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 "replace.h"
  22 #include "system/filesys.h"
  23 #include "system/network.h"
  24 #include "system/time.h"
  25 #include "system/wait.h"
  26
  27 #include <talloc.h>
  28 #include <tevent.h>
  29
  30 #include "lib/util/debug.h"
  31 #include "lib/util/samba_util.h"
  32
  33 #include "ctdb_private.h"
  34 #include "ctdb_logging.h"
  35
  36 #include "common/common.h"
  37
  38
  39 /*
  40   see if any nodes are dead
  41  */
  42 static void ctdb_check_for_dead_nodes(struct tevent_context *ev,
  43                                       struct tevent_timer *te,
  44                                       struct timeval t, void *private_data)
  45 {
  46         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
  47         int i;
  48
  49         /* send a keepalive to all other nodes, unless */
  50         for (i=0;i<ctdb->num_nodes;i++) {
  51                 struct ctdb_node *node = ctdb->nodes[i];
  52
  53                 if (node->flags & NODE_FLAGS_DELETED) {
  54                         continue;
  55                 }
  56
  57                 if (node->pnn == ctdb->pnn) {
  58                         continue;
  59                 }
  60
  61                 if (node->flags & NODE_FLAGS_DISCONNECTED) {
  62                         /* it might have come alive again */
  63                         if (node->rx_cnt != 0) {
  64                                 ctdb_node_connected(node);
  65                         }
  66                         continue;
  67                 }
  68
  69
  70                 if (node->rx_cnt == 0) {
  71                         node->dead_count++;
  72                 } else {
  73                         node->dead_count = 0;
  74                 }
  75
  76                 node->rx_cnt = 0;
  77
  78                 if (node->dead_count >= ctdb->tunable.keepalive_limit) {
  79                         DEBUG(DEBUG_NOTICE,("dead count reached for node %u\n", node->pnn));
  80                         ctdb_node_dead(node);
  81                         ctdb_send_keepalive(ctdb, node->pnn);
  82                         /* maybe tell the transport layer to kill the
  83                            sockets as well?
  84                         */
  85                         continue;
  86                 }
  87
  88                 DEBUG(DEBUG_DEBUG,("sending keepalive to %u\n", node->pnn));
  89                 ctdb_send_keepalive(ctdb, node->pnn);
  90
  91                 node->tx_cnt = 0;
  92         }
  93
  94         tevent_add_timer(ctdb->ev, ctdb->keepalive_ctx,
  95                          timeval_current_ofs(ctdb->tunable.keepalive_interval, 0),
  96                          ctdb_check_for_dead_nodes, ctdb);
  97 }
  98
  99
 100 void ctdb_start_keepalive(struct ctdb_context *ctdb)
 101 {
 102         struct tevent_timer *te;
 103
 104         ctdb->keepalive_ctx = talloc_new(ctdb);
 105         CTDB_NO_MEMORY_FATAL(ctdb, ctdb->keepalive_ctx);
 106
 107         te = tevent_add_timer(ctdb->ev, ctdb->keepalive_ctx,
 108                               timeval_current_ofs(ctdb->tunable.keepalive_interval, 0),
 109                               ctdb_check_for_dead_nodes, ctdb);
 110         CTDB_NO_MEMORY_FATAL(ctdb, te);
 111
 112         DEBUG(DEBUG_NOTICE,("Keepalive monitoring has been started\n"));
 113 }
 114
 115 void ctdb_stop_keepalive(struct ctdb_context *ctdb)
 116 {
 117         talloc_free(ctdb->keepalive_ctx);
 118         ctdb->keepalive_ctx = NULL;
 119 }
 120