static void _set_freq(int type, char *freq, char *freq_def)
{
if ((acct_gather_profile_timer[type].freq =
acct_gather_parse_freq(type, freq)) == -1)
if ((acct_gather_profile_timer[type].freq =
acct_gather_parse_freq(type, freq_def)) == -1)
acct_gather_profile_timer[type].freq = 0;
}
extern int acct_gather_check_acct_freq_task(uint64_t job_mem_lim,
char *acctg_freq)
{
int task_freq;
static uint32_t acct_freq_task = NO_VAL;
if (acct_freq_task == NO_VAL) {
char *acct_freq = slurm_get_jobacct_gather_freq();
int i = acct_gather_parse_freq(PROFILE_TASK, acct_freq);
xfree(acct_freq);
/* If the value is -1 lets set the freq to something
really high so we don't check this again.
*/
if (i == -1)
acct_freq_task = (uint16_t)NO_VAL;
else
acct_freq_task = i;
}
if (!job_mem_lim || !acct_freq_task)
return 0;
task_freq = acct_gather_parse_freq(PROFILE_TASK, acctg_freq);
if (task_freq == -1)
return 0;
if (task_freq == 0) {
error("Can't turn accounting frequency off. "
"We need it to monitor memory usage.");
slurm_seterrno(ESLURMD_INVALID_ACCT_FREQ);
return 1;
} else if (task_freq > acct_freq_task) {
error("Can't set frequency to %d, it is higher than %u. "
"We need it to be at least at this level to "
"monitor memory usage.",
task_freq, acct_freq_task);
slurm_seterrno(ESLURMD_INVALID_ACCT_FREQ);
return 1;
}
return 0;
}
static int _check_acct_freq_task(uint32_t job_mem_lim, char *acctg_freq)
{
int task_freq;
if (!job_mem_lim || !conf->acct_freq_task)
return 0;
task_freq = acct_gather_parse_freq(PROFILE_TASK, acctg_freq);
if (task_freq == -1)
return 0;
if ((task_freq == 0) || (task_freq > conf->acct_freq_task)) {
error("Can't set frequency to %d, it is higher than %u. "
"We need it to be at least at this level to "
"monitor memory usage.",
task_freq, conf->acct_freq_task);
slurm_seterrno (ESLURMD_INVALID_ACCT_FREQ);
return 1;
}
return 0;
}
/*
* 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();
}
static slurmd_conf_t *read_slurmd_conf_lite(int fd)
{
int rc;
int len;
Buf buffer = NULL;
slurmd_conf_t *confl, *local_conf = NULL;
int tmp_int = 0;
/* First check to see if we've already initialized the
* global slurmd_conf_t in 'conf'. Allocate memory if not.
*/
if (conf) {
confl = conf;
} else {
local_conf = xmalloc(sizeof(slurmd_conf_t));
confl = local_conf;
}
safe_read(fd, &len, sizeof(int));
buffer = init_buf(len);
safe_read(fd, buffer->head, len);
rc = unpack_slurmd_conf_lite_no_alloc(confl, buffer);
if (rc == SLURM_ERROR)
fatal("slurmstepd: problem with unpack of slurmd_conf");
free_buf(buffer);
confl->log_opts.prefix_level = 1;
confl->log_opts.stderr_level = confl->debug_level;
confl->log_opts.logfile_level = confl->debug_level;
confl->log_opts.syslog_level = confl->debug_level;
/*
* If daemonizing, turn off stderr logging -- also, if
* logging to a file, turn off syslog.
*
* Otherwise, if remaining in foreground, turn off logging
* to syslog (but keep logfile level)
*/
if (confl->daemonize) {
confl->log_opts.stderr_level = LOG_LEVEL_QUIET;
if (confl->logfile)
confl->log_opts.syslog_level = LOG_LEVEL_QUIET;
} else
confl->log_opts.syslog_level = LOG_LEVEL_QUIET;
confl->acct_freq_task = (uint16_t)NO_VAL;
tmp_int = acct_gather_parse_freq(PROFILE_TASK,
confl->job_acct_gather_freq);
if (tmp_int != -1)
confl->acct_freq_task = tmp_int;
return (confl);
rwfail:
FREE_NULL_BUFFER(buffer);
xfree(local_conf);
return (NULL);
}
