/* Convert an array of task IDs into a list of host names
* RET: the string, caller must xfree() this value */
static char *_task_ids_to_host_list(int ntasks, uint32_t taskids[])
{
int i;
hostset_t hs;
char *hosts;
slurm_step_layout_t *sl;
if ((sl = launch_common_get_slurm_step_layout(local_srun_job)) == NULL)
return (xstrdup("Unknown"));
hs = hostset_create(NULL);
for (i = 0; i < ntasks; i++) {
char *host = slurm_step_layout_host_name(sl, taskids[i]);
if (host) {
hostset_insert(hs, host);
free(host);
} else {
error("Could not identify host name for task %u",
taskids[i]);
}
}
hosts = _hostset_to_string(hs);
hostset_destroy(hs);
return (hosts);
}
/* Convert an array of task IDs into a list of host names
* RET: the string, caller must xfree() this value */
static char *_task_ids_to_host_list(int ntasks, uint32_t *taskids)
{
int i, task_cnt = 0;
hostset_t hs;
char *hosts;
slurm_step_layout_t *sl;
if ((sl = launch_common_get_slurm_step_layout(local_srun_job)) == NULL)
return (xstrdup("Unknown"));
/* If overhead of determining the hostlist is too high then srun
* communications will timeout and fail, so return "Unknown" instead.
*
* See slurm_step_layout_host_id() in src/common/slurm_step_layout.c
* for details. */
for (i = 0; i < sl->node_cnt; i++) {
task_cnt += sl->tasks[i];
}
if (task_cnt > 100000)
return (xstrdup("Unknown"));
hs = hostset_create(NULL);
for (i = 0; i < ntasks; i++) {
char *host = slurm_step_layout_host_name(sl, taskids[i]);
if (host) {
hostset_insert(hs, host);
free(host);
} else {
error("Could not identify host name for task %u",
taskids[i]);
}
}
hosts = _hostset_to_string(hs);
hostset_destroy(hs);
return (hosts);
}
/*
* slurm_pack_job_will_run - determine if a heterogenous job would execute
* immediately if submitted now
* IN job_req_list - List of job_desc_msg_t structures describing the resource
* allocation request
* RET SLURM_SUCCESS on success, otherwise return SLURM_ERROR with errno set
*/
extern int slurm_pack_job_will_run(List job_req_list)
{
job_desc_msg_t *req;
will_run_response_msg_t *will_run_resp;
char buf[64], *sep = "";
int rc = SLURM_SUCCESS, inx = 0;
ListIterator iter, itr;
time_t first_start = (time_t) 0;
uint32_t first_job_id = 0, tot_proc_count = 0, *job_id_ptr;
hostset_t hs = NULL;
char *job_list = NULL;
if (!job_req_list || (list_count(job_req_list) == 0)) {
error("No job descriptors input");
return SLURM_ERROR;
}
iter = list_iterator_create(job_req_list);
while ((req = (job_desc_msg_t *) list_next(iter))) {
will_run_resp = NULL;
rc = slurm_job_will_run2(req, &will_run_resp);
if (will_run_resp)
print_multi_line_string(
will_run_resp->job_submit_user_msg,
inx, LOG_LEVEL_INFO);
if ((rc == SLURM_SUCCESS) && will_run_resp) {
if (first_job_id == 0)
first_job_id = will_run_resp->job_id;
if ((first_start == 0) ||
(first_start < will_run_resp->start_time))
first_start = will_run_resp->start_time;
tot_proc_count += will_run_resp->proc_cnt;
if (hs)
hostset_insert(hs, will_run_resp->node_list);
else
hs = hostset_create(will_run_resp->node_list);
if (will_run_resp->preemptee_job_id) {
itr = list_iterator_create(will_run_resp->
preemptee_job_id);
while ((job_id_ptr = list_next(itr))) {
if (job_list)
sep = ",";
xstrfmtcat(job_list, "%s%u", sep,
*job_id_ptr);
}
list_iterator_destroy(itr);
}
slurm_free_will_run_response_msg(will_run_resp);
}
if (rc != SLURM_SUCCESS)
break;
inx++;
}
list_iterator_destroy(iter);
if (rc == SLURM_SUCCESS) {
char node_list[1028] = "";
if (hs)
hostset_ranged_string(hs, sizeof(node_list), node_list);
slurm_make_time_str(&first_start, buf, sizeof(buf));
info("Job %u to start at %s using %u processors on %s",
first_job_id, buf, tot_proc_count, node_list);
if (job_list)
info(" Preempts: %s", job_list);
}
if (hs)
hostset_destroy(hs);
xfree(job_list);
return rc;
}
/*
* slurm_pack_job_will_run - determine if a heterogenous job would execute
* immediately if submitted now
* IN job_req_list - List of job_desc_msg_t structures describing the resource
* allocation request
* RET 0 on success, otherwise return -1 and set errno to indicate the error
*/
extern int slurm_pack_job_will_run(List job_req_list)
{
job_desc_msg_t *req;
will_run_response_msg_t *will_run_resp;
char buf[64], local_hostname[64] = "", *sep = "";
int rc = SLURM_SUCCESS;
char *type = "processors";
ListIterator iter, itr;
time_t first_start = (time_t) 0;
uint32_t first_job_id = 0, tot_proc_count = 0, *job_id_ptr;
hostset_t hs = NULL;
char *job_list = NULL;
if (!job_req_list || (list_count(job_req_list) == 0)) {
error("No job descriptors input");
return SLURM_ERROR;
}
(void) gethostname_short(local_hostname, sizeof(local_hostname));
iter = list_iterator_create(job_req_list);
while ((req = (job_desc_msg_t *) list_next(iter))) {
if ((req->alloc_node == NULL) && local_hostname[0])
req->alloc_node = local_hostname;
will_run_resp = NULL;
rc = slurm_job_will_run2(req, &will_run_resp);
if ((rc == SLURM_SUCCESS) && will_run_resp) {
if (first_job_id == 0)
first_job_id = will_run_resp->job_id;
if ((first_start == 0) ||
(first_start < will_run_resp->start_time))
first_start = will_run_resp->start_time;
tot_proc_count += will_run_resp->proc_cnt;
if (hs)
hostset_insert(hs, will_run_resp->node_list);
else
hs = hostset_create(will_run_resp->node_list);
if (will_run_resp->preemptee_job_id) {
itr = list_iterator_create(will_run_resp->
preemptee_job_id);
while ((job_id_ptr = list_next(itr))) {
if (job_list)
sep = ",";
xstrfmtcat(job_list, "%s%u", sep,
*job_id_ptr);
}
list_iterator_destroy(itr);
}
slurm_free_will_run_response_msg(will_run_resp);
}
if (req->alloc_node == local_hostname)
req->alloc_node = NULL;
if (rc != SLURM_SUCCESS)
break;
}
list_iterator_destroy(iter);
if (rc == SLURM_SUCCESS) {
uint32_t cluster_flags = slurmdb_setup_cluster_flags();
char node_list[1028] = "";
if (cluster_flags & CLUSTER_FLAG_BG)
type = "cnodes";
if (hs)
hostset_ranged_string(hs, sizeof(node_list), node_list);
slurm_make_time_str(&first_start, buf, sizeof(buf));
info("Job %u to start at %s using %u %s on %s",
first_job_id, buf, tot_proc_count, type, node_list);
if (job_list)
info(" Preempts: %s", job_list);
}
if (hs)
hostset_destroy(hs);
xfree(job_list);
return rc;
}
/*
* parse_command_line
*/
extern void
parse_command_line( int argc, char* argv[] )
{
char *env_val = NULL;
bool override_format_env = false;
int opt_char;
int option_index;
static struct option long_options[] = {
{"accounts", required_argument, 0, 'A'},
{"all", no_argument, 0, 'a'},
{"format", required_argument, 0, 'o'},
{"help", no_argument, 0, OPT_LONG_HELP},
{"hide", no_argument, 0, OPT_LONG_HIDE},
{"iterate", required_argument, 0, 'i'},
{"jobs", optional_argument, 0, 'j'},
{"long", no_argument, 0, 'l'},
{"cluster", required_argument, 0, 'M'},
{"clusters", required_argument, 0, 'M'},
{"node", required_argument, 0, 'n'},
{"nodes", required_argument, 0, 'n'},
{"noheader", no_argument, 0, 'h'},
{"partitions", required_argument, 0, 'p'},
{"qos", required_argument, 0, 'q'},
{"reservation",required_argument, 0, 'R'},
{"sort", required_argument, 0, 'S'},
{"start", no_argument, 0, OPT_LONG_START},
{"steps", optional_argument, 0, 's'},
{"states", required_argument, 0, 't'},
{"usage", no_argument, 0, OPT_LONG_USAGE},
{"user", required_argument, 0, 'u'},
{"users", required_argument, 0, 'u'},
{"verbose", no_argument, 0, 'v'},
{"version", no_argument, 0, 'V'},
{NULL, 0, 0, 0}
};
if (getenv("SQUEUE_ALL"))
params.all_flag = true;
if ( ( env_val = getenv("SQUEUE_SORT") ) )
params.sort = xstrdup(env_val);
if ( ( env_val = getenv("SLURM_CLUSTERS") ) ) {
if (!(params.clusters = slurmdb_get_info_cluster(env_val))) {
error("'%s' can't be reached now, "
"or it is an invalid entry for "
"SLURM_CLUSTERS. Use 'sacctmgr --list "
"cluster' to see avaliable clusters.",
env_val);
exit(1);
}
working_cluster_rec = list_peek(params.clusters);
}
while ((opt_char = getopt_long(argc, argv,
"A:ahi:j::ln:M:o:p:q:R:s::S:t:u:U:vV",
long_options, &option_index)) != -1) {
switch (opt_char) {
case (int)'?':
fprintf(stderr, "Try \"squeue --help\" "
"for more information\n");
exit(1);
case (int) 'A':
case (int) 'U': /* backwards compatibility */
xfree(params.accounts);
params.accounts = xstrdup(optarg);
params.account_list =
_build_str_list( params.accounts );
break;
case (int)'a':
params.all_flag = true;
break;
case (int)'h':
params.no_header = true;
break;
case (int) 'i':
params.iterate= atoi(optarg);
if (params.iterate <= 0) {
error ("--iterate=%s\n", optarg);
exit(1);
}
break;
case (int) 'j':
if (optarg) {
params.jobs = xstrdup(optarg);
params.job_list =
_build_job_list(params.jobs);
}
params.job_flag = true;
break;
case (int) 'l':
params.long_list = true;
override_format_env = true;
break;
case (int) 'M':
if (params.clusters)
list_destroy(params.clusters);
if (!(params.clusters =
slurmdb_get_info_cluster(optarg))) {
error("'%s' can't be reached now, "
"or it is an invalid entry for "
"--cluster. Use 'sacctmgr --list "
"cluster' to see avaliable clusters.",
optarg);
exit(1);
}
working_cluster_rec = list_peek(params.clusters);
break;
case (int) 'n':
if (params.nodes)
hostset_destroy(params.nodes);
params.nodes = hostset_create(optarg);
if (params.nodes == NULL) {
error("'%s' invalid entry for --nodes",
optarg);
exit(1);
}
break;
case (int) 'o':
xfree(params.format);
params.format = xstrdup(optarg);
override_format_env = true;
break;
case (int) 'p':
xfree(params.partitions);
params.partitions = xstrdup(optarg);
params.part_list =
_build_str_list( params.partitions );
params.all_flag = true;
break;
case (int) 'q':
xfree(params.qoss);
params.qoss = xstrdup(optarg);
params.qos_list =
_build_str_list( params.qoss );
break;
case (int) 'R':
xfree(params.reservation);
params.reservation = xstrdup(optarg);
break;
case (int) 's':
if (optarg) {
params.steps = xstrdup(optarg);
params.step_list =
_build_step_list(params.steps);
}
params.step_flag = true;
override_format_env = true;
break;
case (int) 'S':
xfree(params.sort);
