CDRIVER-2075 retry ismaster once

Server Discovery and Monitoring Spec: "Once a server is connected, the
client MUST change its type to Unknown only after it has retried the
server once."

Author: ajdavis

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

@@ -104,16 +104,15 @@ mongoc_topology_scanner_new (
 void
 mongoc_topology_scanner_destroy (mongoc_topology_scanner_t *ts);
 
-mongoc_topology_scanner_node_t *
+void
 mongoc_topology_scanner_add (mongoc_topology_scanner_t *ts,
                              const mongoc_host_list_t *host,
                              uint32_t id);
 
 void
-mongoc_topology_scanner_add_and_scan (mongoc_topology_scanner_t *ts,
-                                      const mongoc_host_list_t *host,
-                                      uint32_t id,
-                                      int64_t timeout_msec);
+mongoc_topology_scanner_scan (mongoc_topology_scanner_t *ts,
+                              uint32_t id,
+                              int64_t timeout_msec);
 
 void
 mongoc_topology_scanner_node_retire (mongoc_topology_scanner_node_t *node);

src/mongoc/mongoc-topology-scanner.c

@@ -217,7 +217,7 @@ mongoc_topology_scanner_destroy (mongoc_topology_scanner_t *ts)
    bson_free (ts);
 }
 
-mongoc_topology_scanner_node_t *
+void
 mongoc_topology_scanner_add (mongoc_topology_scanner_t *ts,
                              const mongoc_host_list_t *host,
                              uint32_t id)
@@ -234,29 +234,25 @@ mongoc_topology_scanner_add (mongoc_topology_scanner_t *ts,
    node->last_used = -1;
 
    DL_APPEND (ts->nodes, node);
-
-   return node;
 }
 
 void
-mongoc_topology_scanner_add_and_scan (mongoc_topology_scanner_t *ts,
-                                      const mongoc_host_list_t *host,
-                                      uint32_t id,
-                                      int64_t timeout_msec)
+mongoc_topology_scanner_scan (mongoc_topology_scanner_t *ts,
+                              uint32_t id,
+                              int64_t timeout_msec)
 {
    mongoc_topology_scanner_node_t *node;
 
    BSON_ASSERT (timeout_msec < INT32_MAX);
 
-   node = mongoc_topology_scanner_add (ts, host, id);
+   node = mongoc_topology_scanner_get_node (ts, id);
 
    /* begin non-blocking connection, don't wait for success */
    if (node && mongoc_topology_scanner_node_setup (node, &node->last_error)) {
       _begin_ismaster_cmd (ts, node, timeout_msec);
    }
 
    /* if setup fails the node stays in the scanner. destroyed after the scan. */
-   return;
 }
 
 void

src/mongoc/mongoc-topology.c

@@ -46,8 +46,9 @@ _mongoc_topology_reconcile_add_nodes (void *item, void *ctx)
     * this scan. */
    if (!mongoc_topology_scanner_get_node (scanner, sd->id) &&
        !mongoc_topology_scanner_has_node_for_host (scanner, &sd->host)) {
-      mongoc_topology_scanner_add_and_scan (
-         scanner, &sd->host, sd->id, topology->connect_timeout_msec);
+      mongoc_topology_scanner_add (scanner, &sd->host, sd->id);
+      mongoc_topology_scanner_scan (
+         scanner, sd->id, topology->connect_timeout_msec);
    }
 
    return true;
@@ -151,16 +152,34 @@ _mongoc_topology_scanner_cb (uint32_t id,
                              const bson_error_t *error /* IN */)
 {
    mongoc_topology_t *topology;
+   mongoc_server_description_t *sd;
 
    BSON_ASSERT (data);
 
    topology = (mongoc_topology_t *) data;
 
    mongoc_mutex_lock (&topology->mutex);
-   _mongoc_topology_update_no_lock (
-      id, ismaster_response, rtt_msec, topology, error);
+   sd = mongoc_topology_description_server_by_id (
+      &topology->description, id, NULL);
+
+   /* Server Discovery and Monitoring Spec: "Once a server is connected, the
+    * client MUST change its type to Unknown only after it has retried the
+    * server once." */
+   if (!ismaster_response && sd && sd->type != MONGOC_SERVER_UNKNOWN) {
+      _mongoc_topology_update_no_lock (
+         id, ismaster_response, rtt_msec, topology, error);
+
+      /* add another ismaster call to the current scan - the scan continues
+       * until all commands are done */
+      mongoc_topology_scanner_scan (
+         topology->scanner, sd->id, topology->connect_timeout_msec);
+   } else {
+      _mongoc_topology_update_no_lock (
+         id, ismaster_response, rtt_msec, topology, error);
+
+      mongoc_cond_broadcast (&topology->cond_client);
+   }
 
-   mongoc_cond_broadcast (&topology->cond_client);
    mongoc_mutex_unlock (&topology->mutex);
 }
 

tests/TestSuite.h

@@ -444,6 +444,23 @@ test_error (const char *format, ...) BSON_GNUC_PRINTF (1, 2);
       }                                                                  \
    } while (0)
 
+#define WAIT_UNTIL(_pred)                                              \
+   do {                                                                \
+      int64_t _start = bson_get_monotonic_time ();                     \
+      while (!(_pred)) {                                               \
+         if (bson_get_monotonic_time () - _start > 10 * 1000 * 1000) { \
+            fprintf (stderr,                                           \
+                     "Predicate \"%s\" timed out\n"                    \
+                     "   %s:%d  %s()\n",                               \
+                     #_pred,                                           \
+                     __FILE__,                                         \
+                     __LINE__,                                         \
+                     BSON_FUNC);                                       \
+            abort ();                                                  \
+         }                                                             \
+      }                                                                \
+   } while (0)
+
 #define MAX_TEST_NAME_LENGTH 500
 
 

tests/test-mongoc-client.c

@@ -2689,6 +2689,7 @@ _test_client_sends_handshake (bool pooled)
    mock_server_run (server);
    uri = mongoc_uri_copy (mock_server_get_uri (server));
    mongoc_uri_set_option_as_int32 (uri, "heartbeatFrequencyMS", heartbeat_ms);
+   mongoc_uri_set_option_as_int32 (uri, "connectTimeoutMS", 100);
 
    if (pooled) {
       pool = mongoc_client_pool_new (uri);
@@ -2730,7 +2731,12 @@ _test_client_sends_handshake (bool pooled)
     * time the server hangs up */
    request = mock_server_receives_ismaster (server);
    _assert_ismaster_valid (request, false);
+   mock_server_hangs_up (request);
+   request_destroy (request);
 
+   /* Client retries once (CDRIVER-2075) */
+   request = mock_server_receives_ismaster (server);
+   _assert_ismaster_valid (request, true);
    mock_server_hangs_up (request);
    request_destroy (request);