/*
* Send a msg to the next msg aggregation collector node. If primary
* collector is unavailable or returns error, try backup collector.
* If backup collector is unavailable or returns error, send msg
* directly to controller.
*/
static int _send_to_next_collector(slurm_msg_t *msg)
{
slurm_addr_t *next_dest = NULL;
bool i_am_collector;
int rc = SLURM_SUCCESS;
if (msg_collection.debug_flags & DEBUG_FLAG_ROUTE)
info("msg aggr: send_to_next_collector: getting primary next "
"collector");
if ((next_dest = route_g_next_collector(&i_am_collector))) {
if (msg_collection.debug_flags & DEBUG_FLAG_ROUTE) {
char addrbuf[100];
slurm_print_slurm_addr(next_dest, addrbuf, 32);
info("msg aggr: send_to_next_collector: *next_dest is "
"%s", addrbuf);
}
memcpy(&msg->address, next_dest, sizeof(slurm_addr_t));
rc = slurm_send_only_node_msg(msg);
}
if (!next_dest || (rc != SLURM_SUCCESS))
rc = _send_to_backup_collector(msg, rc);
return rc;
}
static int _send_to_backup_collector(slurm_msg_t *msg, int rc)
{
slurm_addr_t *next_dest = NULL;
if (msg_collection.debug_flags & DEBUG_FLAG_ROUTE) {
info("_send_to_backup_collector: primary %s, "
"getting backup",
rc ? "can't be reached" : "is null");
}
if ((next_dest = route_g_next_collector_backup())) {
if (msg_collection.debug_flags & DEBUG_FLAG_ROUTE) {
char addrbuf[100];
slurm_print_slurm_addr(next_dest, addrbuf, 32);
info("_send_to_backup_collector: *next_dest is "
"%s", addrbuf);
}
memcpy(&msg->address, next_dest, sizeof(slurm_addr_t));
rc = slurm_send_only_node_msg(msg);
}
if (!next_dest || (rc != SLURM_SUCCESS)) {
if (msg_collection.debug_flags & DEBUG_FLAG_ROUTE)
info("_send_to_backup_collector: backup %s, "
"sending msg to controller",
rc ? "can't be reached" : "is null");
rc = slurm_send_only_controller_msg(msg, working_cluster_rec);
}
return rc;
}
int p_mpi_hook_slurmstepd_task(const mpi_plugin_task_info_t *job,
char ***env)
{
char addrbuf[1024];
char *p;
char *addr = getenvp(*env, "SLURM_LAUNCH_NODE_IPADDR");
debug("Using mpi/mpich-gm");
slurm_print_slurm_addr (job->self, addrbuf, sizeof(addrbuf));
if ((p = strchr (addrbuf, ':')) != NULL)
*p = '\0';
env_array_overwrite_fmt(env, "GMPI_MASTER", "%s", addr);
env_array_overwrite_fmt(env, "GMPI_SLAVE", "%s", addrbuf);
env_array_overwrite_fmt(env, "GMPI_ID", "%u", job->gtaskid);
if (!getenv("GMPI_RECV")) {
env_array_overwrite_fmt(env, "GMPI_RECV", "%s", "hybrid");
}
env_array_overwrite_fmt(env, "MXMPI_MASTER", "%s", addr);
env_array_overwrite_fmt(env, "MXMPI_ID", "%u", job->gtaskid);
env_array_overwrite_fmt(env, "MXMPI_SLAVE", "%s", addrbuf);
if (!getenv("MXMPI_RECV")) {
env_array_overwrite_fmt(env, "MXMPI_RECV", "%s", "hybrid");
}
debug2("init for mpi rank %u", job->gtaskid);
return SLURM_SUCCESS;
}
extern void msg_aggr_resp(slurm_msg_t *msg)
{
slurm_msg_t *next_msg;
composite_msg_t *comp_msg;
msg_aggr_t *msg_aggr;
ListIterator itr;
comp_msg = (composite_msg_t *)msg->data;
