

Hide file tree

Showing 3 changed files with 39 additions and 2 deletions.
diff --git a/jvb/src/main/java/org/jitsi/videobridge/EndpointMessageTransport.java b/jvb/src/main/java/org/jitsi/videobridge/EndpointMessageTransport.java
@@ -353,9 +353,18 @@ public void webSocketClosed(ColibriWebSocket ws, int statusCode, String reason)
                 webSocket = null;
                 webSocketLastActive = false;
                 getLogger().info(() -> "Websocket closed, statusCode " + statusCode + " ( " + reason + ").");
+                // 1000 is normal, 1001 is e.g. a tab closing. 1005 is "No Status Rcvd" and we see the majority of
+                // sockets close this way.
+                if (statusCode == 1000 || statusCode == 1001 || statusCode == 1005)
+                {
+                    VideobridgeMetrics.colibriWebSocketCloseNormal.inc();
+                }
+                else
+                {
+                    VideobridgeMetrics.colibriWebSocketCloseAbnormal.inc();
+                }
             }
         }
-
     }
 
     /**
@@ -365,6 +374,7 @@ public void webSocketClosed(ColibriWebSocket ws, int statusCode, String reason)
     public void webSocketError(ColibriWebSocket ws, Throwable cause)
     {
         getLogger().error("Colibri websocket error: " +  cause.getMessage());
+        VideobridgeMetrics.colibriWebSocketErrors.inc();
     }
 
     /**

diff --git a/jvb/src/main/kotlin/org/jitsi/videobridge/metrics/VideobridgeMetrics.kt b/jvb/src/main/kotlin/org/jitsi/videobridge/metrics/VideobridgeMetrics.kt
@@ -72,6 +72,24 @@ object VideobridgeMetrics {
         "Number of messages sent via the data channels of the endpoints of this conference."
     )
 
+    @JvmField
+    val colibriWebSocketCloseNormal = metricsContainer.registerCounter(
+        "colibri_web_socket_close_normal",
+        "Number of times a colibri web socket was closed normally."
+    )
+
+    @JvmField
+    val colibriWebSocketCloseAbnormal = metricsContainer.registerCounter(
+        "colibri_web_socket_close_abnormal",
+        "Number of times a colibri web socket was closed abnormally."
+    )
+
+    @JvmField
+    val colibriWebSocketErrors = metricsContainer.registerCounter(
+        "colibri_web_socket_error",
+        "Number of times a colibri web socket reported an error."
+    )
+
     @JvmField
     val packetsReceived = metricsContainer.registerCounter(
         "packets_received",

diff --git a/jvb/src/main/kotlin/org/jitsi/videobridge/relay/RelayMessageTransport.kt b/jvb/src/main/kotlin/org/jitsi/videobridge/relay/RelayMessageTransport.kt
@@ -340,6 +340,13 @@ class RelayMessageTransport(
                 webSocket = null
                 webSocketLastActive = false
                 logger.debug { "Web socket closed, statusCode $statusCode ( $reason)." }
+                // 1000 is normal, 1001 is e.g. a tab closing. 1005 is "No Status Rcvd" and we see the majority of
+                // sockets close this way.
+                if (statusCode == 1000 || statusCode == 1001 || statusCode == 1005) {
+                    VideobridgeMetrics.colibriWebSocketCloseNormal.inc()
+                } else {
+                    VideobridgeMetrics.colibriWebSocketCloseAbnormal.inc()
+                }
             }
         }
 
@@ -350,8 +357,10 @@ class RelayMessageTransport(
         }
     }
 
-    override fun webSocketError(ws: ColibriWebSocket, cause: Throwable) =
+    override fun webSocketError(ws: ColibriWebSocket, cause: Throwable) {
         logger.error("Colibri websocket error: ${cause.message}")
+        VideobridgeMetrics.colibriWebSocketErrors.inc()
+    }
 
     /**
      * {@inheritDoc}