CBL-8076: Do not retry replicator connections if WebSocket being clos… (#2479)

The Abnormal Close occurs after the client is already decided to close the socket but fails to receive response from the server. We don't want to retry replicating on this condition.

We fix it by a change in WebSocket, that it now treats a connection closure that is  not fully complete only due to a pending final server acknowledgement (i.e. CLOSE frame was already sent).

Author: jianminzhao

Networking/WebSockets/WebSocketImpl.cc

@@ -534,7 +534,8 @@ namespace litecore::websocket {
 
                     if ( clean ) {
                         status.reason = kWebSocketClose;
-                        if ( !expected ) status.code = kCodeAbnormal;
+                        // if (!expected) then _closeSent => !_closeReceived
+                        if ( !expected ) status.code = _closeSent ? kCodeNormal : kCodeAbnormal;
                         else if ( !_closeMessage )
                             status.code = kCodeNormal;
                         else {

Replicator/tests/ReplicatorAPITest.cc

@@ -816,6 +816,86 @@ TEST_CASE_METHOD(ReplicatorAPITest, "Stop after transient connect failure", "[C]
     waitForStatus(kC4Stopped);
 }
 
+// CBL-8074
+TEST_CASE_METHOD(ReplicatorAPITest, "WebSocket Peer Going Away", "[C][Push][Pull]") {
+    bool            afterClose = false;
+    C4SocketFactory factory    = {};
+    C4Socket*       c4socket   = nullptr;
+    factory.context            = &c4socket;
+    factory.open               = [](C4Socket* socket, const C4Address* addr, C4Slice options, void* context) {
+        c4socket_opened(socket);
+        *(C4Socket**)context = socket;
+    };
+
+    factory.close = [](C4Socket* socket) {
+        // Not invoked
+        REQUIRE(false);
+    };
+
+    // "peer going away" before CLOSE is sent
+    // Replicator receives error code 1006, which is transient.
+    // C4Replicator goes to offline and waiting for retry.
+    SECTION("CLOSE Not Sent") {
+        afterClose    = false;
+        _mayGoOffline = true;
+        factory.write = [](C4Socket* socket, C4SliceResult msg) {
+            // Simulate Peer-Going-Away before Replicator calling Stop.
+            // Socket is closed unexpectedly, without the client sending CLOSE
+            FLSliceResult_Release(msg);
+            c4socket_closed(socket, {WebSocketDomain, websocket::kCodeGoingAway});
+        };
+    }
+
+    // "peer going away" after CLOSE frame was already sent
+    // Since the replicator is already stopped when the peer goes away, WebSocket will
+    // treat it as Normal Close.
+    SECTION("CLOSE Has Been Sent") {
+        afterClose    = true;
+        _mayGoOffline = false;
+        factory.write = [](C4Socket* socket, C4SliceResult msg) {
+            // Do nothing
+            FLSliceResult_Release(msg);
+        };
+    }
+
+    _socketFactory = &factory;
+    C4Error err;
+    importJSONLines(sFixturesDir + "names_100.json");
+
+    if ( !afterClose ) {
+        // WebSocket code 1006, transient error
+        REQUIRE(startReplicator(kC4Disabled, kC4OneShot, WITH_ERROR(&err)));
+        _numCallbacksWithLevel[kC4Offline] = 0;
+        waitForStatus(kC4Offline);
+    } else {
+        REQUIRE(startReplicator(kC4Disabled, kC4Continuous, WITH_ERROR(&err)));
+        // Making sure the WebSocket is open/connected
+        waitForStatus(kC4Busy);
+    }
+
+    c4repl_stop(_repl);
+
+    if ( afterClose ) {
+        // Give some time for Replicator::_stop to be called, but before timeout in WebSocketImpl
+        // to not get Timeout error.
+        std::this_thread::sleep_for(1s);
+        // WebSocket will treat it as Normal Close
+        c4socket_closed(c4socket, {WebSocketDomain, websocket::kCodeGoingAway});
+    }
+
+    waitForStatus(kC4Stopped);
+
+    auto status = c4repl_getStatus(_repl);
+
+    if ( !afterClose ) {
+        // kCodeAbnormal == 1006
+        CHECK((status.error.domain == WebSocketDomain && status.error.code == websocket::kCodeAbnormal));
+    } else {
+        // "peer going away" after stop results in normal Stop.
+        CHECK(status.error.code == 0);
+    }
+}
+
 TEST_CASE_METHOD(ReplicatorAPITest, "Calling c4socket_ method after STOP", "[C][Push][Pull]") {
     // c.f. the flow with test case "Stop after transient connect failure"
     _mayGoOffline            = true;