C++ slurm_set_addr函数代码示例

/*
 * srun_node_fail - notify srun of a node's failure
 * IN job_ptr - job to notify
 * IN node_name - name of failed node
 */
extern void srun_node_fail(struct job_record *job_ptr, char *node_name)
{
#ifndef HAVE_FRONT_END
	struct node_record *node_ptr;
#endif
	int bit_position = -1;
	slurm_addr_t * addr;
	srun_node_fail_msg_t *msg_arg;
	ListIterator step_iterator;
	struct step_record *step_ptr;

	xassert(job_ptr);
	xassert(node_name);
	if (!job_ptr || !IS_JOB_RUNNING(job_ptr))
		return;

#ifdef HAVE_FRONT_END
	/* Purge all jobs steps in front end mode */
#else
	if (!node_name || (node_ptr = find_node_record(node_name)) == NULL)
		return;
	bit_position = node_ptr - node_record_table_ptr;
#endif

	step_iterator = list_iterator_create(job_ptr->step_list);
	while ((step_ptr = (struct step_record *) list_next(step_iterator))) {
		if (step_ptr->step_node_bitmap == NULL)   /* pending step */
			continue;
		if ((bit_position >= 0) &&
		    (!bit_test(step_ptr->step_node_bitmap, bit_position)))
			continue;	/* job step not on this node */
		if ( (step_ptr->port    == 0)    ||
		     (step_ptr->host    == NULL) ||
		     (step_ptr->batch_step)      ||
		     (step_ptr->host[0] == '\0') )
			continue;
		addr = xmalloc(sizeof(struct sockaddr_in));
		slurm_set_addr(addr, step_ptr->port, step_ptr->host);
		msg_arg = xmalloc(sizeof(srun_node_fail_msg_t));
		msg_arg->job_id   = job_ptr->job_id;
		msg_arg->step_id  = step_ptr->step_id;
		msg_arg->nodelist = xstrdup(node_name);
		_srun_agent_launch(addr, step_ptr->host, SRUN_NODE_FAIL,
				   msg_arg, step_ptr->start_protocol_ver);
	}
	list_iterator_destroy(step_iterator);

	if (job_ptr->other_port && job_ptr->alloc_node && job_ptr->resp_host) {
		addr = xmalloc(sizeof(struct sockaddr_in));
		slurm_set_addr(addr, job_ptr->other_port, job_ptr->resp_host);
		msg_arg = xmalloc(sizeof(srun_node_fail_msg_t));
		msg_arg->job_id   = job_ptr->job_id;
		msg_arg->step_id  = NO_VAL;
		msg_arg->nodelist = xstrdup(node_name);
		_srun_agent_launch(addr, job_ptr->alloc_node, SRUN_NODE_FAIL,
				   msg_arg, job_ptr->start_protocol_ver);
	}
}

/*
 * srun_allocate - notify srun of a resource allocation
 * IN job_ptr - job allocated resources
 */
extern void srun_allocate(struct job_record *job_ptr)
{
	struct job_record *pack_job, *pack_leader;
	resource_allocation_response_msg_t *msg_arg = NULL;
	slurm_addr_t *addr;
	ListIterator iter;
	List job_resp_list = NULL;

	xassert(job_ptr);
	if (!job_ptr || !job_ptr->alloc_resp_port || !job_ptr->alloc_node ||
	    !job_ptr->resp_host || !job_ptr->job_resrcs ||
	    !job_ptr->job_resrcs->cpu_array_cnt)
		return;

	if (job_ptr->pack_job_id == 0) {
		addr = xmalloc(sizeof(struct sockaddr_in));
		slurm_set_addr(addr, job_ptr->alloc_resp_port,
			job_ptr->resp_host);

		msg_arg = build_alloc_msg(job_ptr, SLURM_SUCCESS, NULL);
		_srun_agent_launch(addr, job_ptr->alloc_node,
				   RESPONSE_RESOURCE_ALLOCATION, msg_arg,
				   job_ptr->start_protocol_ver);
	} else if (_pending_pack_jobs(job_ptr)) {
		return;
	} else if ((pack_leader = find_job_record(job_ptr->pack_job_id))) {
		addr = xmalloc(sizeof(struct sockaddr_in));
		slurm_set_addr(addr, pack_leader->alloc_resp_port,
			       pack_leader->resp_host);
		job_resp_list = list_create(_free_srun_alloc);
		iter = list_iterator_create(pack_leader->pack_job_list);
		while ((pack_job = (struct job_record *) list_next(iter))) {
			if (pack_leader->pack_job_id != pack_job->pack_job_id) {
				error("%s: Bad pack_job_list for %pJ",
				      __func__, pack_leader);
				continue;
			}
			msg_arg = build_alloc_msg(pack_job, SLURM_SUCCESS,
						  NULL);
			list_append(job_resp_list, msg_arg);
			msg_arg = NULL;
		}
		list_iterator_destroy(iter);
		_srun_agent_launch(addr, job_ptr->alloc_node,
				   RESPONSE_JOB_PACK_ALLOCATION, job_resp_list,
				   job_ptr->start_protocol_ver);
	} else {
		error("%s: Can not find pack job leader %pJ",
		      __func__, job_ptr);
	}
}