itr = list_iterator_create(comp_msg->msg_list);
if (msg_collection.debug_flags & DEBUG_FLAG_ROUTE)
info("msg_aggr_resp: processing composite msg_list...");
while ((next_msg = list_next(itr))) {
switch (next_msg->msg_type) {
case RESPONSE_SLURM_RC:
/* signal sending thread that slurmctld received this
* epilog complete msg */
if (msg_collection.debug_flags & DEBUG_FLAG_ROUTE)
info("msg_aggr_resp: rc message found for "
"index %u signaling sending thread",
next_msg->msg_index);
slurm_mutex_lock(&msg_collection.aggr_mutex);
if (!(msg_aggr = _handle_msg_aggr_ret(
next_msg->msg_index, 1))) {
debug2("msg_aggr_resp: error: unable to "
"locate aggr message struct for job %u",
next_msg->msg_index);
slurm_mutex_unlock(&msg_collection.aggr_mutex);
continue;
}
pthread_cond_signal(&msg_aggr->wait_cond);
slurm_mutex_unlock(&msg_collection.aggr_mutex);
break;
case RESPONSE_MESSAGE_COMPOSITE:
comp_msg = (composite_msg_t *)next_msg->data;
/* set up the address here for the next node */
memcpy(&next_msg->address, &comp_msg->sender,
sizeof(slurm_addr_t));
if (msg_collection.debug_flags & DEBUG_FLAG_ROUTE) {
char addrbuf[100];
slurm_print_slurm_addr(&next_msg->address,
addrbuf, 32);
info("msg_aggr_resp: composite response msg "
"found for %s", addrbuf);
}
slurm_send_only_node_msg(next_msg);
break;
default:
error("_rpc_composite_resp: invalid msg type in "
"composite msg_list");
break;
}
}
list_iterator_destroy(itr);
if (msg_collection.debug_flags & DEBUG_FLAG_ROUTE)
info("msg aggr: _rpc_composite_resp: finished processing "
"composite msg_list...");
}
/*
* _set_collectors call the split_hostlist API on the all nodes hostlist
* to set the node to be used as a collector for unsolicited node aggregation.
*
* If this node is a forwarding node (first node in any hostlist),
* then its collector and backup are the ControlMachine and it's backup.
*
* Otherwise, we find the hostlist containing this node.
* The forwarding node in that hostlist becomes a collector, the next node
* which is not this node becomes the backup.
* That list is split, we iterate through it and searching for a list in
* which this node is a forwarding node. If found, we set the collector and
* backup, else this process is repeated.
*/
static void _set_collectors(char *this_node_name)
{
slurm_ctl_conf_t *conf;
hostlist_t nodes;
hostlist_t* hll = NULL;
char *parent = NULL, *backup = NULL;
char addrbuf[32];
int i, j, f = -1;
int hl_count = 0;
uint16_t parent_port;
uint16_t backup_port;
bool found = false;
bool ctldparent = true;
#ifdef HAVE_FRONT_END
return; /* on a FrontEnd system this would never be useful. */
#endif
if (!run_in_daemon("slurmd"))
return; /* Only compute nodes have collectors */
/* Set the initial iteration, collector is controller,
* full list is split */
xassert(this_node_name);
conf = slurm_conf_lock();
nodes = _get_all_nodes();
parent = strdup(conf->control_addr);
if (conf->backup_addr) {
backup = strdup(conf->backup_addr);
