static char *
_guess_nodename()
{
char host[256];
char *nodename = NULL;
if (gethostname_short(host, 256) != 0)
return NULL;
nodename = slurm_conf_get_nodename(host);
if (nodename == NULL)
nodename = slurm_conf_get_aliased_nodename();
if (nodename == NULL) /* if no match, try localhost */
nodename = slurm_conf_get_nodename("localhost");
return nodename;
}
/*
* _find_alias_node_record - find a record for node with the alias of
* the specified name supplied
* input: name - name to be aliased of the desired node
* output: return pointer to node record or NULL if not found
* global: node_record_table_ptr - pointer to global node table
* node_hash_table - table of hash indexes
*/
static struct node_record *_find_alias_node_record (char *name)
{
int i;
char *alias = NULL;
if ((name == NULL) || (name[0] == '\0')) {
info("_find_alias_node_record: passed NULL name");
return NULL;
}
/* Get the alias we have just to make sure the user isn't
* trying to use the real hostname to run on something that has
* been aliased.
*/
alias = slurm_conf_get_nodename(name);
if (!alias)
return NULL;
/* try to find via hash table, if it exists */
if (node_hash_table) {
struct node_record *node_ptr;
i = _hash_index (alias);
node_ptr = node_hash_table[i];
while (node_ptr) {
xassert(node_ptr->magic == NODE_MAGIC);
if (!strcmp(node_ptr->name, alias)) {
xfree(alias);
return node_ptr;
}
node_ptr = node_ptr->node_next;
}
error ("_find_alias_node_record: lookup failure for %s", name);
}
/* revert to sequential search */
else {
for (i = 0; i < node_record_count; i++) {
if (!strcmp (alias, node_record_table_ptr[i].name)) {
xfree(alias);
return (&node_record_table_ptr[i]);
}
}
}
xfree(alias);
return (struct node_record *) NULL;
}
extern void create_daemon_popup(GtkAction *action, gpointer user_data)
{
GtkWidget *popup = gtk_dialog_new_with_buttons(
"SLURM Daemons running",
GTK_WINDOW(user_data),
GTK_DIALOG_DESTROY_WITH_PARENT,
GTK_STOCK_CLOSE,
GTK_RESPONSE_OK,
NULL);
int update = 0;
slurm_ctl_conf_info_msg_t *conf;
char me[MAX_SLURM_NAME], *b, *c, *n;
int actld = 0, ctld = 0, d = 0;
GtkTreeStore *treestore =
_local_create_treestore_2cols(popup, 300, 100);
GtkTreeIter iter;
g_signal_connect(G_OBJECT(popup), "delete_event",
G_CALLBACK(_delete_popup), NULL);
g_signal_connect(G_OBJECT(popup), "response",
G_CALLBACK(_delete_popup), NULL);
slurm_conf_init(NULL);
conf = slurm_conf_lock();
gethostname_short(me, MAX_SLURM_NAME);
if ((b = conf->backup_controller)) {
if ((strcmp(b, me) == 0) ||
(strcasecmp(b, "localhost") == 0))
ctld = 1;
}
if ((c = conf->control_machine)) {
actld = 1;
if ((strcmp(c, me) == 0) ||
(strcasecmp(c, "localhost") == 0))
ctld = 1;
}
slurm_conf_unlock();
if ((n = slurm_conf_get_nodename(me))) {
d = 1;
xfree(n);
} else if ((n = slurm_conf_get_aliased_nodename())) {
d = 1;
xfree(n);
} else if ((n = slurm_conf_get_nodename("localhost"))) {
d = 1;
xfree(n);
}
if (actld && ctld)
add_display_treestore_line(update, treestore, &iter,
"Slurmctld", "1");
if (actld && d)
add_display_treestore_line(update, treestore, &iter,
"Slurmd", "1");
gtk_widget_show_all(popup);
return;
}
/*
* Read the slurm configuration file (slurm.conf) and substitute some
* values into the slurmd configuration in preference of the defaults.