/*
 * srun_timeout - notify srun of a job's imminent timeout
 * IN job_ptr - pointer to the slurmctld job record
 */
extern void srun_timeout (struct job_record *job_ptr)
{
	slurm_addr_t * addr;
	srun_timeout_msg_t *msg_arg;
	ListIterator step_iterator;
	struct step_record *step_ptr;

	xassert(job_ptr);
	if (!IS_JOB_RUNNING(job_ptr))
		return;

	if (job_ptr->other_port && job_ptr->alloc_node && job_ptr->resp_host) {
		addr = xmalloc(sizeof(struct sockaddr_in));
		slurm_set_addr(addr, job_ptr->other_port, job_ptr->resp_host);
		msg_arg = xmalloc(sizeof(srun_timeout_msg_t));
		msg_arg->job_id   = job_ptr->job_id;
		msg_arg->step_id  = NO_VAL;
		msg_arg->timeout  = job_ptr->end_time;
		_srun_agent_launch(addr, job_ptr->alloc_node, SRUN_TIMEOUT,
				   msg_arg, job_ptr->start_protocol_ver);
	}


	step_iterator = list_iterator_create(job_ptr->step_list);
	while ((step_ptr = (struct step_record *) list_next(step_iterator)))
		srun_step_timeout(step_ptr, job_ptr->end_time);
	list_iterator_destroy(step_iterator);
}

extern void msg_aggr_sender_init(char *host, uint16_t port, uint64_t window,
				 uint64_t max_msg_cnt)
{
	if (msg_collection.running || (max_msg_cnt <= 1))
		return;

	memset(&msg_collection, 0, sizeof(msg_collection_type_t));

	slurm_mutex_init(&msg_collection.aggr_mutex);
	slurm_mutex_init(&msg_collection.mutex);

	slurm_mutex_lock(&msg_collection.mutex);
	slurm_mutex_lock(&msg_collection.aggr_mutex);
	slurm_cond_init(&msg_collection.cond, NULL);
	slurm_set_addr(&msg_collection.node_addr, port, host);
	msg_collection.window = window;
	msg_collection.max_msg_cnt = max_msg_cnt;
	msg_collection.msg_aggr_list = list_create(_msg_aggr_free);
	msg_collection.msg_list = list_create(slurm_free_comp_msg_list);
	msg_collection.max_msgs = false;
	msg_collection.debug_flags = slurm_get_debug_flags();
	slurm_mutex_unlock(&msg_collection.aggr_mutex);
	slurm_mutex_unlock(&msg_collection.mutex);

	slurm_thread_create(&msg_collection.thread_id,
			    &_msg_aggregation_sender, NULL);
}

/* Ping primary ControlMachine
 * RET 0 if no error */
static int _ping_controller(void)
{
	int rc;
	slurm_msg_t req;
	/* Locks: Read configuration */
	slurmctld_lock_t config_read_lock = {
		READ_LOCK, NO_LOCK, NO_LOCK, NO_LOCK };

	/*
	 *  Set address of controller to ping
	 */
	slurm_msg_t_init(&req);
	lock_slurmctld(config_read_lock);
	debug3("pinging slurmctld at %s", slurmctld_conf.control_addr);
	slurm_set_addr(&req.address, slurmctld_conf.slurmctld_port,
	               slurmctld_conf.control_addr);
	unlock_slurmctld(config_read_lock);

	req.msg_type = REQUEST_PING;

	if (slurm_send_recv_rc_msg_only_one(&req, &rc, 0) < 0) {
		error("_ping_controller/slurm_send_node_msg error: %m");
		return SLURM_ERROR;
	}

	if (rc) {
		error("_ping_controller/response error %d", rc);
		return SLURM_PROTOCOL_ERROR;
	}