}
parent_port = conf->slurmctld_port;
backup_port = parent_port;
slurm_conf_unlock();
while (!found) {
if ( route_g_split_hostlist(nodes, &hll, &hl_count) ) {
error("unable to split forward hostlist");
goto clean; /* collector addrs remains null */
}
/* Find which hostlist contains this node */
for (i=0; i < hl_count; i++) {
f = hostlist_find(hll[i], this_node_name);
if (f != -1)
break;
}
if (i == hl_count) {
fatal("ROUTE -- %s not found in node_record_table",
this_node_name);
}
if (f == 0) {
/* we are a forwarded to node,
* so our parent is parent */
if (hostlist_count(hll[i]) > 1)
this_is_collector = true;
xfree(msg_collect_node);
msg_collect_node = xmalloc(sizeof(slurm_addr_t));
if (ctldparent)
slurm_set_addr(msg_collect_node, parent_port,
parent);
else {
slurm_conf_get_addr(parent, msg_collect_node);
msg_collect_node->sin_port = htons(parent_port);
}
if (debug_flags & DEBUG_FLAG_ROUTE) {
slurm_print_slurm_addr(msg_collect_node,
addrbuf, 32);
info("ROUTE -- message collector address is %s",
addrbuf);
}
xfree(msg_collect_backup);
if (backup) {
msg_collect_backup =
xmalloc(sizeof(slurm_addr_t));
if (ctldparent) {
slurm_set_addr(msg_collect_backup,
backup_port, backup);
} else {
slurm_conf_get_addr(backup,
msg_collect_backup);
msg_collect_backup->sin_port =
htons(backup_port);
}
if (debug_flags & DEBUG_FLAG_ROUTE) {
slurm_print_slurm_addr(
msg_collect_backup,
addrbuf, 32);
info("ROUTE -- message collector backup"
" address is %s", addrbuf);
}
} else {
if (debug_flags & DEBUG_FLAG_ROUTE) {
info("ROUTE -- no message collector "
"backup");
}
}
found = true;
goto clean;
}
/* We are not a forwarding node, the first node in this list
* will split the forward_list.
* We also know that the forwarding node is not a controller.
*
* clean up parent context */
ctldparent = false;
hostlist_destroy(nodes);
if (parent)
free(parent);
if (backup)
free(backup);
nodes = hostlist_copy(hll[i]);
for (j=0; j < hl_count; j++) {
hostlist_destroy(hll[j]);
}
xfree(hll);
/* set our parent, backup, and continue search */
parent = hostlist_shift(nodes);
backup = hostlist_nth(nodes, 0);
if (strcmp(backup, this_node_name) == 0) {
free(backup);
backup = NULL;
if (hostlist_count(nodes) > 1)
backup = hostlist_nth(nodes, 1);
}
parent_port = slurm_conf_get_port(parent);
if (backup) {
backup_port = slurm_conf_get_port(backup);
} else
backup_port = 0;
}
clean:
if (debug_flags & DEBUG_FLAG_ROUTE) {
if (this_is_collector)
info("ROUTE -- %s is a collector node", this_node_name);
else
info("ROUTE -- %s is a leaf node", this_node_name);
}
hostlist_destroy(nodes);
if (parent)
free(parent);
if (backup)
free(backup);
for (i=0; i < hl_count; i++) {
hostlist_destroy(hll[i]);
}
xfree(hll);
}
int setup_env(env_t *env, bool preserve_env)
{
int rc = SLURM_SUCCESS;
char *dist = NULL, *lllp_dist = NULL;
char addrbuf[INET_ADDRSTRLEN];
uint32_t cluster_flags = slurmdb_setup_cluster_flags();
if (env == NULL)
