CDRIVER-4207 handshake on new stream in scanner

src/libmongoc/src/mongoc/mongoc-topology-scanner-private.h

@@ -54,6 +54,9 @@ typedef struct mongoc_topology_scanner_node {
    mongoc_stream_t *stream;
 
    int64_t last_used;
+   /* last_failed is set upon a network error trying to check a server.
+    * last_failed is used to enforce cooldownMS.
+    * last_failed is not set upon a network error during an application operation on @stream. */
    int64_t last_failed;
    bool has_auth;
    bool hello_ok;

src/libmongoc/src/mongoc/mongoc-topology-scanner.c

@@ -374,12 +374,13 @@ _begin_hello_cmd (mongoc_topology_scanner_node_t *node,
                   mongoc_stream_t *stream,
                   bool is_setup_done,
                   struct addrinfo *dns_result,
-                  int64_t initiate_delay_ms)
+                  int64_t initiate_delay_ms,
+                  bool use_handshake)
 {
    mongoc_topology_scanner_t *ts = node->ts;
    bson_t cmd;
 
-   if (node->last_used != -1 && node->last_failed == -1) {
+   if (node->last_used != -1 && node->last_failed == -1 && !use_handshake) {
       /* The node's been used before and not failed recently */
       bson_copy_to (
          _mongoc_topology_scanner_get_monitoring_cmd (ts, node->hello_ok),
@@ -946,11 +947,21 @@ mongoc_topology_scanner_node_setup_tcp (mongoc_topology_scanner_node_t *node,
    }
 
    if (node->successful_dns_result) {
-      _begin_hello_cmd (node, NULL, false, node->successful_dns_result, 0);
+      _begin_hello_cmd (node,
+                        NULL /* stream */,
+                        false /* is_setup_done */,
+                        node->successful_dns_result,
+                        0 /* initiate_delay_ms */,
+                        true /* use_handshake */);
    } else {
       LL_FOREACH2 (node->dns_results, iter, ai_next)
       {
-         _begin_hello_cmd (node, NULL, false, iter, delay);
+         _begin_hello_cmd (node,
+                           NULL /* stream */,
+                           false /* is_setup_done */,
+                           iter,
+                           delay,
+                           true /* use_handshake */);
          /* each subsequent DNS result will have an additional 250ms delay. */
          delay += HAPPY_EYEBALLS_DELAY_MS;
       }
@@ -1017,7 +1028,8 @@ mongoc_topology_scanner_node_connect_unix (mongoc_topology_scanner_node_t *node,
                         stream,
                         false /* is_setup_done */,
                         NULL /* dns result */,
-                        0 /* delay */);
+                        0 /* delay */,
+                        true /* use_handshake */);
       RETURN (true);
    }
    bson_set_error (error,
@@ -1055,7 +1067,12 @@ mongoc_topology_scanner_node_setup (mongoc_topology_scanner_node_t *node,
 
    /* if there is already a working stream, push it back to be re-scanned. */
    if (node->stream) {
-      _begin_hello_cmd (node, node->stream, true /* is_setup_done */, NULL, 0);
+      _begin_hello_cmd (node,
+                        node->stream,
+                        true /* is_setup_done */,
+                        NULL /* dns_result */,
+                        0 /* initiate_delay_ms */,
+                        false /* use_handshake */);
       node->stream = NULL;
       return;
    }
@@ -1067,7 +1084,12 @@ mongoc_topology_scanner_node_setup (mongoc_topology_scanner_node_t *node,
          node->ts->uri, &node->host, node->ts->initiator_context, error);
       if (stream) {
          success = true;
-         _begin_hello_cmd (node, stream, false, NULL, 0);
+         _begin_hello_cmd (node,
+                           stream,
+                           false /* is_setup_done */,
+                           NULL /* dns_result */,
+                           0 /* initiate_delay_ms */,
+                           true /* use_handshake */);
       }
    } else {
       if (node->host.family == AF_UNIX) {

src/libmongoc/tests/test-mongoc-client.c

@@ -4145,6 +4145,65 @@ test_mongoc_client_get_handshake_establishes_connection_pooled (void)
    mongoc_client_pool_destroy (pool);
 }
 
+/* Regression test for CDRIVER-4207. */
+void
+test_mongoc_client_resends_handshake_on_network_error (void)
+{
+   mongoc_client_t *client;
+   mock_server_t *server;
+   future_t *future;
+   request_t *request;
+   bson_error_t error;
+   bson_t *ping = tmp_bson ("{'ping': 1}");
+
+   server = mock_server_new ();
+   mock_server_run (server);
+   client = mongoc_client_new_from_uri (mock_server_get_uri (server));
+   mongoc_client_set_appname (client, "foo");
+
+   /* Send a "ping" command. */
+   future = future_client_command_simple (
+      client, "db", ping, NULL /* read_prefs */, NULL /* reply */, &error);
+   /* The first command on the new connection is handshake. It uses the legacy
+    * hello and includes the client.application.name. */
+   request = mock_server_receives_legacy_hello (
+      server,
+      "{'" HANDSHAKE_CMD_LEGACY_HELLO
+      "': 1, 'client': {'application': {'name': 'foo'}}}");
+   mock_server_replies_simple (request, "{'ok': 1, 'maxWireVersion': 14 }");
+   request_destroy (request);
+   request = mock_server_receives_msg (
+      server, MONGOC_QUERY_NONE, tmp_bson ("{'ping': 1}"));
+   mock_server_hangs_up (request);
+   future_wait (future);
+   future_destroy (future);
+   ASSERT_ERROR_CONTAINS (error,
+                          MONGOC_ERROR_STREAM,
+                          MONGOC_ERROR_STREAM_SOCKET,
+                          "socket error or timeout");
+   request_destroy (request);
+
+   /* Send another "ping" command. */
+   future = future_client_command_simple (
+      client, "db", ping, NULL /* read_prefs */, NULL /* reply */, &error);
+   /* Expect the new connection to send the full handshake. */
+   request = mock_server_receives_legacy_hello (
+      server,
+      "{'" HANDSHAKE_CMD_LEGACY_HELLO
+      "': 1, 'client': {'application': {'name': 'foo'}}}");
+   mock_server_replies_simple (request, "{'ok': 1, 'maxWireVersion': 14 }");
+   request_destroy (request);
+
+   request = mock_server_receives_msg (
+      server, MONGOC_QUERY_NONE, tmp_bson ("{'ping': 1}"));
+   mock_server_replies_ok_and_destroys (request);
+   ASSERT (future_get_bool (future));
+   future_destroy (future);
+
+   mongoc_client_destroy (client);
+   mock_server_destroy (server);
+}
+
 void
 test_client_install (TestSuite *suite)
 {
@@ -4452,4 +4511,8 @@ test_client_install (TestSuite *suite)
       suite,
       "/Client/get_handshake_establishes_connection/pooled",
       test_mongoc_client_get_handshake_establishes_connection_pooled);
+   TestSuite_AddMockServerTest (
+      suite,
+      "/Client/resends_handshake_on_network_error",
+      test_mongoc_client_resends_handshake_on_network_error);
 }