Adds checks & log messages on semaphore errors

src/conf.c

@@ -260,7 +260,7 @@ int pyfcgi_wd_init(void (*wd_sig_cleaner)(int), const struct timespec *delay)
 	PyFCGI_conf.context.wd_killdelay = timeout;
 
 	pyfcgi_log(LOG_DEBUG,
-		"Set watchdog with %d.%09ds timeout (%d.%09ds for SIGKILL)",
+		"Set KILL watchdog with %d.%09ds timeout (%d.%09ds)",
 		PyFCGI_conf.context.wd_delay.tv_sec,
 		PyFCGI_conf.context.wd_delay.tv_nsec,
 		PyFCGI_conf.context.wd_killdelay.tv_sec,

src/pyworker.c

@@ -618,17 +618,14 @@ static void worker_set_busy()
 	 * maybe to nanosleep 0.01s and retry a sem_trywait... SysV sem are
 	 * better T_T
 	 */
-#ifdef DEBUG
 	if(sem_trywait(PyFCGI_SEM(SEM_WSTATE).sem) < 0)
-#else
-	if(sem_wait(PyFCGI_SEM(SEM_WSTATE).sem) < 0)
-#endif
 	{
 		err = errno;
 		if(err == EAGAIN)
 		{ //panic
 			pyfcgi_log(LOG_ALERT, "Unable to set busy ! WSTATE sem is allready 0 !!!");
 			_worker_idle = 0;
+			kill(PyFCGI_conf.context.ppid, SIGTERM);
 			return;
 		}
 		pyfcgi_log(LOG_ERR, "error decrementing the WSTATE semaphore : %s",

src/responder.c

@@ -22,6 +22,10 @@
  * @ingroup work_master_proc */
 static void clean_exit(int status)
 {
+	if(PyFCGI_conf.context.n_wrk)
+	{
+		kill(PyFCGI_conf.context.pid, SIGTERM);
+	}
 	pyfcgi_IPC_close(IPC_WSTATE | IPC_WREQS | IPC_SEMST | IPC_SHMST);
 	pyfcgi_IPC_destroy(IPC_WSTATE | IPC_WREQS | IPC_SEMST | IPC_SHMST);
 	exit(status);
@@ -166,6 +170,8 @@ int responder_loop()
 		PyFCGI_conf.context.n_wrk = n_wrk;
 		if(last_update != (now = time(NULL)))
 		{
+pyfcgi_log(LOG_DEBUG, "Infos : n_wrk=%d max=%d min=%d",
+	n_wrk, PyFCGI_conf.max_wrk, PyFCGI_conf.min_wrk);
 			pyfcgi_pool_shm_update(n_wrk);
 			last_update = now;
 		}
@@ -191,11 +197,19 @@ int responder_loop()
 			}
 			if(WIFSIGNALED(status))
 			{
+				if(WTERMSIG(status) == 9)
+				{
+					pyfcgi_log(LOG_ALERT,
+						"Worker[%d] get killed ! No guaranty that SEM_WSTATE is OK, exiting...",
+						n);
+					clean_exit(PYFCGI_WORKER_FAIL);
+				}
 				if(WTERMSIG(status) == 11)
 				{
 					pyfcgi_log(LOG_ALERT,
-						"Worker[%d] segfault !",
+						"Worker[%d] segfault ! No guaranty that SEM_WSTATE is OK, exiting...",
 						n);
+					clean_exit(PYFCGI_WORKER_FAIL);
 				}
 				else
 				{
@@ -442,7 +456,13 @@ int pyfcgi_pool_idle(const struct timespec *timeout)
 				clean_exit(PYFCGI_FATAL);
 		}
 	}
-	sem_post(PyFCGI_SEM(SEM_WSTATE).sem); //Hope no worker fails to set busy...
+	if(sem_post(PyFCGI_SEM(SEM_WSTATE).sem) < 0)
+	{
+		pyfcgi_log(LOG_ALERT,
+			"Unable to sempost after a sem_timedwait : %s",
+			strerror(errno));
+		clean_exit(PYFCGI_FATAL);
+	}
 	return 1; //idle
 }
 
@@ -493,6 +513,7 @@ void pool_sighandler(int signum)
 		}
 		
 	}
+	PyFCGI_conf.context.n_wrk = 0;
 	clean_exit(0);
 }
 
@@ -553,9 +574,14 @@ void pyfcgi_pool_shm_update(int nworker)
 
 	data = (pyfcgi_stats_shm_t*)PyFCGI_conf.shm.ptr;
 	data->nworker = nworker;
+	err = 0;
 	if(sem_getvalue(PyFCGI_SEM(SEM_WSTATE).sem, &(data->pool_load)) < 0)
 	{
 		data->pool_load = -1;
+		pyfcgi_log(LOG_ALERT,
+			"Unable to get semaphore value for SEM_WSTATE : ",
+			strerror(errno));
+		err = 1;
 	}
 
 	if(sem_post(PyFCGI_SEM(SEM_STATS).sem) < 0)
@@ -564,5 +590,9 @@ void pyfcgi_pool_shm_update(int nworker)
 			strerror(errno));
 		clean_exit(PYFCGI_FATAL);
 	}
+	if(err)
+	{
+		clean_exit(PYFCGI_FATAL);
+	}
 }
 

src/stats.c

@@ -359,13 +359,21 @@ int pyfcgi_stats_get_shm(pyfcgi_stats_shm_t *res)
 
 	*res = *(pyfcgi_stats_shm_t*)PyFCGI_conf.shm.ptr;
 
+	err = 0;
+	if(res->pool_load == -1)
+	{
+		pyfcgi_log(LOG_INFO,
+			"Dropping data : pool_handler seems to encounter problems");
+		err = 1;
+		memset(res, 0, sizeof(pyfcgi_stats_shm_t));
+	}
 	if(sem_post(PyFCGI_SEM(SEM_STATS).sem) < 0)
 	{
 		pyfcgi_log(LOG_ALERT, "Unable to post sem at shm update : %s",
 			strerror(errno));
 		return -1;
 	}
-	return 0;
+	return err?-1:0;
 }