/*****************************************************************************\
* message hander thread
\*****************************************************************************/
static void *_msg_thread(void *no_data)
{
slurm_fd_t sock_fd = -1, new_fd;
slurm_addr_t cli_addr;
char *msg;
int i;
/* If JobSubmitDynAllocPort is already taken, keep trying to open it
* once per minute. Slurmctld will continue to function
* during this interval even if nothing can be scheduled. */
for (i=0; (!thread_shutdown); i++) {
if (i > 0)
sleep(60);
sock_fd = slurm_init_msg_engine_port(sched_port);
if (sock_fd != SLURM_SOCKET_ERROR)
break;
error("dynalloc: slurm_init_msg_engine_port %u %m",
sched_port);
error("dynalloc: Unable to communicate with ORTE RAS");
}
/* Process incoming RPCs until told to shutdown */
while (!thread_shutdown) {
if ((new_fd = slurm_accept_msg_conn(sock_fd, &cli_addr))
== SLURM_SOCKET_ERROR) {
if (errno != EINTR)
error("dyalloc: slurm_accept_msg_conn %m");
continue;
}
if (thread_shutdown) {
close(new_fd);
break;
}
err_code = 0;
err_msg = "";
msg = _recv_msg(new_fd);
if (msg) {
_proc_msg(new_fd, msg);
xfree(msg);
}
slurm_close_accepted_conn(new_fd);
}
verbose("dynalloc: message engine shutdown");
if (sock_fd > 0)
(void) slurm_shutdown_msg_engine(sock_fd);
pthread_exit((void *) 0);
return NULL;
}
static void *_msg_thread(void *no_data)
{
int sock_fd = -1, new_fd;
slurm_addr_t cli_addr;
char *msg;
int i;
/* If Port is already taken, keep trying to open it 10 secs */
for (i = 0; (!thread_shutdown); i++) {
if (i > 0)
sleep(10);
sock_fd = slurm_init_msg_engine_port(nonstop_comm_port);
if (sock_fd != SLURM_SOCKET_ERROR)
break;
error("slurmctld/nonstop: can not open port: %hu %m",
nonstop_comm_port);
}
/* Process incoming RPCs until told to shutdown */
while (!thread_shutdown) {
new_fd = slurm_accept_msg_conn(sock_fd, &cli_addr);
if (new_fd == SLURM_SOCKET_ERROR) {
if (errno != EINTR) {
info("slurmctld/nonstop: "
"slurm_accept_msg_conn %m");
}
continue;
}
if (thread_shutdown) {
close(new_fd);
break;
}
/* It would be nice to create a pthread for each new
* RPC, but that leaks memory on some systems when
* done from a plugin. Alternately, we could maintain
* a pool of pthreads and reuse them. */
msg = _recv_msg(new_fd);
if (msg) {
_proc_msg(new_fd, msg, cli_addr);
xfree(msg);
}
slurm_close(new_fd);
}
debug("slurmctld/nonstop: message engine shutdown");
if (sock_fd > 0)
(void) slurm_shutdown_msg_engine(sock_fd);
pthread_exit((void *) 0);
return NULL;
}
/*****************************************************************************\
* message hander thread
\*****************************************************************************/
static void *_msg_thread(void *no_data)
{
slurm_fd_t sock_fd = -1, new_fd;
slurm_addr_t cli_addr;
char *msg;
slurm_ctl_conf_t *conf;
int i;
/* Locks: Write configuration, job, node, and partition */
slurmctld_lock_t config_write_lock = {
WRITE_LOCK, WRITE_LOCK, WRITE_LOCK, WRITE_LOCK };
conf = slurm_conf_lock();
sched_port = conf->schedport;
slurm_conf_unlock();
/* Wait until configuration is completely loaded */
lock_slurmctld(config_write_lock);
unlock_slurmctld(config_write_lock);
/* If SchedulerPort is already taken, keep trying to open it
* once per minute. Slurmctld will continue to function
* during this interval even if nothing can be scheduled. */
for (i=0; (!thread_shutdown); i++) {
if (i > 0)
sleep(60);
sock_fd = slurm_init_msg_engine_port(sched_port);
if (sock_fd != SLURM_SOCKET_ERROR)
break;
error("wiki: slurm_init_msg_engine_port %u %m",
sched_port);
error("wiki: Unable to communicate with Moab");
}
/* Process incoming RPCs until told to shutdown */
while (!thread_shutdown) {
if ((new_fd = slurm_accept_msg_conn(sock_fd, &cli_addr))
== SLURM_SOCKET_ERROR) {
if (errno != EINTR)
error("wiki: slurm_accept_msg_conn %m");
continue;
}
if (thread_shutdown) {
close(new_fd);
break;
}
/* It would be nice to create a pthread for each new
* RPC, but that leaks memory on some systems when
* done from a plugin.
* FIXME: Maintain a pool of and reuse them. */
err_code = 0;
err_msg = "";
msg = _recv_msg(new_fd);
if (msg) {
_proc_msg(new_fd, msg);
xfree(msg);
}
slurm_close_accepted_conn(new_fd);
}
if (sock_fd > 0)
(void) slurm_shutdown_msg_engine(sock_fd);
pthread_exit((void *) 0);
return NULL;
}
