Changeset 23040

Timestamp:

02/07/07 15:39:58 (22 months ago)

Author:

zarzycki@…

Message:

Add a per job exit timeout.

Location:

trunk/launchd/src

Files:

• launchd.plist.5 (modified) (1 diff)
• launchd_core_logic.c (modified) (12 diffs)
• liblaunch_public.h (modified) (1 diff)

trunk/launchd/src/launchd.plist.5

@@ -177 +177 @@
 .Nm launchd
 for use by the job at check in time.
+.It Sy ExitTimeOut <integer>
+The amount of time
+.Nm launchd
+waits before sending a SIGKILL signal. The default value is 20 seconds. The value zero is interpreted as infinity.
 .It Sy ThrottleInterval <integer>
 This key lets one override the default throttling policy imposed on jobs by

trunk/launchd/src/launchd_core_logic.c

@@ -88 +88 @@
 #define LAUNCHD_MIN_JOB_RUN_TIME 10
 #define LAUNCHD_ADVISABLE_IDLE_TIMEOUT 30
+#define LAUNCHD_DEFAULT_EXIT_TIMEOUT 20
 
 extern char **environ;
@@ -255 +256 @@
         int log_redirect_fd;
         int nice;
-        int timeout;
+        unsigned int timeout;
+        unsigned int exit_timeout;
         int stdout_err_fd;
+        struct timeval sent_sigterm_time;
         time_t start_time;
         time_t min_run_time;
@@ -406 +409 @@
         if (j->p) {
                 job_assumes(j, kill(j->p, SIGTERM) != -1);
+                job_assumes(j, gettimeofday(&j->sent_sigterm_time, NULL) != -1);
+                if (j->exit_timeout) {
+                        job_assumes(j, kevent_mod((uintptr_t)&j->exit_timeout, EVFILT_TIMER,
+                                                EV_ADD|EV_ONESHOT, NOTE_SECONDS, j->exit_timeout, j) != -1);
+                }
         }
 }
@@ -517 +525 @@
                         job_remove(ji);
                 } else {
-                        if (debug_shutdown_hangs) {
-                                job_assumes(ji, kevent_mod((uintptr_t)ji, EVFILT_TIMER, EV_ADD|EV_ONESHOT, NOTE_SECONDS, 8, ji) != -1);
-                        }
                         job_stop(ji);
                 }
@@ -663 +668 @@
                 job_assumes(j, kevent_mod((uintptr_t)&j->start_interval, EVFILT_TIMER, EV_DELETE, 0, 0, NULL) != -1);
         }
+        if (j->exit_timeout) {
+                kevent_mod((uintptr_t)&j->exit_timeout, EVFILT_TIMER, EV_DELETE, 0, 0, NULL);
+        }
 
         kevent_mod((uintptr_t)j, EVFILT_TIMER, EV_DELETE, 0, 0, NULL);
@@ -908 +916 @@
         j->min_run_time = LAUNCHD_MIN_JOB_RUN_TIME;
         j->timeout = LAUNCHD_ADVISABLE_IDLE_TIMEOUT;
+        j->exit_timeout = LAUNCHD_DEFAULT_EXIT_TIMEOUT;
         j->currently_ignored = true;
         j->ondemand = true;
@@ -1163 +1172 @@
 {
         switch (key[0]) {
+        case 'e':
+        case 'E':
+                if (strcasecmp(key, LAUNCH_JOBKEY_EXITTIMEOUT) == 0) {
+                        if (value < 0) {
+                                job_log(j, LOG_WARNING, "Exit timeout less zero. Ignoring.");
+                        } else {
+                                j->exit_timeout = value;
+                        }
+                }
+                break;
         case 'n':
         case 'N':
@@ -1578 +1597 @@
 job_reap(job_t j)
 {
+        struct timeval tve, tvd;
         struct rusage ru;
         int status;
@@ -1596 +1616 @@
                 return;
         }
+
+        job_assumes(j, gettimeofday(&tve, NULL) != -1);
 
         if (j->wait_reply_port) {
@@ -1601 +1623 @@
                 job_assumes(j, job_mig_wait_reply(j->wait_reply_port, 0, status) == 0);
                 j->wait_reply_port = MACH_PORT_NULL;
+        }
+
+        if (j->sent_sigterm_time.tv_sec) {
+                double delta;
+
+                timersub(&tve, &j->sent_sigterm_time,  &tvd);
+
+                delta = (double)tvd.tv_sec + (double)tvd.tv_usec / (double)1000000;
+
+                job_log(j, tvd.tv_sec ? LOG_NOTICE : LOG_INFO, "Exited %f seconds after SIGTERM was sent", delta);
         }
 
@@ -1725 +1757 @@
                 break;
         case EVFILT_TIMER:
-                if ((uintptr_t)j == kev->ident) {
-                        if (j->p && job_assumes(j, debug_shutdown_hangs)) {
+                if ((uintptr_t)j == kev->ident || (uintptr_t)&j->start_interval == kev->ident) {
+                        job_dispatch(j, true);
+                } else if ((uintptr_t)&j->exit_timeout == kev->ident) {
+                        if (debug_shutdown_hangs) {
                                 job_force_sampletool(j);
-                        } else {
-                                job_dispatch(j, true);
                         }
-                } else if ((uintptr_t)&j->start_interval == kev->ident) {
-                        job_dispatch(j, true);
+                        job_log(j, LOG_WARNING, "Exit timeout elapsed (%u seconds). Killing.", j->exit_timeout);
+                        job_assumes(j, kill(j->p, SIGKILL) != -1);
                 } else {
                         calendarinterval_callback(j, kev);
@@ -1788 +1820 @@
 
         job_log(j, LOG_DEBUG, "Starting");
+
+        j->sent_sigterm_time.tv_sec = 0;
+        j->sent_sigterm_time.tv_usec = 0;
 
         /* FIXME, using stdinpath is a hack for re-reading the conf file */

trunk/launchd/src/liblaunch_public.h

@@ -59 +59 @@
 #define LAUNCH_JOBKEY_GROUPNAME                 "GroupName"
 #define LAUNCH_JOBKEY_TIMEOUT                   "TimeOut"
+#define LAUNCH_JOBKEY_EXITTIMEOUT               "ExitTimeOut"
 #define LAUNCH_JOBKEY_INITGROUPS                "InitGroups"
 #define LAUNCH_JOBKEY_SOCKETS                   "Sockets"