*/
static void
_read_config(void)
{
char *path_pubkey = NULL;
slurm_ctl_conf_t *cf = NULL;
uint16_t tmp16 = 0;
#ifndef HAVE_FRONT_END
bool cr_flag = false, gang_flag = false;
#endif
cf = slurm_conf_lock();
slurm_mutex_lock(&conf->config_mutex);
if (conf->conffile == NULL)
conf->conffile = xstrdup(cf->slurm_conf);
conf->slurm_user_id = cf->slurm_user_id;
conf->cr_type = cf->select_type_param;
path_pubkey = xstrdup(cf->job_credential_public_certificate);
if (!conf->logfile)
conf->logfile = xstrdup(cf->slurmd_logfile);
#ifndef HAVE_FRONT_END
if (!strcmp(cf->select_type, "select/cons_res"))
cr_flag = true;
if (cf->preempt_mode & PREEMPT_MODE_GANG)
gang_flag = true;
#endif
slurm_conf_unlock();
/* node_name may already be set from a command line parameter */
if (conf->node_name == NULL)
conf->node_name = slurm_conf_get_nodename(conf->hostname);
/* if we didn't match the form of the hostname already
* stored in conf->hostname, check to see if we match any
* valid aliases */
if (conf->node_name == NULL)
conf->node_name = slurm_conf_get_aliased_nodename();
if (conf->node_name == NULL)
conf->node_name = slurm_conf_get_nodename("localhost");
if (conf->node_name == NULL)
fatal("Unable to determine this slurmd's NodeName");
_massage_pathname(&conf->logfile);
/* set node_addr if relevant */
if ((conf->node_addr == NULL) &&
(conf->node_addr = slurm_conf_get_nodeaddr(conf->hostname)) &&
(strcmp(conf->node_addr, conf->hostname) == 0)) {
xfree(conf->node_addr); /* Sets to NULL */
}
conf->port = slurm_conf_get_port(conf->node_name);
slurm_conf_get_cpus_bsct(conf->node_name,
&conf->conf_cpus, &conf->conf_boards,
&conf->conf_sockets, &conf->conf_cores,
&conf->conf_threads);
/* store hardware properties in slurmd_config */
xfree(conf->block_map);
xfree(conf->block_map_inv);
_update_logging();
_update_nice();
get_cpuinfo(&conf->actual_cpus,
&conf->actual_boards,
&conf->actual_sockets,
&conf->actual_cores,
&conf->actual_threads,
&conf->block_map_size,
&conf->block_map, &conf->block_map_inv);
#ifdef HAVE_FRONT_END
/*
* When running with multiple frontends, the slurmd S:C:T values are not
* relevant, hence ignored by both _register_front_ends (sets all to 1)
* and validate_nodes_via_front_end (uses slurm.conf values).
* Report actual hardware configuration, irrespective of FastSchedule.
*/
conf->cpus = conf->actual_cpus;
conf->boards = conf->actual_boards;
conf->sockets = conf->actual_sockets;
conf->cores = conf->actual_cores;
conf->threads = conf->actual_threads;
#else
/* If the actual resources on a node differ than what is in
* the configuration file and we are using
* cons_res or gang scheduling we have to use what is in the
* configuration file because the slurmctld creates bitmaps
* for scheduling before these nodes check in.
