/* _print_job_step - print the specified job step's information */
static int
_print_job_steps( bool clear_old )
{
int error_code;
static job_step_info_response_msg_t * old_step_ptr = NULL;
static job_step_info_response_msg_t * new_step_ptr;
uint16_t show_flags = 0;
if (params.all_flag)
show_flags |= SHOW_ALL;
if (old_step_ptr) {
if (clear_old)
old_step_ptr->last_update = 0;
/* Use a last_update time of 0 so that we can get an updated
* run_time for jobs rather than just its start_time */
error_code = slurm_get_job_steps((time_t) 0, NO_VAL, NO_VAL,
&new_step_ptr, show_flags);
if (error_code == SLURM_SUCCESS)
slurm_free_job_step_info_response_msg( old_step_ptr );
else if (slurm_get_errno () == SLURM_NO_CHANGE_IN_DATA) {
error_code = SLURM_SUCCESS;
new_step_ptr = old_step_ptr;
}
} else {
error_code = slurm_get_job_steps((time_t) 0, NO_VAL, NO_VAL,
&new_step_ptr, show_flags);
}
if (error_code) {
slurm_perror ("slurm_get_job_steps error");
return SLURM_ERROR;
}
old_step_ptr = new_step_ptr;
if (params.verbose) {
printf ("last_update_time=%ld records=%u\n",
(long) new_step_ptr->last_update,
new_step_ptr->job_step_count);
}
if (!params.format && !params.format_long)
params.format = "%.15i %.8j %.9P %.8u %.9M %N";
if (!params.format_list) {
if (params.format)
parse_format(params.format);
else if (params.format_long)
parse_long_format(params.format_long);
}
print_steps_array( new_step_ptr->job_steps,
new_step_ptr->job_step_count,
params.format_list );
return SLURM_SUCCESS;
}
/*
* slurm_terminate_job_step - terminates a job step by sending a
* REQUEST_TERMINATE_TASKS rpc to all slurmd of a job step.
* IN job_id - the job's id
* IN step_id - the job step's id - use SLURM_BATCH_SCRIPT as the step_id
* to terminate a job's batch script
* RET 0 on success, otherwise return -1 and set errno to indicate the error
*/
extern int
slurm_terminate_job_step (uint32_t job_id, uint32_t step_id)
{
resource_allocation_response_msg_t *alloc_info = NULL;
job_step_info_response_msg_t *step_info = NULL;
int rc = 0;
int i;
int save_errno = 0;
if (slurm_allocation_lookup_lite(job_id, &alloc_info)) {
return -1;
}
/*
* The controller won't give us info about the batch script job step,
* so we need to handle that seperately.
*/
if (step_id == SLURM_BATCH_SCRIPT) {
rc = _terminate_batch_script_step(alloc_info);
slurm_free_resource_allocation_response_msg(alloc_info);
errno = rc;
return rc ? -1 : 0;
}
/*
* Otherwise, look through the list of job step info and find
* the one matching step_id. Terminate that step.
*/
rc = slurm_get_job_steps((time_t)0, job_id, step_id,
&step_info, SHOW_ALL);
if (rc != 0) {
save_errno = errno;
goto fail;
}
for (i = 0; i < step_info->job_step_count; i++) {
if ((step_info->job_steps[i].job_id == job_id) &&
(step_info->job_steps[i].step_id == step_id)) {
rc = _terminate_job_step(&step_info->job_steps[i],
alloc_info);
save_errno = errno;
break;
}
}
slurm_free_job_step_info_response_msg(step_info);
fail:
slurm_free_resource_allocation_response_msg(alloc_info);
errno = save_errno;
return rc ? -1 : 0;
}
/* Return the current time limit of the specified job/step_id or NO_VAL if the
* information is not available */
static uint32_t _get_step_time(uint32_t job_id, uint32_t step_id)
{
uint32_t time_limit = NO_VAL;
int i, rc;
job_step_info_response_msg_t *resp;
rc = slurm_get_job_steps((time_t) 0, job_id, step_id, &resp, SHOW_ALL);
if (rc == SLURM_SUCCESS) {
for (i = 0; i < resp->job_step_count; i++) {
if ((resp->job_steps[i].job_id != job_id) ||
(resp->job_steps[i].step_id != step_id))
continue; /* should not happen */
time_limit = resp->job_steps[i].time_limit;
break;
}
slurm_free_job_step_info_response_msg(resp);
} else {
error("Could not load state information for step %u.%u: %m",
job_id, step_id);
}
return time_limit;
}
int main(int argc, char **argv)
{
ListIterator itr = NULL;
uint32_t req_cpufreq = NO_VAL;
uint32_t stepid = NO_VAL;
slurmdb_selected_step_t *selected_step = NULL;
#ifdef HAVE_ALPS_CRAY
error("The sstat command is not supported on Cray systems");
return 1;
#endif
#ifdef HAVE_BG
error("The sstat command is not supported on IBM BlueGene systems");
return 1;
#endif
slurm_conf_init(NULL);
print_fields_list = list_create(NULL);
print_fields_itr = list_iterator_create(print_fields_list);
parse_command_line(argc, argv);
if (!params.opt_job_list || !list_count(params.opt_job_list)) {
error("You didn't give me any jobs to stat.");
return 1;
}
print_fields_header(print_fields_list);
itr = list_iterator_create(params.opt_job_list);
while ((selected_step = list_next(itr))) {
char *nodelist = NULL;
bool free_nodelist = false;
if (selected_step->stepid == INFINITE) {
/* get the batch step info */