	return SLURM_PROTOCOL_SUCCESS;
}

/*
 * srun_job_complete - notify srun of a job's termination
 * IN job_ptr - pointer to the slurmctld job record
 */
extern void srun_job_complete (struct job_record *job_ptr)
{
	slurm_addr_t * addr;
	srun_job_complete_msg_t *msg_arg;
	ListIterator step_iterator;
	struct step_record *step_ptr;

	xassert(job_ptr);

	if (job_ptr->other_port && job_ptr->alloc_node && job_ptr->resp_host) {
		addr = xmalloc(sizeof(struct sockaddr_in));
		slurm_set_addr(addr, job_ptr->other_port, job_ptr->resp_host);
		msg_arg = xmalloc(sizeof(srun_job_complete_msg_t));
		msg_arg->job_id   = job_ptr->job_id;
		msg_arg->step_id  = NO_VAL;
		_srun_agent_launch(addr, job_ptr->alloc_node,
				   SRUN_JOB_COMPLETE, msg_arg);
	}

	step_iterator = list_iterator_create(job_ptr->step_list);
	while ((step_ptr = (struct step_record *) list_next(step_iterator))) {
		if (step_ptr->batch_step)	/* batch script itself */
			continue;
		srun_step_complete(step_ptr);
	}
	list_iterator_destroy(step_iterator);
}

/* srun_ping - ping all srun commands that have not been heard from recently */
extern void srun_ping (void)
{
	ListIterator job_iterator;
	struct job_record *job_ptr;
	slurm_addr_t * addr;
	time_t now = time(NULL);
	time_t old = now - (slurmctld_conf.inactive_limit / 3) +
			   slurmctld_conf.msg_timeout + 1;
	srun_ping_msg_t *msg_arg;

	if (slurmctld_conf.inactive_limit == 0)
		return;		/* No limit, don't bother pinging */

	job_iterator = list_iterator_create(job_list);
	while ((job_ptr = (struct job_record *) list_next(job_iterator))) {
		xassert (job_ptr->magic == JOB_MAGIC);

		if (!IS_JOB_RUNNING(job_ptr))
			continue;

		if ((job_ptr->time_last_active <= old) && job_ptr->other_port
		    &&  job_ptr->alloc_node && job_ptr->resp_host) {
			addr = xmalloc(sizeof(struct sockaddr_in));
			slurm_set_addr(addr, job_ptr->other_port,
				job_ptr->resp_host);
			msg_arg = xmalloc(sizeof(srun_ping_msg_t));
			msg_arg->job_id  = job_ptr->job_id;
			msg_arg->step_id = NO_VAL;
			_srun_agent_launch(addr, job_ptr->alloc_node,
					   SRUN_PING, msg_arg);
		}
	}

	list_iterator_destroy(job_iterator);
}

/* Forward keypair info to other tasks as required.
 * Clear message forward structure upon completion.
 * The messages are forwarded sequentially. */
static int _forward_comm_set(struct kvs_comm_set *kvs_set_ptr)
{
	int i, rc = SLURM_SUCCESS;
	int tmp_host_cnt = kvs_set_ptr->host_cnt;
	slurm_msg_t msg_send;
	int msg_rc;

	kvs_set_ptr->host_cnt = 0;
	for (i=0; i<tmp_host_cnt; i++) {
		if (kvs_set_ptr->kvs_host_ptr[i].port == 0)
			continue;	/* empty */
		slurm_msg_t_init(&msg_send);
		msg_send.msg_type = PMI_KVS_GET_RESP;
		msg_send.data = (void *) kvs_set_ptr;
		slurm_set_addr(&msg_send.address,
			kvs_set_ptr->kvs_host_ptr[i].port,
			kvs_set_ptr->kvs_host_ptr[i].hostname);
		if (slurm_send_recv_rc_msg_only_one(&msg_send,
				&msg_rc, 0) < 0) {
			error("Could not forward msg to %s",
				kvs_set_ptr->kvs_host_ptr[i].hostname);
			msg_rc = 1;
		}
		rc = MAX(rc, msg_rc);
		xfree(kvs_set_ptr->kvs_host_ptr[i].hostname);
	}
	xfree(kvs_set_ptr->kvs_host_ptr);
	return rc;
}

