ctdb/common/ctdb_monitor.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 2 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, write to the Free Software
  19    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20 */
  21
  22 #include "includes.h"
  23 #include "lib/events/events.h"
  24 #include "system/filesys.h"
  25 #include "system/wait.h"
  26 #include "../include/ctdb_private.h"
  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         if (ctdb->monitoring_mode == CTDB_MONITORING_DISABLED) {
  38                 event_add_timed(ctdb->ev, ctdb,
  39                         timeval_current_ofs(ctdb->tunable.keepalive_interval, 0),
  40                         ctdb_check_for_dead_nodes, ctdb);
  41                 return;
  42         }
  43
  44         /* send a keepalive to all other nodes, unless */
  45         for (i=0;i<ctdb->num_nodes;i++) {
  46                 struct ctdb_node *node = ctdb->nodes[i];
  47                 if (node->vnn == ctdb->vnn) {
  48                         continue;
  49                 }
  50
  51                 if (!(node->flags & NODE_FLAGS_CONNECTED)) {
  52                         /* it might have come alive again */
  53                         if (node->rx_cnt != 0) {
  54                                 ctdb_node_connected(node);
  55                         }
  56                         continue;
  57                 }
  58
  59
  60                 if (node->rx_cnt == 0) {
  61                         node->dead_count++;
  62                 } else {
  63                         node->dead_count = 0;
  64                 }
  65
  66                 node->rx_cnt = 0;
  67
  68                 if (node->dead_count >= ctdb->tunable.keepalive_limit) {
  69                         DEBUG(0,("dead count reached for node %u\n", node->vnn));
  70                         ctdb_node_dead(node);
  71                         ctdb_send_keepalive(ctdb, node->vnn);
  72                         /* maybe tell the transport layer to kill the
  73                            sockets as well?
  74                         */
  75                         continue;
  76                 }
  77
  78                 if (node->tx_cnt == 0) {
  79                         DEBUG(5,("sending keepalive to %u\n", node->vnn));
  80                         ctdb_send_keepalive(ctdb, node->vnn);
  81                 }
  82
  83                 node->tx_cnt = 0;
  84         }
  85
  86         event_add_timed(ctdb->ev, ctdb,
  87                         timeval_current_ofs(ctdb->tunable.keepalive_interval, 0),
  88                         ctdb_check_for_dead_nodes, ctdb);
  89 }
  90
  91 static void ctdb_check_health(struct event_context *ev, struct timed_event *te,
  92                               struct timeval t, void *private_data);
  93
  94 /*
  95   called when a health monitoring event script finishes
  96  */
  97 static void ctdb_health_callback(struct ctdb_context *ctdb, int status, void *p)
  98 {
  99         struct ctdb_node *node = ctdb->nodes[ctdb->vnn];
 100         TDB_DATA data;
 101         struct ctdb_node_flag_change c;
 102
 103         event_add_timed(ctdb->ev, ctdb,
 104                         timeval_current_ofs(ctdb->tunable.monitor_interval, 0),
 105                         ctdb_check_health, ctdb);
 106
 107         if (status != 0 && !(node->flags & NODE_FLAGS_DISABLED)) {
 108                 DEBUG(0,("monitor event failed - disabling node\n"));
 109                 node->flags |= NODE_FLAGS_DISABLED;
 110         } else if (status == 0 && (node->flags & NODE_FLAGS_DISABLED)) {
 111                 DEBUG(0,("monitor event OK - node re-enabled\n"));
 112                 ctdb->nodes[ctdb->vnn]->flags &= ~NODE_FLAGS_DISABLED;
 113         } else {
 114                 /* no change */
 115                 return;
 116         }
 117
 118         c.vnn = ctdb->vnn;
 119         c.flags = node->flags;
 120
 121         data.dptr = (uint8_t *)&c;
 122         data.dsize = sizeof(c);
 123
 124         /* tell the recmaster that something has changed */
 125         ctdb_send_message(ctdb, ctdb->recovery_master, CTDB_SRVID_NODE_FLAGS_CHANGED, data);
 126 }
 127
 128
 129 /*
 130   see if the event scripts think we are healthy
 131  */
 132 static void ctdb_check_health(struct event_context *ev, struct timed_event *te,
 133                               struct timeval t, void *private_data)
 134 {
 135         struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
 136         int ret;
 137
 138         if (ctdb->monitoring_mode == CTDB_MONITORING_DISABLED) {
 139                 event_add_timed(ctdb->ev, ctdb,
 140                                 timeval_current_ofs(ctdb->tunable.monitor_interval, 0),
 141                                 ctdb_check_health, ctdb);
 142                 return;
 143         }
 144
 145         ret = ctdb_event_script_callback(ctdb, ctdb, ctdb_health_callback, ctdb, "monitor");
 146         if (ret != 0) {
 147                 DEBUG(0,("Unable to launch monitor event script\n"));
 148                 event_add_timed(ctdb->ev, ctdb,
 149                                 timeval_current_ofs(ctdb->tunable.monitor_interval, 0),
 150                                 ctdb_check_health, ctdb);
 151         }
 152 }
 153
 154
 155 /*
 156   start watching for nodes that might be dead
 157  */
 158 int ctdb_start_monitoring(struct ctdb_context *ctdb)
 159 {
 160         event_add_timed(ctdb->ev, ctdb,
 161                         timeval_current_ofs(ctdb->tunable.keepalive_interval, 0),
 162                         ctdb_check_for_dead_nodes, ctdb);
 163         event_add_timed(ctdb->ev, ctdb,
 164                         timeval_current_ofs(ctdb->tunable.monitor_interval, 0),
 165                         ctdb_check_health, ctdb);
 166         return 0;
 167 }