return SLURM_ERROR;
if (env->task_pid
&& setenvf(&env->env, "SLURM_TASK_PID", "%d", (int)env->task_pid)) {
error("Unable to set SLURM_TASK_PID environment variable");
rc = SLURM_FAILURE;
}
if (!preserve_env && env->ntasks) {
if(setenvf(&env->env, "SLURM_NTASKS", "%d", env->ntasks)) {
error("Unable to set SLURM_NTASKS "
"environment variable");
rc = SLURM_FAILURE;
}
if(setenvf(&env->env, "SLURM_NPROCS", "%d", env->ntasks)) {
error("Unable to set SLURM_NPROCS "
"environment variable");
rc = SLURM_FAILURE;
}
}
if (env->cpus_per_task
&& setenvf(&env->env, "SLURM_CPUS_PER_TASK", "%d",
env->cpus_per_task) ) {
error("Unable to set SLURM_CPUS_PER_TASK");
rc = SLURM_FAILURE;
}
if (env->ntasks_per_node
&& setenvf(&env->env, "SLURM_NTASKS_PER_NODE", "%d",
env->ntasks_per_node) ) {
error("Unable to set SLURM_NTASKS_PER_NODE");
rc = SLURM_FAILURE;
}
if (env->ntasks_per_socket
&& setenvf(&env->env, "SLURM_NTASKS_PER_SOCKET", "%d",
env->ntasks_per_socket) ) {
error("Unable to set SLURM_NTASKS_PER_SOCKET");
rc = SLURM_FAILURE;
}
if (env->ntasks_per_core
&& setenvf(&env->env, "SLURM_NTASKS_PER_CORE", "%d",
env->ntasks_per_core) ) {
error("Unable to set SLURM_NTASKS_PER_CORE");
rc = SLURM_FAILURE;
}
if (env->cpus_on_node
&& setenvf(&env->env, "SLURM_CPUS_ON_NODE", "%d",
env->cpus_on_node) ) {
error("Unable to set SLURM_CPUS_ON_NODE");
rc = SLURM_FAILURE;
}
_set_distribution(env->distribution, &dist, &lllp_dist);
if(dist)
if (setenvf(&env->env, "SLURM_DISTRIBUTION", "%s", dist)) {
error("Can't set SLURM_DISTRIBUTION env variable");
rc = SLURM_FAILURE;
}
if(env->distribution == SLURM_DIST_PLANE)
if (setenvf(&env->env, "SLURM_DIST_PLANESIZE", "%u",
env->plane_size)) {
error("Can't set SLURM_DIST_PLANESIZE "
"env variable");
rc = SLURM_FAILURE;
}
if(lllp_dist)
if (setenvf(&env->env, "SLURM_DIST_LLLP", "%s", lllp_dist)) {
error("Can't set SLURM_DIST_LLLP env variable");
rc = SLURM_FAILURE;
}
if (env->cpu_bind_type) {
char *str_verbose, *str_bind_type, *str_bind_list;
char *str_bind;
int len;
if (env->batch_flag) {
unsetenvp(env->env, "SBATCH_CPU_BIND_VERBOSE");
unsetenvp(env->env, "SBATCH_CPU_BIND_TYPE");
unsetenvp(env->env, "SBATCH_CPU_BIND_LIST");
unsetenvp(env->env, "SBATCH_CPU_BIND");
} else {
unsetenvp(env->env, "SLURM_CPU_BIND_VERBOSE");
unsetenvp(env->env, "SLURM_CPU_BIND_TYPE");
unsetenvp(env->env, "SLURM_CPU_BIND_LIST");
unsetenvp(env->env, "SLURM_CPU_BIND");
}
str_verbose = xstrdup ("");
if (env->cpu_bind_type & CPU_BIND_VERBOSE) {
xstrcat(str_verbose, "verbose");
} else {
xstrcat(str_verbose, "quiet");
}
str_bind_type = xstrdup ("");
if (env->cpu_bind_type & CPU_BIND_TO_THREADS) {
xstrcat(str_bind_type, "threads,");
} else if (env->cpu_bind_type & CPU_BIND_TO_CORES) {
xstrcat(str_bind_type, "cores,");
} else if (env->cpu_bind_type & CPU_BIND_TO_SOCKETS) {
xstrcat(str_bind_type, "sockets,");
} else if (env->cpu_bind_type & CPU_BIND_TO_LDOMS) {
xstrcat(str_bind_type, "ldoms,");
