Uploaded image for project: 'C Driver'
  1. C Driver
  2. CDRIVER-4622

OP_QUERY egress counter may overcount

    • Type: Icon: Bug Bug
    • Resolution: Fixed
    • Priority: Icon: Minor - P4 Minor - P4
    • 1.24.0
    • Affects Version/s: None
    • Component/s: None
    • Labels:
      None

      Proposal

      Only increment egress OP_QUERY counter for each command sent. Do not increment egress OP_QUERY counter for each resolved DNS record. This may result in overcounting.

      Background

      _mongoc_rpc_gather increments the egress OP_QUERY counter

      _mongoc_rpc_gather may be called when constructing hello commands for each DNS record in mongoc_topology_scanner_node_setup_tcp:

      LL_FOREACH2 (node->dns_results, iter, ai_next)
      {
          _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;
      }
      

      _mongoc_rpc_gather is called through this call path:

      test-libmongoc!_mongoc_rpc_gather (/Users/kevin.albertson/review/mongo-c-driver-1242/src/libmongoc/src/mongoc/mongoc-rpc.c:606)
      test-libmongoc!_mongoc_async_cmd_init_send (/Users/kevin.albertson/review/mongo-c-driver-1242/src/libmongoc/src/mongoc/mongoc-async-cmd.c:168)
      test-libmongoc!mongoc_async_cmd_new (/Users/kevin.albertson/review/mongo-c-driver-1242/src/libmongoc/src/mongoc/mongoc-async-cmd.c:232)
      test-libmongoc!_begin_hello_cmd (/Users/kevin.albertson/review/mongo-c-driver-1242/src/libmongoc/src/mongoc/mongoc-topology-scanner.c:431)
      test-libmongoc!mongoc_topology_scanner_node_setup_tcp (/Users/kevin.albertson/review/mongo-c-driver-1242/src/libmongoc/src/mongoc/mongoc-topology-scanner.c:975)
      

      The egress OP_QUERY counter is documented as "The number of sent Query operations."

      The expected behavior is to only increment egress OP_QUERY when an OP_QUERY is sent, not just constructed.

      Here is a test that shows the issue:

      static void
      test_overcounting_opquery (void *unused)
      {
         BSON_UNUSED (unused);
      
         bson_error_t error = {0};
         // Connect to exactly one host that resolves to both IPv4 and IPv6.
         mongoc_client_t *const client = mongoc_client_new ("mongodb://ipv4_and_ipv6.test.build.10gen.cc:27017");
      
         mongoc_counter_op_egress_query_reset ();
      
         ASSERT_OR_PRINT (
            mongoc_client_get_server_status (client, NULL, NULL, &error), error);
      
         const int32_t sent_queries = mongoc_counter_op_egress_query_count ();
         // Expect one OP_QUERY for handshake.
         ASSERT_WITH_MSG (
            sent_queries == 1,
            "expected exactly one OP_QUERY requests, but observed %" PRId32
            " requests",
            sent_queries);
         // May fail with: 'expected exactly one OP_QUERY requests, but observed 2
         // requests' if 'localhost' resolves to more than one DNS record.
         mongoc_client_destroy (client);
      }
      

      For assistance with testing:

      • localhost.test.build.10gen.cc only resolves to IPv4 127.0.0.1
      • ipv4_and_ipv6.test.build.10gen.cc resolves to both IPv4 12 127.0.0.1 and IPv6 ::1

            Assignee:
            ezra.chung@mongodb.com Ezra Chung
            Reporter:
            kevin.albertson@mongodb.com Kevin Albertson
            Votes:
            0 Vote for this issue
            Watchers:
            3 Start watching this issue

              Created:
              Updated:
              Resolved: