static int
eliminate_nodes (char **hosts)
{
hostlist_t hl = NULL;
hostlist_t hlnew = NULL;
hostlist_iterator_t hitr = NULL;
ipmidetect_t id = NULL;
char *host = NULL;
char hostbuf[HOSTLIST_BUFLEN + 1];
int rv = -1;
assert (hosts);
assert (*hosts);
if (!(id = ipmidetect_handle_create ()))
{
fprintf (stderr,
"ipmidetect_handle_create\n");
goto cleanup;
}
if (ipmidetect_load_data (id,
NULL,
0,
0) < 0)
{
if (ipmidetect_errnum (id) == IPMIDETECT_ERR_CONNECT
|| ipmidetect_errnum (id) == IPMIDETECT_ERR_CONNECT_TIMEOUT)
fprintf (stderr,
"Error connecting to ipmidetect daemon\n");
else
fprintf (stderr,
"ipmidetect_load_data: %s\n", ipmidetect_errormsg (id));
goto cleanup;
}
if (!(hl = hostlist_create (*hosts)))
{
fprintf (stderr,
"hostlist_create: %s\n",
strerror (errno));
goto cleanup;
}
if (!(hlnew = hostlist_create (*hosts)))
{
fprintf (stderr,
"hostlist_create: %s\n",
strerror (errno));
goto cleanup;
}
if (!(hitr = hostlist_iterator_create (hl)))
{
fprintf (stderr,
"hostlist_iterator_create: %s\n",
strerror (errno));
goto cleanup;
}
while ((host = hostlist_next (hitr)))
{
int ret;
if ((ret = ipmidetect_is_node_detected (id, host)) < 0)
{
if (ipmidetect_errnum (id) == IPMIDETECT_ERR_NOTFOUND)
fprintf (stderr,
"Node '%s' unrecognized by ipmidetect\n", host);
else
fprintf (stderr,
"ipmidetect_is_node_detected: %s\n", ipmidetect_errormsg (id));
goto cleanup;
}
if (!ret)
hostlist_delete (hlnew, host);
free (host);
}
host = NULL;
if (!hostlist_count (hlnew))
{
rv = 0;
goto cleanup;
}
memset (hostbuf, '\0', HOSTLIST_BUFLEN + 1);
if (hostlist_ranged_string (hlnew, HOSTLIST_BUFLEN, hostbuf) < 0)
{
fprintf (stderr,
"hostlist_ranged_string: truncation\n");
goto cleanup;
}
free (*hosts);
if (!(*hosts = strdup (hostbuf)))
{
fprintf (stderr, "strdup: %s\n", strerror (errno));
goto cleanup;
}
rv = hostlist_count (hlnew);
cleanup:
if (id)
ipmidetect_handle_destroy (id);
if (hitr)
hostlist_iterator_destroy (hitr);
if (hl)
hostlist_destroy (hl);
if (hlnew)
hostlist_destroy (hlnew);
free (host);
return (rv);
}
/*
* Create an srun job structure for a step w/out an allocation response msg.
* (i.e. inside an allocation)
*/
srun_job_t *
job_step_create_allocation(resource_allocation_response_msg_t *resp)
{
uint32_t job_id = resp->job_id;
srun_job_t *job = NULL;
allocation_info_t *ai = xmalloc(sizeof(*ai));
hostlist_t hl = NULL;
char *buf = NULL;
int count = 0;
uint32_t alloc_count = 0;
ai->jobid = job_id;
ai->stepid = NO_VAL;
ai->nodelist = opt.alloc_nodelist;
hl = hostlist_create(ai->nodelist);
hostlist_uniq(hl);
alloc_count = hostlist_count(hl);
ai->nnodes = alloc_count;
hostlist_destroy(hl);
if (opt.exc_nodes) {
hostlist_t exc_hl = hostlist_create(opt.exc_nodes);
hostlist_t inc_hl = NULL;
char *node_name = NULL;
hl = hostlist_create(ai->nodelist);
if(opt.nodelist) {
inc_hl = hostlist_create(opt.nodelist);
}
hostlist_uniq(hl);
//info("using %s or %s", opt.nodelist, ai->nodelist);
while ((node_name = hostlist_shift(exc_hl))) {
int inx = hostlist_find(hl, node_name);
if (inx >= 0) {
debug("excluding node %s", node_name);
hostlist_delete_nth(hl, inx);
ai->nnodes--; /* decrement node count */
}
if(inc_hl) {
inx = hostlist_find(inc_hl, node_name);
if (inx >= 0) {
error("Requested node %s is also "
"in the excluded list.",
node_name);
error("Job not submitted.");
hostlist_destroy(exc_hl);
hostlist_destroy(inc_hl);
goto error;
}
}
free(node_name);
}
hostlist_destroy(exc_hl);
/* we need to set this here so if there are more nodes
* available than we requested we can set it
* straight. If there is no exclude list then we set
* the vars then.
*/
if (!opt.nodes_set) {
/* we don't want to set the number of nodes =
* to the number of requested processes unless we
* know it is less than the number of nodes
* in the allocation
*/
if(opt.ntasks_set && (opt.ntasks < ai->nnodes))
opt.min_nodes = opt.ntasks;
else
opt.min_nodes = ai->nnodes;
opt.nodes_set = true;
}
if(!opt.max_nodes)
opt.max_nodes = opt.min_nodes;
if((opt.max_nodes > 0) && (opt.max_nodes < ai->nnodes))
ai->nnodes = opt.max_nodes;
count = hostlist_count(hl);
if(!count) {
error("Hostlist is now nothing! Can't run job.");
hostlist_destroy(hl);
goto error;
}
if(inc_hl) {
count = hostlist_count(inc_hl);
if(count < ai->nnodes) {
/* add more nodes to get correct number for
allocation */
hostlist_t tmp_hl = hostlist_copy(hl);
int i=0;
int diff = ai->nnodes - count;
buf = hostlist_ranged_string_xmalloc(inc_hl);
hostlist_delete(tmp_hl, buf);
xfree(buf);
while ((node_name = hostlist_shift(tmp_hl)) &&
(i < diff)) {
hostlist_push(inc_hl, node_name);
i++;
}
hostlist_destroy(tmp_hl);
}
buf = hostlist_ranged_string_xmalloc(inc_hl);
hostlist_destroy(inc_hl);
xfree(opt.nodelist);
opt.nodelist = buf;
} else {
if (count > ai->nnodes) {
/* remove more nodes than needed for
allocation */
int i=0;
for (i=count; i>ai->nnodes; i--)
hostlist_delete_nth(hl, i);
}
xfree(opt.nodelist);
opt.nodelist = hostlist_ranged_string_xmalloc(hl);
}
hostlist_destroy(hl);
} else {
if (!opt.nodes_set) {
/* we don't want to set the number of nodes =
* to the number of requested processes unless we
* know it is less than the number of nodes
* in the allocation
*/
if(opt.ntasks_set && (opt.ntasks < ai->nnodes))
opt.min_nodes = opt.ntasks;
else
opt.min_nodes = ai->nnodes;
opt.nodes_set = true;
}
if(!opt.max_nodes)
opt.max_nodes = opt.min_nodes;
if((opt.max_nodes > 0) && (opt.max_nodes < ai->nnodes))
ai->nnodes = opt.max_nodes;
/* Don't reset the ai->nodelist because that is the
* nodelist we want to say the allocation is under
* opt.nodelist is what is used for the allocation.
*/
/* xfree(ai->nodelist); */
/* ai->nodelist = xstrdup(buf); */
}
/* get the correct number of hosts to run tasks on */
if (opt.nodelist) {
hl = hostlist_create(opt.nodelist);
if (opt.distribution != SLURM_DIST_ARBITRARY)
hostlist_uniq(hl);
if (!hostlist_count(hl)) {
error("Hostlist is now nothing! Can not run job.");
hostlist_destroy(hl);
goto error;
}
buf = hostlist_ranged_string_xmalloc(hl);
count = hostlist_count(hl);
hostlist_destroy(hl);
/* Don't reset the ai->nodelist because that is the
* nodelist we want to say the allocation is under
* opt.nodelist is what is used for the allocation.
*/
/* xfree(ai->nodelist); */
/* ai->nodelist = xstrdup(buf); */
xfree(opt.nodelist);
opt.nodelist = buf;
}
if (opt.distribution == SLURM_DIST_ARBITRARY) {
if (count != opt.ntasks) {
error("You asked for %d tasks but specified %d nodes",
opt.ntasks, count);
goto error;
}
}
if (ai->nnodes == 0) {
error("No nodes in allocation, can't run job");
goto error;
}
ai->num_cpu_groups = resp->num_cpu_groups;
ai->cpus_per_node = resp->cpus_per_node;
ai->cpu_count_reps = resp->cpu_count_reps;
/* info("looking for %d nodes out of %s with a must list of %s", */
/* ai->nnodes, ai->nodelist, opt.nodelist); */
/*
* Create job
*/
job = _job_create_structure(ai);
error:
xfree(ai);
return (job);
}