}
if (env->cpu_bind_type & CPU_BIND_NONE) {
xstrcat(str_bind_type, "none");
} else if (env->cpu_bind_type & CPU_BIND_RANK) {
xstrcat(str_bind_type, "rank");
} else if (env->cpu_bind_type & CPU_BIND_MAP) {
xstrcat(str_bind_type, "map_cpu:");
} else if (env->cpu_bind_type & CPU_BIND_MASK) {
xstrcat(str_bind_type, "mask_cpu:");
} else if (env->cpu_bind_type & CPU_BIND_LDRANK) {
xstrcat(str_bind_type, "rank_ldom");
} else if (env->cpu_bind_type & CPU_BIND_LDMAP) {
xstrcat(str_bind_type, "map_ldom:");
} else if (env->cpu_bind_type & CPU_BIND_LDMASK) {
xstrcat(str_bind_type, "mask_ldom:");
}
len = strlen(str_bind_type);
if (len) { /* remove a possible trailing ',' */
if (str_bind_type[len-1] == ',') {
str_bind_type[len-1] = '\0';
}
}
str_bind_list = xstrdup ("");
if (env->cpu_bind) {
xstrcat(str_bind_list, env->cpu_bind);
}
str_bind = xstrdup ("");
xstrcat(str_bind, str_verbose);
if (str_bind[0] && str_bind_type && str_bind_type[0])
xstrcatchar(str_bind, ',');
xstrcat(str_bind, str_bind_type);
xstrcat(str_bind, str_bind_list);
if (env->batch_flag) {
if (setenvf(&env->env, "SBATCH_CPU_BIND_VERBOSE",
str_verbose)) {
error("Unable to set SBATCH_CPU_BIND_VERBOSE");
rc = SLURM_FAILURE;
}
if (setenvf(&env->env, "SBATCH_CPU_BIND_TYPE",
str_bind_type)) {
error("Unable to set SBATCH_CPU_BIND_TYPE");
rc = SLURM_FAILURE;
}
if (setenvf(&env->env, "SBATCH_CPU_BIND_LIST",
str_bind_list)) {
error("Unable to set SBATCH_CPU_BIND_LIST");
rc = SLURM_FAILURE;
}
if (setenvf(&env->env, "SBATCH_CPU_BIND", str_bind)) {
error("Unable to set SBATCH_CPU_BIND");
rc = SLURM_FAILURE;
}
} else {
if (setenvf(&env->env, "SLURM_CPU_BIND_VERBOSE",
str_verbose)) {
error("Unable to set SLURM_CPU_BIND_VERBOSE");
rc = SLURM_FAILURE;
}
if (setenvf(&env->env, "SLURM_CPU_BIND_TYPE",
str_bind_type)) {
error("Unable to set SLURM_CPU_BIND_TYPE");
rc = SLURM_FAILURE;
}
if (setenvf(&env->env, "SLURM_CPU_BIND_LIST",
str_bind_list)) {
error("Unable to set SLURM_CPU_BIND_LIST");
rc = SLURM_FAILURE;
}
if (setenvf(&env->env, "SLURM_CPU_BIND", str_bind)) {
error("Unable to set SLURM_CPU_BIND");
rc = SLURM_FAILURE;
}
}
}
if (env->mem_bind_type) {
char *str_verbose, *str_bind_type, *str_bind_list;
char *str_bind;
if (env->batch_flag) {
unsetenvp(env->env, "SBATCH_MEM_BIND_VERBOSE");
unsetenvp(env->env, "SBATCH_MEM_BIND_TYPE");
unsetenvp(env->env, "SBATCH_MEM_BIND_LIST");
unsetenvp(env->env, "SBATCH_MEM_BIND");
} else {
unsetenvp(env->env, "SLURM_MEM_BIND_VERBOSE");
unsetenvp(env->env, "SLURM_MEM_BIND_TYPE");
unsetenvp(env->env, "SLURM_MEM_BIND_LIST");
unsetenvp(env->env, "SLURM_MEM_BIND");
}
str_verbose = xstrdup ("");
if (env->mem_bind_type & MEM_BIND_VERBOSE) {
xstrcat(str_verbose, "verbose");
} else {
xstrcat(str_verbose, "quiet");
}
str_bind_type = xstrdup ("");
if (env->mem_bind_type & MEM_BIND_NONE) {
xstrcat(str_bind_type, "none");
} else if (env->mem_bind_type & MEM_BIND_RANK) {
xstrcat(str_bind_type, "rank");
