Seen once in the ""/Topology/invalidate_server/pooled" test: if an application thread fails to open a new socket in _mongoc_cluster_stream_for_server_description, it calls mongoc_topology_invalidate_server, which locks the topology and calls mongoc_topology_description_invalidate_server.

However, the topology scanner thread might have detected the down server and removed its server description before the application thread got the topology lock. In that case we get a NULL server description and pass it to mongoc_topology_description_invalidate_server:

void
mongoc_topology_description_invalidate_server (mongoc_topology_description_t *topology,
                                               uint32_t                       id,
                                               const bson_error_t            *error)
{
   mongoc_server_description_t *sd;

   sd = mongoc_topology_description_server_by_id (topology, id, NULL);
   mongoc_topology_description_handle_ismaster (topology, sd, NULL, 0, NULL);
}