params.sort = xstrdup(optarg);
break;
case (int) 't':
xfree(params.states);
params.states = xstrdup(optarg);
params.state_list =
_build_state_list( params.states );
break;
case (int) 'u':
xfree(params.users);
params.users = xstrdup(optarg);
params.user_list =
_build_user_list( params.users );
break;
case (int) 'v':
params.verbose++;
break;
case (int) 'V':
print_slurm_version();
exit(0);
case OPT_LONG_HELP:
_help();
exit(0);
case OPT_LONG_HIDE:
params.all_flag = false;
break;
case OPT_LONG_START:
params.start_flag = true;
break;
case OPT_LONG_USAGE:
_usage();
exit(0);
}
}
if ( override_format_env == false ) {
if ( ( env_val = getenv("SQUEUE_FORMAT") ) )
params.format = xstrdup(env_val);
}
params.cluster_flags = slurmdb_setup_cluster_flags();
if (optind < argc) {
if (params.job_flag) {
params.jobs = xstrdup(argv[optind++]);
params.job_list = _build_job_list(params.jobs);
} else if (params.step_flag) {
params.steps = xstrdup(argv[optind++]);
params.step_list = _build_step_list(params.steps);
}
if (optind < argc) {
error("Unrecognized option: %s",argv[optind]);
_usage();
exit(1);
}
}
if ( params.job_flag && params.step_flag) {
if (params.job_list) {
verbose("Printing job steps with job filter");
params.job_flag = false;
} else {
error("Incompatible options --jobs and --steps");
exit(1);
}
}
if ( params.nodes ) {
char *name1 = NULL;
char *name2 = NULL;
hostset_t nodenames = hostset_create(NULL);
if (nodenames == NULL)
fatal("malloc failure");
while ( hostset_count(params.nodes) > 0 ) {
name1 = hostset_pop(params.nodes);
/* localhost = use current host name */
if ( strcasecmp("localhost", name1) == 0 ) {
name2 = xmalloc(128);
gethostname_short(name2, 128);
} else {
/* translate NodeHostName to NodeName */
name2 = slurm_conf_get_nodename(name1);
/* use NodeName if translation failed */
if ( name2 == NULL )
name2 = xstrdup(name1);
}
hostset_insert(nodenames, name2);
free(name1);
xfree(name2);
}
/* Replace params.nodename with the new one */
hostset_destroy(params.nodes);
params.nodes = nodenames;
}
if ( ( params.accounts == NULL ) &&
( env_val = getenv("SQUEUE_ACCOUNT") ) ) {
params.accounts = xstrdup(env_val);
params.account_list = _build_str_list( params.accounts );
}
if ( ( params.partitions == NULL ) &&
( env_val = getenv("SQUEUE_PARTITION") ) ) {
params.partitions = xstrdup(env_val);
params.part_list = _build_str_list( params.partitions );
params.all_flag = true;
}
if ( ( params.qoss == NULL ) &&
( env_val = getenv("SQUEUE_QOS") ) ) {
params.qoss = xstrdup(env_val);
params.qos_list = _build_str_list( params.qoss );
}
if ( ( params.states == NULL ) &&
( env_val = getenv("SQUEUE_STATES") ) ) {
params.states = xstrdup(env_val);
params.state_list = _build_state_list( params.states );
}
if ( ( params.users == NULL ) &&
( env_val = getenv("SQUEUE_USERS") ) ) {
params.users = xstrdup(env_val);
params.user_list = _build_user_list( params.users );
}
if ( params.start_flag && !params.step_flag ) {
/* Set more defaults */
if (params.format == NULL)
params.format = xstrdup("%.7i %.9P %.8j %.8u %.2t %.19S %.6D %R");
if (params.sort == NULL)
params.sort = xstrdup("S");
if (params.states == NULL) {
params.states = xstrdup("PD");
params.state_list = _build_state_list( params.states );
}
}
params.max_cpus = _max_cpus_per_node();
if ( params.verbose )
_print_options();
}