} else if (env->mem_bind_type & MEM_BIND_MAP) {
xstrcat(str_bind_type, "map_mem:");
} else if (env->mem_bind_type & MEM_BIND_MASK) {
xstrcat(str_bind_type, "mask_mem:");
} else if (env->mem_bind_type & MEM_BIND_LOCAL) {
xstrcat(str_bind_type, "local");
}
str_bind_list = xstrdup ("");
if (env->mem_bind) {
xstrcat(str_bind_list, env->mem_bind);
}
str_bind = xstrdup ("");
xstrcat(str_bind, str_verbose);
if (str_bind[0]) { /* add ',' if str_verbose */
xstrcatchar(str_bind, ',');
}
xstrcat(str_bind, str_bind_type);
xstrcat(str_bind, str_bind_list);
if (env->batch_flag) {
if (setenvf(&env->env, "SBATCH_MEM_BIND_VERBOSE",
str_verbose)) {
error("Unable to set SBATCH_MEM_BIND_VERBOSE");
rc = SLURM_FAILURE;
}
if (setenvf(&env->env, "SBATCH_MEM_BIND_TYPE",
str_bind_type)) {
error("Unable to set SBATCH_MEM_BIND_TYPE");
rc = SLURM_FAILURE;
}
if (setenvf(&env->env, "SBATCH_MEM_BIND_LIST",
str_bind_list)) {
error("Unable to set SBATCH_MEM_BIND_LIST");
rc = SLURM_FAILURE;
}
if (setenvf(&env->env, "SBATCH_MEM_BIND", str_bind)) {
error("Unable to set SBATCH_MEM_BIND");
rc = SLURM_FAILURE;
}
} else {
if (setenvf(&env->env, "SLURM_MEM_BIND_VERBOSE",
str_verbose)) {
error("Unable to set SLURM_MEM_BIND_VERBOSE");
rc = SLURM_FAILURE;
}
if (setenvf(&env->env, "SLURM_MEM_BIND_TYPE",
str_bind_type)) {
error("Unable to set SLURM_MEM_BIND_TYPE");
rc = SLURM_FAILURE;
}
if (setenvf(&env->env, "SLURM_MEM_BIND_LIST",
str_bind_list)) {
error("Unable to set SLURM_MEM_BIND_LIST");
rc = SLURM_FAILURE;
}
if (setenvf(&env->env, "SLURM_MEM_BIND", str_bind)) {
error("Unable to set SLURM_MEM_BIND");
rc = SLURM_FAILURE;
}
}
}
if (env->overcommit
&& (setenvf(&env->env, "SLURM_OVERCOMMIT", "1"))) {
error("Unable to set SLURM_OVERCOMMIT environment variable");
rc = SLURM_FAILURE;
}
if (env->slurmd_debug
&& setenvf(&env->env, "SLURMD_DEBUG", "%d", env->slurmd_debug)) {
error("Can't set SLURMD_DEBUG environment variable");
rc = SLURM_FAILURE;
}
if (env->labelio
&& setenvf(&env->env, "SLURM_LABELIO", "1")) {
error("Unable to set SLURM_LABELIO environment variable");
rc = SLURM_FAILURE;
}
if (env->select_jobinfo) {
_setup_particulars(cluster_flags, &env->env,
env->select_jobinfo);
}
if (env->jobid >= 0) {
if (setenvf(&env->env, "SLURM_JOB_ID", "%d", env->jobid)) {
error("Unable to set SLURM_JOB_ID environment");
rc = SLURM_FAILURE;
}
/* and for backwards compatability... */
if (setenvf(&env->env, "SLURM_JOBID", "%d", env->jobid)) {
error("Unable to set SLURM_JOBID environment");
rc = SLURM_FAILURE;
}
}
if (env->nodeid >= 0
&& setenvf(&env->env, "SLURM_NODEID", "%d", env->nodeid)) {
error("Unable to set SLURM_NODEID environment");
rc = SLURM_FAILURE;
}
if (env->procid >= 0
&& setenvf(&env->env, "SLURM_PROCID", "%d", env->procid)) {
error("Unable to set SLURM_PROCID environment");
rc = SLURM_FAILURE;
}
if (env->localid >= 0
&& setenvf(&env->env, "SLURM_LOCALID", "%d", env->localid)) {
