2 * Bacula thread watchdog routine. General routine that monitors
3 * the daemon and signals a thread if it is blocked on a BSOCK
4 * too long. This prevents catastropic long waits -- generally
5 * due to Windows "hanging" the app.
7 * Kern Sibbald, January MMII
11 Copyright (C) 2000-2003 Kern Sibbald and John Walker
13 This program is free software; you can redistribute it and/or
14 modify it under the terms of the GNU General Public License as
15 published by the Free Software Foundation; either version 2 of
16 the License, or (at your option) any later version.
18 This program is distributed in the hope that it will be useful,
19 but WITHOUT ANY WARRANTY; without even the implied warranty of
20 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
21 General Public License for more details.
23 You should have received a copy of the GNU General Public
24 License along with this program; if not, write to the Free
25 Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
33 /* Exported globals */
34 time_t watchdog_time; /* this has granularity of SLEEP_TIME */
36 #define SLEEP_TIME 30 /* examine things every 30 seconds */
38 /* Forward referenced functions */
39 static void *btimer_thread(void *arg);
40 static void stop_btimer(btimer_id wid);
41 static btimer_id btimer_start_common(uint32_t wait);
44 static pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
45 static pthread_cond_t timer = PTHREAD_COND_INITIALIZER;
47 static btimer_t *timer_chain = NULL;
51 * Timeout signal comes here
53 void timeout_handler(int sig)
55 return; /* thus interrupting the function */
60 * Start watchdog thread
62 * Returns: 0 on success
65 int start_watchdog(void)
69 struct sigaction sigtimer;
71 sigtimer.sa_flags = 0;
72 sigtimer.sa_handler = timeout_handler;
73 sigfillset(&sigtimer.sa_mask);
74 sigaction(TIMEOUT_SIGNAL, &sigtimer, NULL);
75 watchdog_time = time(NULL);
77 if ((stat = pthread_create(&wdid, NULL, btimer_thread, (void *)NULL)) != 0) {
84 * Terminate the watchdog thread
86 * Returns: 0 on success
89 int stop_watchdog(void)
95 if ((stat = pthread_cond_signal(&timer)) != 0) {
105 * This is the actual watchdog thread.
107 static void *btimer_thread(void *arg)
113 Dmsg0(200, "Start watchdog thread\n");
114 pthread_detach(pthread_self());
117 time_t timer_start, now;
119 Dmsg0(200, "Top of watchdog loop\n");
121 watchdog_time = time(NULL); /* update timer */
123 /* Walk through all JCRs checking if any one is
124 * blocked for more than specified max time.
127 for (jcr=NULL; (jcr=get_next_jcr(jcr)); ) {
128 free_locked_jcr(jcr);
129 if (jcr->JobId == 0) {
132 fd = jcr->store_bsock;
134 timer_start = fd->timer_start;
135 if (timer_start && (watchdog_time - timer_start) > fd->timeout) {
136 fd->timed_out = TRUE;
137 Jmsg(jcr, M_ERROR, 0, _(
138 "Watchdog sending kill after %d secs to thread stalled reading Storage daemon.\n"),
139 watchdog_time - timer_start);
140 pthread_kill(jcr->my_thread_id, TIMEOUT_SIGNAL);
143 fd = jcr->file_bsock;
145 timer_start = fd->timer_start;
146 if (timer_start && (watchdog_time - timer_start) > fd->timeout) {
147 fd->timed_out = TRUE;
148 Jmsg(jcr, M_ERROR, 0, _(
149 "Watchdog sending kill after %d secs to thread stalled reading File daemon.\n"),
150 watchdog_time - timer_start);
151 pthread_kill(jcr->my_thread_id, TIMEOUT_SIGNAL);
156 timer_start = fd->timer_start;
157 if (timer_start && (watchdog_time - timer_start) > fd->timeout) {
158 fd->timed_out = TRUE;
159 Jmsg(jcr, M_ERROR, 0, _(
160 "Watchdog sending kill after %d secs to thread stalled reading Director.\n"),
161 watchdog_time - timer_start);
162 pthread_kill(jcr->my_thread_id, TIMEOUT_SIGNAL);
169 Dmsg0(200, "Watchdog sleep.\n");
170 bmicrosleep(SLEEP_TIME, 0);
174 * Now handle child and thread timers set by the code.
176 /* Walk child chain killing off any process overdue */
178 for (wid = timer_chain; wid; wid=wid->next) {
180 /* First ask him politely to go away */
181 if (!wid->killed && now > (wid->start_time + wid->wait)) {
182 // Dmsg1(000, "Watchdog sigterm pid=%d\n", wid->pid);
183 if (wid->type == TYPE_CHILD) {
184 kill(wid->pid, SIGTERM);
187 Dmsg1(200, "watchdog kill thread %d\n", wid->tid);
188 pthread_kill(wid->tid, TIMEOUT_SIGNAL);
192 /* If we asked a child to die, wait 3 seconds and slam him */
196 for (wid1 = timer_chain; wid1; wid1=wid1->next) {
197 if (wid->type == TYPE_CHILD &&
198 !wid1->killed && now > (wid1->start_time + wid1->wait)) {
199 kill(wid1->pid, SIGKILL);
200 // Dmsg1(000, "Watchdog killed pid=%d\n", wid->pid);
207 } /* end of big for loop */
209 Dmsg0(200, "End watchdog\n");
214 * Start a timer on a child process of pid, kill it after wait seconds.
215 * NOTE! Granularity is SLEEP_TIME (i.e. 30 seconds)
217 * Returns: btimer_id (pointer to btimer_t struct) on success
220 btimer_id start_child_timer(pid_t pid, uint32_t wait)
223 wid = btimer_start_common(wait);
225 wid->type = TYPE_CHILD;
226 Dmsg2(200, "Start child timer 0x%x for %d secs.\n", wid, wait);
231 * Start a timer on a thread. kill it after wait seconds.
232 * NOTE! Granularity is SLEEP_TIME (i.e. 30 seconds)
234 * Returns: btimer_id (pointer to btimer_t struct) on success
237 btimer_id start_thread_timer(pthread_t tid, uint32_t wait)
240 wid = btimer_start_common(wait);
242 wid->type = TYPE_PTHREAD;
243 Dmsg2(200, "Start thread timer 0x%x for %d secs.\n", wid, wait);
247 static btimer_id btimer_start_common(uint32_t wait)
249 btimer_id wid = (btimer_id)malloc(sizeof(btimer_t));
252 /* Chain it into timer_chain as the first item */
254 wid->next = timer_chain;
256 timer_chain->prev = wid;
259 wid->start_time = time(NULL);
269 void stop_child_timer(btimer_id wid)
271 Dmsg2(200, "Stop child timer 0x%x for %d secs.\n", wid, wid->wait);
278 void stop_thread_timer(btimer_id wid)
283 Dmsg2(200, "Stop thread timer 0x%x for %d secs.\n", wid, wid->wait);
291 static void stop_btimer(btimer_id wid)
294 Emsg0(M_ABORT, 0, _("NULL btimer_id.\n"));
297 /* Remove wid from timer_chain */
298 if (!wid->prev) { /* if no prev */
299 timer_chain = wid->next; /* set new head */
301 wid->prev->next = wid->next; /* update prev */
304 wid->next->prev = wid->prev; /* unlink it */