server/ctdb_keepalive.c

   1 /*
   2    monitoring links to all other nodes to detect dead nodes
   3
   4
   5    Copyright (C) Ronnie Sahlberg 2007
   6
   7    This program is free software; you can redistribute it and/or modify
   8    it under the terms of the GNU General Public License as published by
   9    the Free Software Foundation; either version 3 of the License, or
  10    (at your option) any later version.
  11
  12    This program is distributed in the hope that it will be useful,
  13    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15    GNU General Public License for more details.
  16
  17    You should have received a copy of the GNU General Public License
  18    along with this program; if not, see <http://www.gnu.org/licenses/>.
  19 */
  20
  21 #include "includes.h"
  22 #include "lib/events/events.h"
  23 #include "system/filesys.h"
  24 #include "system/wait.h"
  25 #include "include/ctdb_protocol.h"
  26 #include "include/ctdb_private.h"
  27
  28
  29 /*
  30   see if any nodes are dead
  31  */
  32 static void ctdb_check_for_dead_nodes(struct event_context *ev, struct timed_event *te,
  33                                       struct timeval t, void *private_data)
  34 {
  35         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
  36         int i;
  37
  38         /* send a keepalive to all other nodes, unless */
  39         for (i=0;i<ctdb->num_nodes;i++) {
  40                 struct ctdb_node *node = ctdb->nodes[i];
  41
  42                 if (node->flags & NODE_FLAGS_DELETED) {
  43                         continue;
  44                 }
  45
  46                 if (node->pnn == ctdb->pnn) {
  47                         continue;
  48                 }
  49
  50                 if (node->flags & NODE_FLAGS_DISCONNECTED) {
  51                         /* it might have come alive again */
  52                         if (node->rx_cnt != 0) {
  53                                 ctdb_node_connected(node);
  54                         }
  55                         continue;
  56                 }
  57
  58
  59                 if (node->rx_cnt == 0) {
  60                         node->dead_count++;
  61                 } else {
  62                         node->dead_count = 0;
  63                 }
  64
  65                 node->rx_cnt = 0;
  66
  67                 if (node->dead_count >= ctdb->tunable.keepalive_limit) {
  68                         DEBUG(DEBUG_NOTICE,("dead count reached for node %u\n", node->pnn));
  69                         ctdb_node_dead(node);
  70                         ctdb_send_keepalive(ctdb, node->pnn);
  71                         /* maybe tell the transport layer to kill the
  72                            sockets as well?
  73                         */
  74                         continue;
  75                 }
  76
  77                 DEBUG(DEBUG_DEBUG,("sending keepalive to %u\n", node->pnn));
  78                 ctdb_send_keepalive(ctdb, node->pnn);
  79
  80                 node->tx_cnt = 0;
  81         }
  82
  83         event_add_timed(ctdb->ev, ctdb->keepalive_ctx,
  84                         timeval_current_ofs(ctdb->tunable.keepalive_interval, 0),
  85                         ctdb_check_for_dead_nodes, ctdb);
  86 }
  87
  88
  89 void ctdb_start_keepalive(struct ctdb_context *ctdb)
  90 {
  91         struct timed_event *te;
  92
  93         ctdb->keepalive_ctx = talloc_new(ctdb);
  94         CTDB_NO_MEMORY_FATAL(ctdb, ctdb->keepalive_ctx);
  95
  96         te = event_add_timed(ctdb->ev, ctdb->keepalive_ctx,
  97                              timeval_current_ofs(ctdb->tunable.keepalive_interval, 0),
  98                              ctdb_check_for_dead_nodes, ctdb);
  99         CTDB_NO_MEMORY_FATAL(ctdb, te);
 100
 101         DEBUG(DEBUG_NOTICE,("Keepalive monitoring has been started\n"));
 102 }
 103
 104 void ctdb_stop_keepalive(struct ctdb_context *ctdb)
 105 {
 106         talloc_free(ctdb->keepalive_ctx);
 107         ctdb->keepalive_ctx = NULL;
 108 }
 109