remove custom logic for sigterm (#286)

* remove custom logic for sigterm

* fmt

* add comment

Co-authored-by: reshke <Kirill Reshke>

docker/bin/setup

@@ -1,6 +1,5 @@
 #!/bin/bash
 
-
 export SETUP_LOG=/var/log/postgresql/postgresql-13-main.log
 
 cat > /etc/postgresql/13/main/pg_hba.conf <<-EOF

docker/scram/config.conf

@@ -54,7 +54,7 @@ database "scram_db" {
 }
 
 daemonize yes
-pid_file "odyssey/data/odyssey.pid"
+pid_file "/var/run/odyssey.pid"
 
 unix_socket_dir "/tmp"
 unix_socket_mode "0644"

sources/cron.c

@@ -197,26 +197,36 @@ static void od_cron_err_stat(od_cron_t *cron)
 	od_route_pool_unlock(router->route_pool)
 }
 
-od_attribute_noreturn() static void od_cron(void *arg)
+static void od_cron(void *arg)
 {
 	od_cron_t *cron = arg;
 	od_instance_t *instance = cron->global->instance;
 
 	cron->stat_time_us = machine_time_us();
+	cron->online = 1;
 
 	int stats_tick = 0;
 	for (;;) {
-		/* mark and sweep expired idle server connections */
-		od_cron_expire(cron);
-
-		/* update statistics */
-		if (++stats_tick >= instance->config.stats_interval) {
-			od_cron_stat(cron);
-			stats_tick = 0;
+		if (!cron->online) {
+			return;
 		}
 
-		od_cron_err_stat(cron);
+                // we take a lock here 
+                // to prevent usage routes that are deallocated while shutdown
+                pthread_mutex_lock(&cron->lock);
+		{
+			/* mark and sweep expired idle server connections */
+			od_cron_expire(cron);
 
+			/* update statistics */
+			if (++stats_tick >= instance->config.stats_interval) {
+				od_cron_stat(cron);
+				stats_tick = 0;
+			}
+
+			od_cron_err_stat(cron);
+		}
+		pthread_mutex_unlock(&cron->lock);
 		/* 1 second soft interval */
 		machine_sleep(1000);
 	}
@@ -227,6 +237,9 @@ void od_cron_init(od_cron_t *cron)
 	cron->stat_time_us = 0;
 	cron->global = NULL;
 	cron->startup_errors = 0;
+
+	cron->online = 0;
+	pthread_mutex_init(&cron->lock, NULL);
 }
 
 int od_cron_start(od_cron_t *cron, od_global_t *global)
@@ -235,10 +248,18 @@ int od_cron_start(od_cron_t *cron, od_global_t *global)
 	od_instance_t *instance = global->instance;
 	int64_t coroutine_id;
 	coroutine_id = machine_coroutine_create(od_cron, cron);
-	if (coroutine_id == -1) {
+	if (coroutine_id == INVALID_COROUTINE_ID) {
 		od_error(&instance->logger, "cron", NULL, NULL,
 			 "failed to start cron coroutine");
-		return -1;
+		return NOT_OK_RESPONSE;
 	}
-	return 0;
+
+	return OK_RESPONSE;
+}
+
+od_retcode_t od_cron_stop(od_cron_t *cron)
+{
+	cron->online = 0;
+	pthread_mutex_lock(&cron->lock);
+	return OK_RESPONSE;
 }

sources/cron.h

@@ -13,6 +13,9 @@ struct od_cron {
 	uint64_t stat_time_us;
 	od_global_t *global;
 	od_atomic_u64_t startup_errors;
+
+	pthread_mutex_t lock;
+	int online;
 };
 
 void od_cron_init(od_cron_t *);

sources/frontend.c

@@ -547,7 +547,8 @@ static inline bool od_should_drop_connection(od_client_t *client,
 		/* fall through */
 	case OD_RULE_POOL_TRANSACTION: {
 		//TODO:: drop no more than X connection per sec/min/whatever
-		if (od_likely(instance->shutdown_worker_id == INVALID_ID)) {
+		if (od_likely(instance->shutdown_worker_id ==
+			      INVALID_COROUTINE_ID)) {
 			// try to optimize likely path
 			return false;
 		}

sources/instance.c

@@ -16,7 +16,7 @@ void od_instance_init(od_instance_t *instance)
 	od_config_init(&instance->config);
 
 	instance->config_file = NULL;
-	instance->shutdown_worker_id = INVALID_ID;
+	instance->shutdown_worker_id = INVALID_COROUTINE_ID;
 
 	sigset_t mask;
 	sigemptyset(&mask);

sources/macro.h

@@ -24,6 +24,6 @@ typedef int od_retcode_t;
 
 /*misc*/
 
-#define INVALID_ID -1
+#define INVALID_COROUTINE_ID -1
 
 #endif /* ODYSSEY_MACRO_H */

sources/route.h

@@ -33,6 +33,9 @@ struct od_route {
 static inline void od_route_init(od_route_t *route, bool extra_route_logging)
 {
 	route->rule = NULL;
+
+	route->tcp_connections = 0;
+
 	od_route_id_init(&route->id);
 	od_server_pool_init(&route->server_pool);
 	od_client_pool_init(&route->client_pool);

sources/sighandler.c

@@ -7,7 +7,7 @@ static inline od_retcode_t
 od_system_gracefully_killer_invoke(od_system_t *system)
 {
 	od_instance_t *instance = system->global->instance;
-	if (instance->shutdown_worker_id != INVALID_ID) {
+	if (instance->shutdown_worker_id != INVALID_COROUTINE_ID) {
 		return OK_RESPONSE;
 	}
 	int64_t mid;
@@ -46,6 +46,10 @@ od_attribute_noreturn() void od_system_shutdown(od_system_t *system,
 {
 	od_log(&instance->logger, "system", NULL, NULL,
 	       "SIGINT received, shutting down");
+
+	// lock here
+	od_cron_stop(system->global->cron);
+
 	od_worker_pool_stop(system->global->worker_pool);
 	od_router_free(system->global->router);
 	/* Prevent OpenSSL usage during deinitialization */
@@ -55,22 +59,6 @@ od_attribute_noreturn() void od_system_shutdown(od_system_t *system,
 	exit(0);
 }
 
-od_attribute_noreturn() void od_system_shutdown_fast(od_system_t *system,
-						     od_instance_t *instance)
-{
-	od_log(&instance->logger, "system", NULL, NULL,
-	       "SIGTERM received, shutting down");
-	od_worker_pool_stop(system->global->worker_pool);
-	od_router_free(system->global->router);
-
-	/* No time for caution */
-	od_system_cleanup(system);
-
-	/* TODO:  */
-	od_modules_unload_fast(system->global->modules);
-	exit(0);
-}
-
 void od_system_signal_handler(void *arg)
 {
 	od_system_t *system = arg;
@@ -100,8 +88,6 @@ void od_system_signal_handler(void *arg)
 			break;
 		switch (rc) {
 		case SIGTERM:
-			od_system_shutdown_fast(system, instance);
-			break;
 		case SIGINT:
 			od_system_shutdown(system, instance);
 			break;