CDRIVER-696: Aggregate the scanner node errors when Server Selection fails

Author: bjori

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

@@ -113,6 +113,10 @@ bool
 mongoc_topology_scanner_work (mongoc_topology_scanner_t *ts,
                               int32_t                    timeout_msec);
 
+void
+mongoc_topology_scanner_sum_errors (mongoc_topology_scanner_t *ts,
+                                    bson_error_t              *error);
+
 void
 mongoc_topology_scanner_reset (mongoc_topology_scanner_t *ts);
 

src/mongoc/mongoc-topology-scanner.c

@@ -535,6 +535,45 @@ mongoc_topology_scanner_work (mongoc_topology_scanner_t *ts,
    return r;
 }
 
+/*
+ *--------------------------------------------------------------------------
+ *
+ * mongoc_topology_scanner_sum_errors --
+ *
+ *      Summarizes all scanner node errors into one error message
+ *
+ *--------------------------------------------------------------------------
+ */
+
+void
+mongoc_topology_scanner_sum_errors (mongoc_topology_scanner_t *ts,
+                                    bson_error_t              *error)
+{
+   mongoc_topology_scanner_node_t *node, *tmp;
+
+   DL_FOREACH_SAFE (ts->nodes, node, tmp) {
+      if (node->last_error.code) {
+         char *msg = NULL;
+
+         if (error->code) {
+            msg = bson_strdup(error->message);
+         }
+
+         bson_set_error(error,
+                        MONGOC_ERROR_SERVER_SELECTION,
+                        MONGOC_ERROR_SERVER_SELECTION_FAILURE,
+                        "%s[%s] ",
+                        msg ? msg : "", node->last_error.message);
+         if (msg) {
+            bson_free (msg);
+         }
+      }
+   }
+   if (error->code) {
+      error->message[strlen(error->message)-1] = '\0';
+   }
+}
+
 /*
  *--------------------------------------------------------------------------
  *
@@ -557,3 +596,4 @@ mongoc_topology_scanner_reset (mongoc_topology_scanner_t *ts)
       }
    }
 }
+

src/mongoc/mongoc-topology.c

@@ -321,14 +321,16 @@ _mongoc_topology_run_scanner (mongoc_topology_t *topology,
  *--------------------------------------------------------------------------
  */
 static void
-_mongoc_topology_do_blocking_scan (mongoc_topology_t *topology) {
+_mongoc_topology_do_blocking_scan (mongoc_topology_t *topology, bson_error_t *error) {
    mongoc_topology_scanner_start (topology->scanner,
                                   topology->connect_timeout_msec,
                                   true);
 
    while (_mongoc_topology_run_scanner (topology,
                                         topology->connect_timeout_msec)) {}
 
+   /* Aggregate all scanner errors, if any */
+   mongoc_topology_scanner_sum_errors (topology->scanner, error);
    /* "retired" nodes can be checked again in the next scan */
    mongoc_topology_scanner_reset (topology->scanner);
    topology->last_scan = bson_get_monotonic_time ();
@@ -378,6 +380,7 @@ mongoc_topology_select (mongoc_topology_t         *topology,
    bool try_once;
    int64_t sleep_usec;
    bool tried_once;
+   bson_error_t scanner_error = { 0 };
 
    /* These names come from the Server Selection Spec pseudocode */
    int64_t loop_start;  /* when we entered this function */
@@ -424,7 +427,7 @@ mongoc_topology_select (mongoc_topology_t         *topology,
             }
 
             /* takes up to connectTimeoutMS. sets "last_scan", clears "stale" */
-            _mongoc_topology_do_blocking_scan (topology);
+            _mongoc_topology_do_blocking_scan (topology, &scanner_error);
             tried_once = true;
          }
 
@@ -441,11 +444,19 @@ mongoc_topology_select (mongoc_topology_t         *topology,
 
          if (try_once) {
             if (tried_once) {
-               bson_set_error(error,
-                              MONGOC_ERROR_SERVER_SELECTION,
-                              MONGOC_ERROR_SERVER_SELECTION_FAILURE,
-                              "No suitable servers found: "
-                              "`serverselectiontryonce` set");
+               if (scanner_error.code) {
+                  bson_set_error(error,
+                                 MONGOC_ERROR_SERVER_SELECTION,
+                                 MONGOC_ERROR_SERVER_SELECTION_FAILURE,
+                                 "No suitable servers found "
+                                 "(`serverselectiontryonce` set): %s", scanner_error.message);
+               } else {
+                  bson_set_error(error,
+                                 MONGOC_ERROR_SERVER_SELECTION,
+                                 MONGOC_ERROR_SERVER_SELECTION_FAILURE,
+                                 "No suitable servers found "
+                                 "(`serverselectiontryonce` set)");
+               }
                goto FAIL;
             }
          } else {

tests/test-mongoc-topology.c

@@ -198,7 +198,7 @@ _test_server_selection (bool try_once)
    ASSERT_CMPINT (error.code, ==, MONGOC_ERROR_SERVER_SELECTION_FAILURE);
 
    if (try_once) {
-      ASSERT_CMPSTR ("No suitable servers found: `serverselectiontryonce` set", error.message);
+      ASSERT_CMPSTR ("No suitable servers found (`serverselectiontryonce` set)", error.message);
    } else {
       ASSERT_CMPSTR ("No suitable servers found: `minheartbeatfrequencyms` not reached yet", error.message);
    }