job_info_msg_t *job_ptr = NULL;
hostlist_t hl;
if (slurm_load_job(
&job_ptr, selected_step->jobid, SHOW_ALL)) {
error("couldn't get info for job %u",
selected_step->jobid);
continue;
}
stepid = NO_VAL;
hl = hostlist_create(job_ptr->job_array[0].nodes);
nodelist = hostlist_pop(hl);
free_nodelist = true;
hostlist_destroy(hl);
slurm_free_job_info_msg(job_ptr);
} else if (selected_step->stepid != NO_VAL) {
stepid = selected_step->stepid;
} else if (params.opt_all_steps) {
job_step_info_response_msg_t *step_ptr = NULL;
int i = 0;
if (slurm_get_job_steps(
0, selected_step->jobid, NO_VAL,
&step_ptr, SHOW_ALL)) {
error("couldn't get steps for job %u",
selected_step->jobid);
continue;
}
for (i = 0; i < step_ptr->job_step_count; i++) {
_do_stat(selected_step->jobid,
step_ptr->job_steps[i].step_id,
step_ptr->job_steps[i].nodes,
step_ptr->job_steps[i].cpu_freq);
}
slurm_free_job_step_info_response_msg(step_ptr);
continue;
} else {
/* get the first running step to query against. */
job_step_info_response_msg_t *step_ptr = NULL;
if (slurm_get_job_steps(
0, selected_step->jobid, NO_VAL,
&step_ptr, SHOW_ALL)) {
error("couldn't get steps for job %u",
selected_step->jobid);
continue;
}
if (!step_ptr->job_step_count) {
error("no steps running for job %u",
selected_step->jobid);
continue;
}
stepid = step_ptr->job_steps[0].step_id;
nodelist = step_ptr->job_steps[0].nodes;
req_cpufreq = step_ptr->job_steps[0].cpu_freq;
}
_do_stat(selected_step->jobid, stepid, nodelist, req_cpufreq);
if (free_nodelist && nodelist)
free(nodelist);
}
list_iterator_destroy(itr);
xfree(params.opt_field_list);
if (params.opt_job_list)
list_destroy(params.opt_job_list);
if (print_fields_itr)
list_iterator_destroy(print_fields_itr);
if (print_fields_list)
list_destroy(print_fields_list);
return 0;
}
/*
* scontrol_print_step - print the specified job step's information
* IN job_step_id_str - job step's id or NULL to print information
* about all job steps
*/
extern void
scontrol_print_step (char *job_step_id_str)
{
int error_code, i, print_cnt = 0;
uint32_t job_id = NO_VAL, step_id = NO_VAL;
uint16_t array_id = (uint16_t) NO_VAL;
char *next_str;
job_step_info_response_msg_t *job_step_info_ptr;
job_step_info_t * job_step_ptr;
static uint32_t last_job_id = 0, last_array_id, last_step_id = 0;
static job_step_info_response_msg_t *old_job_step_info_ptr = NULL;
static uint16_t last_show_flags = 0xffff;
uint16_t show_flags = 0;
if (job_step_id_str) {
job_id = (uint32_t) strtol (job_step_id_str, &next_str, 10);
if (next_str[0] == '_')
array_id = (uint16_t) strtol(next_str+1, &next_str, 10);
if (next_str[0] == '.')
step_id = (uint32_t) strtol (next_str+1, NULL, 10);
}
if (all_flag)
show_flags |= SHOW_ALL;
if ((old_job_step_info_ptr) && (last_job_id == job_id) &&
(last_array_id == array_id) && (last_step_id == step_id)) {
if (last_show_flags != show_flags)
old_job_step_info_ptr->last_update = (time_t) 0;
error_code = slurm_get_job_steps (
old_job_step_info_ptr->last_update,
job_id, step_id, &job_step_info_ptr,
show_flags);
if (error_code == SLURM_SUCCESS)
slurm_free_job_step_info_response_msg (
old_job_step_info_ptr);
else if (slurm_get_errno () == SLURM_NO_CHANGE_IN_DATA) {
job_step_info_ptr = old_job_step_info_ptr;
error_code = SLURM_SUCCESS;
if (quiet_flag == -1)
printf ("slurm_get_job_steps no change in data\n");
}
} else {
if (old_job_step_info_ptr) {
slurm_free_job_step_info_response_msg (
old_job_step_info_ptr);
old_job_step_info_ptr = NULL;
}
error_code = slurm_get_job_steps ( (time_t) 0, job_id, step_id,
&job_step_info_ptr,
show_flags);
}
if (error_code) {
exit_code = 1;
if (quiet_flag != 1)
slurm_perror ("slurm_get_job_steps error");
return;
}
old_job_step_info_ptr = job_step_info_ptr;
last_show_flags = show_flags;
last_job_id = job_id;
last_step_id = step_id;
if (quiet_flag == -1) {
char time_str[32];
slurm_make_time_str ((time_t *)&job_step_info_ptr->last_update,
time_str, sizeof(time_str));
printf ("last_update_time=%s, records=%d\n",
time_str, job_step_info_ptr->job_step_count);
}
job_step_ptr = job_step_info_ptr->job_steps ;
for (i = 0, job_step_ptr = job_step_info_ptr->job_steps;
i < job_step_info_ptr->job_step_count; i++, job_step_ptr++) {
if ((array_id != (uint16_t) NO_VAL) &&
(array_id != job_step_ptr->array_task_id))
continue;
slurm_print_job_step_info(stdout, job_step_ptr, one_liner);
print_cnt++;
}
if (print_cnt == 0) {
if (job_step_id_str) {
exit_code = 1;
if (quiet_flag != 1) {
if (array_id == (uint16_t) NO_VAL) {
printf ("Job step %u.%u not found\n",
job_id, step_id);
} else {
printf ("Job step %u_%u.%u not found\n",
job_id, array_id, step_id);
}
}
} else if (quiet_flag != 1)
printf ("No job steps in the system\n");
}
}