/*
 * srun_exec - request that srun execute a specific command
 *	and route it's output to stdout
 * IN step_ptr - pointer to the slurmctld job step record
 * IN argv - command and arguments to execute
 */
extern void srun_exec(struct step_record *step_ptr, char **argv)
{
	slurm_addr_t * addr;
	srun_exec_msg_t *msg_arg;
	int cnt = 1, i;

	xassert(step_ptr);

	if (step_ptr->port && step_ptr->host && step_ptr->host[0]) {
		for (i=0; argv[i]; i++)
			cnt++;	/* start at 1 to include trailing NULL */
		addr = xmalloc(sizeof(struct sockaddr_in));
		slurm_set_addr(addr, step_ptr->port, step_ptr->host);
		msg_arg = xmalloc(sizeof(srun_exec_msg_t));
		msg_arg->job_id  = step_ptr->job_ptr->job_id;
		msg_arg->step_id = step_ptr->step_id;
		msg_arg->argc    = cnt;
		msg_arg->argv    = xmalloc(sizeof(char *) * cnt);
		for (i=0; i<cnt ; i++)
			msg_arg->argv[i] = xstrdup(argv[i]);
		_srun_agent_launch(addr, step_ptr->host, SRUN_EXEC,
				   msg_arg);
	} else {
		error("srun_exec %u.%u lacks communication channel",
			step_ptr->job_ptr->job_id, step_ptr->step_id);
	}
}

extern void
slurm_setup_remote_working_cluster(resource_allocation_response_msg_t *msg)
{
	xassert(msg);
	xassert(msg->working_cluster_rec);
	xassert(msg->node_list);
	xassert(msg->node_addr);

	if (working_cluster_rec)
		slurmdb_destroy_cluster_rec(working_cluster_rec);

	working_cluster_rec = (slurmdb_cluster_rec_t *)msg->working_cluster_rec;
	msg->working_cluster_rec = NULL;

	working_cluster_rec->plugin_id_select =
		select_get_plugin_id_pos(working_cluster_rec->plugin_id_select);

	slurm_set_addr(&working_cluster_rec->control_addr,
		       working_cluster_rec->control_port,
		       working_cluster_rec->control_host);

	if (setenvf(NULL, "SLURM_CLUSTER_NAME", "%s",
		    working_cluster_rec->name) < 0)
		error("unable to set SLURM_CLUSTER_NAME in environment");

	add_remote_nodes_to_conf_tbls(msg->node_list, msg->node_addr);
}

extern void term_msg_thread(void)
{
	slurm_mutex_lock(&thread_flag_mutex);
	if (thread_running) {
		int fd;
		slurm_addr_t addr;

		thread_shutdown = true;

		/* Open and close a connection to the plugin listening port.
		 * Allows slurm_accept_msg_conn() to return in _msg_thread()
		 * so that it can check the thread_shutdown flag.
		 */
		slurm_set_addr(&addr, nonstop_comm_port, "localhost");
		fd = slurm_open_stream(&addr, true);
		if (fd != -1) {
			/* we don't care if the open failed */
			slurm_close(fd);
		}

		debug2("waiting for slurmctld/nonstop thread to exit");
		pthread_join(msg_thread_id, NULL);
		msg_thread_id = 0;
		thread_shutdown = false;
		thread_running = false;
		debug2("join of slurmctld/nonstop thread was successful");
	}
	slurm_mutex_unlock(&thread_flag_mutex);
}

static int
_setup_srun_tree_info(const mpi_plugin_client_info_t *job)
{
	char *p;
	uint16_t p_port;

	memset(&tree_info, 0, sizeof(tree_info));

	tree_info.this_node = "launcher"; /* not used */
	tree_info.parent_id = -2;   /* not used */
	tree_info.parent_node = NULL; /* not used */
	tree_info.num_children = job_info.nnodes;
	tree_info.depth = 0;	 /* not used */
	tree_info.max_depth = 0; /* not used */
	/* pmi_port set in _setup_srun_sockets */
	p = getenv(PMI2_SPAWNER_PORT_ENV);
	if (p) {		/* spawned */
		p_port = atoi(p);
		tree_info.srun_addr = xmalloc(sizeof(slurm_addr_t));
		/* assume there is always a lo interface */
		slurm_set_addr(tree_info.srun_addr, p_port, "127.0.0.1");
	} else
		tree_info.srun_addr = NULL;

	snprintf(tree_sock_addr, 128, PMI2_SOCK_ADDR_FMT,
		 job->jobid, job->stepid);