*/
if (((cf->fast_schedule == 0) && !cr_flag && !gang_flag) ||
((cf->fast_schedule == 1) &&
(conf->actual_cpus < conf->conf_cpus))) {
conf->cpus = conf->actual_cpus;
conf->boards = conf->actual_boards;
conf->sockets = conf->actual_sockets;
conf->cores = conf->actual_cores;
conf->threads = conf->actual_threads;
} else {
conf->cpus = conf->conf_cpus;
conf->boards = conf->conf_boards;
conf->sockets = conf->conf_sockets;
conf->cores = conf->conf_cores;
conf->threads = conf->conf_threads;
}
if ((conf->cpus != conf->actual_cpus) ||
(conf->sockets != conf->actual_sockets) ||
(conf->cores != conf->actual_cores) ||
(conf->threads != conf->actual_threads)) {
if (cf->fast_schedule) {
info("Node configuration differs from hardware: "
"CPUs=%u:%u(hw) Boards=%u:%u(hw) "
"SocketsPerBoard=%u:%u(hw) CoresPerSocket=%u:%u(hw) "
"ThreadsPerCore=%u:%u(hw)",
conf->cpus, conf->actual_cpus,
conf->boards, conf->actual_boards,
conf->sockets, conf->actual_sockets,
conf->cores, conf->actual_cores,
conf->threads, conf->actual_threads);
} else if ((cf->fast_schedule == 0) && (cr_flag || gang_flag)) {
error("You are using cons_res or gang scheduling with "
"Fastschedule=0 and node configuration differs "
"from hardware. The node configuration used "
"will be what is in the slurm.conf because of "
"the bitmaps the slurmctld must create before "
"the slurmd registers.\n"
" CPUs=%u:%u(hw) Boards=%u:%u(hw) "
"SocketsPerBoard=%u:%u(hw) CoresPerSocket=%u:%u(hw) "
"ThreadsPerCore=%u:%u(hw)",
conf->cpus, conf->actual_cpus,
conf->boards, conf->actual_boards,
conf->sockets, conf->actual_sockets,
conf->cores, conf->actual_cores,
conf->threads, conf->actual_threads);
}
}
#endif
get_memory(&conf->real_memory_size);
get_up_time(&conf->up_time);
cf = slurm_conf_lock();
get_tmp_disk(&conf->tmp_disk_space, cf->tmp_fs);
_free_and_set(&conf->epilog, xstrdup(cf->epilog));
_free_and_set(&conf->prolog, xstrdup(cf->prolog));
_free_and_set(&conf->tmpfs, xstrdup(cf->tmp_fs));
_free_and_set(&conf->health_check_program,
xstrdup(cf->health_check_program));
_free_and_set(&conf->spooldir, xstrdup(cf->slurmd_spooldir));
_massage_pathname(&conf->spooldir);
_free_and_set(&conf->pidfile, xstrdup(cf->slurmd_pidfile));
_massage_pathname(&conf->pidfile);
_free_and_set(&conf->select_type, xstrdup(cf->select_type));
_free_and_set(&conf->task_prolog, xstrdup(cf->task_prolog));
_free_and_set(&conf->task_epilog, xstrdup(cf->task_epilog));
_free_and_set(&conf->pubkey, path_pubkey);
conf->debug_flags = cf->debug_flags;
conf->propagate_prio = cf->propagate_prio_process;
_free_and_set(&conf->job_acct_gather_freq,
xstrdup(cf->job_acct_gather_freq));
conf->acct_freq_task = (uint16_t)NO_VAL;
tmp16 = acct_gather_parse_freq(PROFILE_TASK,
conf->job_acct_gather_freq);
if (tmp16 != -1)
conf->acct_freq_task = tmp16;
_free_and_set(&conf->acct_gather_energy_type,
xstrdup(cf->acct_gather_energy_type));
_free_and_set(&conf->acct_gather_filesystem_type,
xstrdup(cf->acct_gather_filesystem_type));
_free_and_set(&conf->acct_gather_infiniband_type,
xstrdup(cf->acct_gather_infiniband_type));
_free_and_set(&conf->acct_gather_profile_type,
xstrdup(cf->acct_gather_profile_type));
_free_and_set(&conf->job_acct_gather_type,
xstrdup(cf->job_acct_gather_type));
if ( (conf->node_name == NULL) ||
(conf->node_name[0] == '\0') )
fatal("Node name lookup failure");
if (cf->control_addr == NULL)
fatal("Unable to establish controller machine");
if (cf->slurmctld_port == 0)
fatal("Unable to establish controller port");
conf->slurmd_timeout = cf->slurmd_timeout;
conf->use_pam = cf->use_pam;
conf->task_plugin_param = cf->task_plugin_param;
slurm_mutex_unlock(&conf->config_mutex);
slurm_conf_unlock();
}
/*
* 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();
}