ctdb/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_private.h"
  26
  27
  28 /*
  29   see if any nodes are dead
  30  */
  31 static void ctdb_check_for_dead_nodes(struct event_context *ev, struct timed_event *te,
  32                                       struct timeval t, void *private_data)
  33 {
  34         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
  35         int i;
  36
  37         /* send a keepalive to all other nodes, unless */
  38         for (i=0;i<ctdb->num_nodes;i++) {
  39                 struct ctdb_node *node = ctdb->nodes[i];
  40                 if (node->pnn == ctdb->pnn) {
  41                         continue;
  42                 }
  43
  44                 if (node->flags & NODE_FLAGS_DISCONNECTED) {
  45                         /* it might have come alive again */
  46                         if (node->rx_cnt != 0) {
  47                                 ctdb_node_connected(node);
  48                         }
  49                         continue;
  50                 }
  51
  52
  53                 if (node->rx_cnt == 0) {
  54                         node->dead_count++;
  55                 } else {
  56                         node->dead_count = 0;
  57                 }
  58
  59                 node->rx_cnt = 0;
  60
  61                 if (node->dead_count >= ctdb->tunable.keepalive_limit) {
  62                         DEBUG(DEBUG_NOTICE,("dead count reached for node %u\n", node->pnn));
  63                         ctdb_node_dead(node);
  64                         ctdb_send_keepalive(ctdb, node->pnn);
  65                         /* maybe tell the transport layer to kill the
  66                            sockets as well?
  67                         */
  68                         continue;
  69                 }
  70
  71                 if (node->tx_cnt == 0) {
  72                         DEBUG(DEBUG_DEBUG,("sending keepalive to %u\n", node->pnn));
  73                         ctdb_send_keepalive(ctdb, node->pnn);
  74                 }
  75
  76                 node->tx_cnt = 0;
  77         }
  78
  79         event_add_timed(ctdb->ev, ctdb->keepalive_ctx,
  80                         timeval_current_ofs(ctdb->tunable.keepalive_interval, 0),
  81                         ctdb_check_for_dead_nodes, ctdb);
  82 }
  83
  84
  85 void ctdb_start_keepalive(struct ctdb_context *ctdb)
  86 {
  87         struct timed_event *te;
  88
  89         ctdb->keepalive_ctx = talloc_new(ctdb);
  90         CTDB_NO_MEMORY_FATAL(ctdb, ctdb->keepalive_ctx);
  91
  92         te = event_add_timed(ctdb->ev, ctdb->keepalive_ctx,
  93                              timeval_current_ofs(ctdb->tunable.keepalive_interval, 0),
  94                              ctdb_check_for_dead_nodes, ctdb);
  95         CTDB_NO_MEMORY_FATAL(ctdb, te);
  96
  97         DEBUG(DEBUG_NOTICE,("Keepalive monitoring has been started\n"));
  98 }
  99
 100 void ctdb_stop_keepalive(struct ctdb_context *ctdb)
 101 {
 102         talloc_free(ctdb->keepalive_ctx);
 103         ctdb->keepalive_ctx = NULL;
 104 }
 105