/* start new worker if needed */
void check_worker_population() {
int x, now, target_number_of_workers;
gm_log( GM_LOG_TRACE3, "check_worker_population()\n");
/* set current worker number */
count_current_worker(GM_ENABLED);
/* check if status worker died */
if( shm[3] == -1 ) {
make_new_child(GM_WORKER_STATUS);
}
/* keep up minimum population */
for (x = current_number_of_workers; x < mod_gm_opt->min_worker; x++) {
make_new_child(GM_WORKER_MULTI);
current_number_of_workers++;
}
/* check every second */
now = (int)time(NULL);
if(last_time_increased >= now)
return;
target_number_of_workers = adjust_number_of_worker(mod_gm_opt->min_worker, mod_gm_opt->max_worker, current_number_of_workers, current_number_of_jobs);
for (x = current_number_of_workers; x < target_number_of_workers; x++) {
last_time_increased = now;
/* top up the worker pool */
make_new_child(GM_WORKER_MULTI);
}
return;
}
/* start new worker if needed */
void check_worker_population() {
int x, now, status, target_number_of_workers;
gm_log( GM_LOG_TRACE3, "check_worker_population()\n");
now = (int)time(NULL);
/* collect finished workers */
while(waitpid(-1, &status, WNOHANG) > 0)
gm_log( GM_LOG_TRACE, "waitpid() worker exited with: %d\n", status);
/* set current worker number */
count_current_worker(GM_ENABLED);
/* check last check time, force restart all worker if there is no result in 2 minutes */
if( shm[SHM_WORKER_LAST_CHECK] < (now - 120) ) {
gm_log( GM_LOG_INFO, "no checks in 2minutes, restarting all workers\n", shm[SHM_WORKER_LAST_CHECK]);
shm[SHM_WORKER_LAST_CHECK] = now;
for(x=SHM_SHIFT; x < mod_gm_opt->max_worker+SHM_SHIFT; x++) {
save_kill(shm[x], SIGINT);
}
sleep(3);
for(x=SHM_SHIFT; x < mod_gm_opt->max_worker+SHM_SHIFT; x++) {
save_kill(shm[x], SIGKILL);
shm[x] = -1;
}
}
/* check if status worker died */
if( shm[SHM_STATUS_WORKER_PID] == -1 ) {
make_new_child(GM_WORKER_STATUS);
}
/* keep up minimum population */
for (x = current_number_of_workers; x < mod_gm_opt->min_worker; x++) {
make_new_child(GM_WORKER_MULTI);
current_number_of_workers++;
}
/* check every second if we need to increase worker population */
if(last_time_increased >= now)
return;
target_number_of_workers = adjust_number_of_worker(mod_gm_opt->min_worker, mod_gm_opt->max_worker, current_number_of_workers, current_number_of_jobs);
for (x = current_number_of_workers; x < target_number_of_workers; x++) {
last_time_increased = now;
/* top up the worker pool */
make_new_child(GM_WORKER_MULTI);
}
return;
}
/* stop all children */
void stop_children(int mode) {
int status, chld;
int waited = 0;
int x;
gm_log( GM_LOG_TRACE, "stop_children(%d)\n", mode);
/* ignore some signals for now */
signal(SIGTERM, SIG_IGN);
signal(SIGINT, SIG_IGN);
/*
* send term signal to our children
* children will finish the current job and exit
*/
killpg(0, SIGTERM);
while(current_number_of_workers > 0) {
gm_log( GM_LOG_TRACE, "send SIGTERM\n");
save_kill(shm[SHM_STATUS_WORKER_PID], SIGTERM);
for(x=SHM_SHIFT; x < mod_gm_opt->max_worker+SHM_SHIFT; x++) {
save_kill(shm[x], SIGTERM);
}
while((chld = waitpid(-1, &status, WNOHANG)) != -1 && chld > 0) {
gm_log( GM_LOG_TRACE, "wait() %d exited with %d\n", chld, status);
}
if(mode == GM_WORKER_RESTART)
break;
sleep(1);
waited++;
if(waited > GM_CHILD_SHUTDOWN_TIMEOUT) {
break;
}
count_current_worker(GM_DISABLED);
if(current_number_of_workers == 0)
break;
gm_log( GM_LOG_TRACE, "still waiting (%d) %d children missing...\n", waited, current_number_of_workers);
}
if(mode == GM_WORKER_STOP) {
killpg(0, SIGINT);
count_current_worker(GM_DISABLED);
if(current_number_of_workers == 0)
return;
gm_log( GM_LOG_TRACE, "sending SIGINT...\n");
save_kill(shm[SHM_STATUS_WORKER_PID], SIGINT);
for(x=SHM_SHIFT; x < mod_gm_opt->max_worker+SHM_SHIFT; x++) {
save_kill(shm[x], SIGINT);
}
/* wait 3 more seconds*/
if(current_number_of_workers > 0)
sleep(3);
while((chld = waitpid(-1, &status, WNOHANG)) != -1 && chld > 0) {
gm_log( GM_LOG_TRACE, "wait() %d exited with %d\n", chld, status);
}
/* kill them the hard way */
count_current_worker(GM_DISABLED);
if(current_number_of_workers == 0)
return;
save_kill(shm[SHM_STATUS_WORKER_PID], SIGKILL);
for(x=SHM_SHIFT; x < mod_gm_opt->max_worker+SHM_SHIFT; x++) {
save_kill(shm[x], SIGKILL);
}
/* count children a last time */
count_current_worker(GM_DISABLED);
if(current_number_of_workers == 0)
return;
/* this will kill us too */
gm_log( GM_LOG_ERROR, "exiting by SIGKILL...\n");
killpg(0, SIGKILL);
}
/* restore signal handlers for a clean exit */
signal(SIGINT, clean_exit);
signal(SIGTERM,clean_exit);
}