error("Unable to set SLURM_LOCALID environment");
rc = SLURM_FAILURE;
}
if (env->stepid >= 0
&& setenvf(&env->env, "SLURM_STEPID", "%d", env->stepid)) {
error("Unable to set SLURM_STEPID environment");
rc = SLURM_FAILURE;
}
if (!preserve_env && env->nhosts
&& setenvf(&env->env, "SLURM_NNODES", "%d", env->nhosts)) {
error("Unable to set SLURM_NNODES environment var");
rc = SLURM_FAILURE;
}
if (env->nodelist
&& setenvf(&env->env, "SLURM_NODELIST", "%s", env->nodelist)) {
error("Unable to set SLURM_NODELIST environment var.");
rc = SLURM_FAILURE;
}
if (!preserve_env && env->task_count
&& setenvf (&env->env,
"SLURM_TASKS_PER_NODE", "%s", env->task_count)) {
error ("Can't set SLURM_TASKS_PER_NODE env variable");
rc = SLURM_FAILURE;
}
if (env->comm_port
&& setenvf (&env->env, "SLURM_SRUN_COMM_PORT", "%u",
env->comm_port)) {
error ("Can't set SLURM_SRUN_COMM_PORT env variable");
rc = SLURM_FAILURE;
}
if (env->cli) {
slurm_print_slurm_addr (env->cli, addrbuf, INET_ADDRSTRLEN);
/*
* XXX: Eventually, need a function for slurm_addrs that
* returns just the IP address (not addr:port)
*/
if ((dist = strchr (addrbuf, ':')) != NULL)
*dist = '\0';
setenvf (&env->env, "SLURM_LAUNCH_NODE_IPADDR", "%s", addrbuf);
}
if (env->sgtids
&& setenvf(&env->env, "SLURM_GTIDS", "%s", env->sgtids)) {
error("Unable to set SLURM_GTIDS environment variable");
rc = SLURM_FAILURE;
}
if(cluster_flags & CLUSTER_FLAG_AIX) {
char res_env[128];
char *debug_env = (char *)getenv("SLURM_LL_API_DEBUG");
int debug_num = 0;
/* MP_POERESTART_ENV causes a warning message for "poe", but
* is needed for "poerestart". Presently we have no means to
* determine what command a user will execute. We could
* possibly add a "srestart" command which would set
* MP_POERESTART_ENV, but that presently seems unnecessary. */
/* setenvf(&env->env, "MP_POERESTART_ENV", res_env); */
if (debug_env)
debug_num = atoi(debug_env);
snprintf(res_env, sizeof(res_env), "SLURM_LL_API_DEBUG=%d",
debug_num);
/* Required for AIX/POE systems indicating pre-allocation */
setenvf(&env->env, "LOADLBATCH", "yes");
setenvf(&env->env, "LOADL_ACTIVE", "3.2.0");
}
if (env->pty_port
&& setenvf(&env->env, "SLURM_PTY_PORT", "%hu", env->pty_port)) {
error("Can't set SLURM_PTY_PORT env variable");
rc = SLURM_FAILURE;
}
if (env->ws_col
&& setenvf(&env->env, "SLURM_PTY_WIN_COL", "%hu", env->ws_col)) {
error("Can't set SLURM_PTY_WIN_COL env variable");
rc = SLURM_FAILURE;
}
if (env->ws_row
&& setenvf(&env->env, "SLURM_PTY_WIN_ROW", "%hu", env->ws_row)) {
error("Can't set SLURM_PTY_WIN_ROW env variable");
rc = SLURM_FAILURE;
}
if (env->ckpt_dir
&& setenvf(&env->env, "SLURM_CHECKPOINT_IMAGE_DIR", "%s",
env->ckpt_dir)) {
error("Can't set SLURM_CHECKPOINT_IMAGE_DIR env variable");
rc = SLURM_FAILURE;
}
if (env->restart_cnt &&
setenvf(&env->env, "SLURM_RESTART_COUNT", "%u", env->restart_cnt)) {
error("Can't set SLURM_RESTART_COUNT env variable");
rc = SLURM_FAILURE;
}
return rc;
}