	/* init kvs seq to 0. TODO: reduce array size */
	tree_info.children_kvs_seq = xmalloc(sizeof(uint32_t) *
					     job_info.nnodes);

	return SLURM_SUCCESS;
}

static void *_msg_thread(void *x)
{
	struct msg_arg *msg_arg_ptr = (struct msg_arg *) x;
	int rc, success = 0, timeout;
	slurm_msg_t msg_send;

	slurm_msg_t_init(&msg_send);

	debug2("KVS_Barrier msg to %s:%u",
		msg_arg_ptr->bar_ptr->hostname,
		msg_arg_ptr->bar_ptr->port);
	msg_send.msg_type = PMI_KVS_GET_RESP;
	msg_send.data = (void *) msg_arg_ptr->kvs_ptr;
	slurm_set_addr(&msg_send.address,
		msg_arg_ptr->bar_ptr->port,
		msg_arg_ptr->bar_ptr->hostname);

	timeout = slurm_get_msg_timeout() * 10000;
	if (slurm_send_recv_rc_msg_only_one(&msg_send, &rc, timeout) < 0) {
		error("slurm_send_recv_rc_msg_only_one: %m");
	} else if (rc != SLURM_SUCCESS) {
		error("KVS_Barrier confirm from %s, rc=%d",
			msg_arg_ptr->bar_ptr->hostname, rc);
	} else {
		/* successfully transmitted KVS keypairs */
		success = 1;
	}

	slurm_mutex_lock(&agent_mutex);
	agent_cnt--;
	pthread_cond_signal(&agent_cond);
	slurm_mutex_unlock(&agent_mutex);
	xfree(x);
	return NULL;
}

/*****************************************************************************\
 * terminate message hander thread
\*****************************************************************************/
extern void term_msg_thread(void)
{
	pthread_mutex_lock(&thread_flag_mutex);
	if (thread_running) {
		int fd;
		slurm_addr_t addr;

		thread_shutdown = true;

		/* Open and close a connection to the listening port.
		 * Allows slurm_accept_msg_conn() to return in
		 * _msg_thread() so that it can check the thread_shutdown
		 * flag.
		 */
		slurm_set_addr(&addr, sched_port, "localhost");
		fd = slurm_open_stream(&addr);
		if (fd != -1) {
			/* we don't care if the open failed */
			slurm_close_stream(fd);
		}

		debug2("waiting for dynalloc thread to exit");
		pthread_join(msg_thread_id, NULL);
		msg_thread_id = 0;
		thread_shutdown = false;
		thread_running = false;
		debug2("join of dynalloc thread successful");
	}
	pthread_mutex_unlock(&thread_flag_mutex);
}

/*
 * srun_timeout - notify srun of a job's imminent timeout
 * IN job_ptr - pointer to the slurmctld job record
 */
extern void srun_timeout (struct job_record *job_ptr)
{
	slurm_addr_t * addr;
	srun_timeout_msg_t *msg_arg;
	ListIterator step_iterator;
	struct step_record *step_ptr;

	xassert(job_ptr);
	if (!IS_JOB_RUNNING(job_ptr))
		return;

	if (job_ptr->other_port && job_ptr->alloc_node && job_ptr->resp_host) {
		addr = xmalloc(sizeof(struct sockaddr_in));
		slurm_set_addr(addr, job_ptr->other_port, job_ptr->resp_host);
		msg_arg = xmalloc(sizeof(srun_timeout_msg_t));
		msg_arg->job_id   = job_ptr->job_id;
		msg_arg->step_id  = NO_VAL;
		msg_arg->timeout  = job_ptr->end_time;
		_srun_agent_launch(addr, job_ptr->alloc_node, SRUN_TIMEOUT,
				   msg_arg);
	}


	step_iterator = list_iterator_create(job_ptr->step_list);
	while ((step_ptr = (struct step_record *) list_next(step_iterator))) {
		if ( (step_ptr->port    == 0)    ||
		     (step_ptr->host    == NULL) ||
		     (step_ptr->batch_step)      ||
		     (step_ptr->host[0] == '\0') )
			continue;
		addr = xmalloc(sizeof(struct sockaddr_in));
		slurm_set_addr(addr, step_ptr->port, step_ptr->host);
		msg_arg = xmalloc(sizeof(srun_timeout_msg_t));
		msg_arg->job_id   = job_ptr->job_id;
		msg_arg->step_id  = step_ptr->step_id;
		msg_arg->timeout  = job_ptr->end_time;
		_srun_agent_launch(addr, step_ptr->host, SRUN_TIMEOUT,
				   msg_arg);
	}
	list_iterator_destroy(step_iterator);
}