static int _ping_controller(slurmdb_cluster_rec_t *cluster)
{
int rc = SLURM_SUCCESS;
slurm_msg_t req_msg;
slurm_msg_t resp_msg;
slurm_msg_t_init(&req_msg);
slurm_msg_t_init(&resp_msg);
req_msg.msg_type = REQUEST_PING;
slurm_mutex_lock(&cluster->lock);
if (slurmctld_conf.debug_flags & DEBUG_FLAG_FEDR)
info("pinging %s(%s:%d)", cluster->name, cluster->control_host,
cluster->control_port);
if ((rc = _send_recv_msg(cluster, &req_msg, &resp_msg, true))) {
error("failed to ping %s(%s:%d)",
cluster->name, cluster->control_host,
cluster->control_port);
} else if ((rc = slurm_get_return_code(resp_msg.msg_type,
resp_msg.data)))
error("ping returned error from %s(%s:%d)",
cluster->name, cluster->control_host,
cluster->control_port);
if (slurmctld_conf.debug_flags & DEBUG_FLAG_FEDR)
info("finished pinging %s(%s:%d)", cluster->name,
cluster->control_host, cluster->control_port);
slurm_mutex_unlock(&cluster->lock);
slurm_free_msg_members(&req_msg);
slurm_free_msg_members(&resp_msg);
return rc;
}
static void *_msg_thr_internal(void *arg)
{
slurm_addr_t cli_addr;
int newsockfd;
slurm_msg_t msg;
int *slurmctld_fd_ptr = (int *)arg;
(void) pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL);
(void) pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, NULL);
while (!srun_shutdown) {
newsockfd = slurm_accept_msg_conn(*slurmctld_fd_ptr, &cli_addr);
if (newsockfd == SLURM_SOCKET_ERROR) {
if (errno != EINTR)
error("slurm_accept_msg_conn: %m");
continue;
}
slurm_msg_t_init(&msg);
if (slurm_receive_msg(newsockfd, &msg, 0) != 0) {
error("slurm_receive_msg: %m");
/* close the new socket */
close(newsockfd);
continue;
}
_handle_msg(&msg);
slurm_free_msg_members(&msg);
close(newsockfd);
}
return NULL;
}
/* _background_rpc_mgr - Read and process incoming RPCs to the background
* controller (that's us) */
static void *_background_rpc_mgr(void *no_data)
{
int newsockfd;
int sockfd;
slurm_addr_t cli_addr;
slurm_msg_t msg;
int error_code;
char* node_addr = NULL;
/* Read configuration only */
slurmctld_lock_t config_read_lock = {
READ_LOCK, NO_LOCK, NO_LOCK, NO_LOCK, NO_LOCK };
int sigarray[] = {SIGUSR1, 0};
(void) pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL);
(void) pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, NULL);
debug3("_background_rpc_mgr pid = %lu", (unsigned long) getpid());
/* initialize port for RPCs */
lock_slurmctld(config_read_lock);
/* set node_addr to bind to (NULL means any) */
if ((xstrcmp(slurmctld_conf.backup_controller,
slurmctld_conf.backup_addr) != 0)) {
node_addr = slurmctld_conf.backup_addr ;
}
if ((sockfd =
slurm_init_msg_engine_addrname_port(node_addr,
slurmctld_conf.
slurmctld_port))
== SLURM_SOCKET_ERROR)
fatal("slurm_init_msg_engine_addrname_port error %m");
unlock_slurmctld(config_read_lock);
/* Prepare to catch SIGUSR1 to interrupt accept().
* This signal is generated by the slurmctld signal
* handler thread upon receipt of SIGABRT, SIGINT,
* or SIGTERM. That thread does all processing of
* all signals. */
xsignal(SIGUSR1, _sig_handler);
xsignal_unblock(sigarray);
/*
* Process incoming RPCs indefinitely
*/
while (slurmctld_config.shutdown_time == 0) {
/* accept needed for stream implementation
* is a no-op in message implementation that just passes
* sockfd to newsockfd */
if ((newsockfd = slurm_accept_msg_conn(sockfd, &cli_addr))
== SLURM_SOCKET_ERROR) {
if (errno != EINTR)
error("slurm_accept_msg_conn: %m");
continue;
}
slurm_msg_t_init(&msg);
if (slurm_receive_msg(newsockfd, &msg, 0) != 0)
error("slurm_receive_msg: %m");
error_code = _background_process_msg(&msg);
if ((error_code == SLURM_SUCCESS) &&
(msg.msg_type == REQUEST_SHUTDOWN_IMMEDIATE) &&
(slurmctld_config.shutdown_time == 0))
slurmctld_config.shutdown_time = time(NULL);
slurm_free_msg_members(&msg);
slurm_close(newsockfd); /* close new socket */
}
debug3("_background_rpc_mgr shutting down");
slurm_close(sockfd); /* close the main socket */
pthread_exit((void *) 0);
return NULL;
}
