Add rolling statistics that are collected across 10 second intervals.
[metze/ctdb/wip.git] / 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/tevent/tevent.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
41                 if (node->flags & NODE_FLAGS_DELETED) {
42                         continue;
43                 }
44
45                 if (node->pnn == ctdb->pnn) {
46                         continue;
47                 }
48                 
49                 if (node->flags & NODE_FLAGS_DISCONNECTED) {
50                         /* it might have come alive again */
51                         if (node->rx_cnt != 0) {
52                                 ctdb_node_connected(node);
53                         }
54                         continue;
55                 }
56
57
58                 if (node->rx_cnt == 0) {
59                         node->dead_count++;
60                 } else {
61                         node->dead_count = 0;
62                 }
63
64                 node->rx_cnt = 0;
65
66                 if (node->dead_count >= ctdb->tunable.keepalive_limit) {
67                         DEBUG(DEBUG_NOTICE,("dead count reached for node %u\n", node->pnn));
68                         ctdb_node_dead(node);
69                         ctdb_send_keepalive(ctdb, node->pnn);
70                         /* maybe tell the transport layer to kill the
71                            sockets as well?
72                         */
73                         continue;
74                 }
75                 
76                 DEBUG(DEBUG_DEBUG,("sending keepalive to %u\n", node->pnn));
77                 ctdb_send_keepalive(ctdb, node->pnn);
78
79                 node->tx_cnt = 0;
80         }
81         
82         event_add_timed(ctdb->ev, ctdb->keepalive_ctx,
83                         timeval_current_ofs(ctdb->tunable.keepalive_interval, 0), 
84                         ctdb_check_for_dead_nodes, ctdb);
85 }
86
87
88 void ctdb_start_keepalive(struct ctdb_context *ctdb)
89 {
90         struct timed_event *te;
91
92         ctdb->keepalive_ctx = talloc_new(ctdb);
93         CTDB_NO_MEMORY_FATAL(ctdb, ctdb->keepalive_ctx);
94
95         te = event_add_timed(ctdb->ev, ctdb->keepalive_ctx,
96                              timeval_current_ofs(ctdb->tunable.keepalive_interval, 0), 
97                              ctdb_check_for_dead_nodes, ctdb);
98         CTDB_NO_MEMORY_FATAL(ctdb, te);
99
100         DEBUG(DEBUG_NOTICE,("Keepalive monitoring has been started\n"));
101 }
102
103 void ctdb_stop_keepalive(struct ctdb_context *ctdb)
104 {
105         talloc_free(ctdb->keepalive_ctx);
106         ctdb->keepalive_ctx = NULL;
107 }
108