SERVER-103650: Add FTDC metrics for reconnect attempts in DBClientSession (#51054)

GitOrigin-RevId: 8fec0f1ac9d74c783c0645a0254f37607cf9ac09

jstests/replsets/dbclient_session_reconnect_metrics.js

@@ -0,0 +1,38 @@
+/**
+ * Test that reconnect metrics for DBClientSession exist and are appropriately incremented.
+ *
+ * This test cannot be run on older binaries because the metrics do not exist yet.
+ * * @tags: [multiversion_incompatible]
+ */
+
+import {ReplSetTest} from "jstests/libs/replsettest.js";
+
+function getMetrics(conn) {
+    const status = assert.commandWorked(conn.adminCommand({serverStatus: 1}));
+    return Object.fromEntries([
+        'dbClientSessionReconnectAttempts',
+        'dbClientSessionWithoutAutoReconnectFailures',
+    ].map(name => [name, status.metrics.network[name]]));
+}
+
+const rst = new ReplSetTest(
+    {name: 'testSet', useBridge: true, nodes: 3, settings: {chainingAllowed: false}});
+const nodes = rst.startSet();
+rst.initiate();
+const primary = rst.getPrimary();
+rst.awaitSecondaryNodes();
+const secondary = rst.getSecondaries()[0];
+
+const before_disconnect = getMetrics(secondary);
+// dbClientSessionWithoutAutoReconnectFailures should never be incremented because the connection
+// created by the oplog fetcher is always created with autoReconnect = true.
+assert.eq(before_disconnect.dbClientSessionWithoutAutoReconnectFailures, 0);
+
+secondary.disconnect(primary);
+assert.soon(() => {
+    const after_disconnect = getMetrics(secondary);
+    return after_disconnect.dbClientSessionReconnectAttempts >
+        before_disconnect.dbClientSessionReconnectAttempts;
+});
+
+rst.stopSet();

src/mongo/client/dbclient_session.cpp

@@ -88,6 +88,16 @@
 
 
 namespace mongo {
+namespace {
+// Tracks the number of times DbClientSession detects that the connection has been broken in
+// ensureConnection() and attempts to reconnect.
+auto& dbClientSessionReconnectAttempts =
+    *MetricBuilder<Counter64>("network.dbClientSessionReconnectAttempts");
+// Tracks the number of times DbClientSession detects that the connection has been broken in
+// ensureConnection() with autoReconnect = false so does not try to reconnect.
+auto& dbClientSessionWithoutAutoReconnectFailures =
+    *MetricBuilder<Counter64>("network.dbClientSessionWithoutAutoReconnectFailures");
+}  // namespace
 
 using std::string;
 
@@ -579,4 +589,16 @@ bool DBClientSession::isTLS() {
 }
 #endif
 
+void DBClientSession::ensureConnection() {
+    if (!_failed.load()) {
+        return;
+    }
+    if (!_autoReconnect) {
+        dbClientSessionWithoutAutoReconnectFailures.increment();
+        throwSocketError(SocketErrorKind::FAILED_STATE, toString());
+    }
+    dbClientSessionReconnectAttempts.increment();
+    _reconnectSession();
+}
+
 }  // namespace mongo

src/mongo/client/dbclient_session.h

@@ -213,14 +213,7 @@ public:
     }
 
     // Throws a NetworkException if in failed state and not reconnecting or if waiting to reconnect.
-    void ensureConnection() override {
-        if (_failed.load()) {
-            if (!_autoReconnect) {
-                throwSocketError(SocketErrorKind::FAILED_STATE, toString());
-            }
-            _reconnectSession();
-        }
-    }
+    void ensureConnection() override;
 
     bool isReplicaSetMember() const override {
         return _isReplicaSetMember;