extern uint16_t
scontrol_get_job_state(uint32_t job_id)
{
job_info_msg_t * job_buffer_ptr = NULL;
int error_code = SLURM_SUCCESS, i;
job_info_t *job_ptr = NULL;
error_code = scontrol_load_job(&job_buffer_ptr, job_id);
if (error_code) {
exit_code = 1;
if (quiet_flag == -1)
slurm_perror ("slurm_load_job error");
return (uint16_t) NO_VAL;
}
if (quiet_flag == -1) {
char time_str[32];
slurm_make_time_str((time_t *)&job_buffer_ptr->last_update,
time_str, sizeof(time_str));
printf("last_update_time=%s, records=%d\n",
time_str, job_buffer_ptr->record_count);
}
job_ptr = job_buffer_ptr->job_array ;
for (i = 0; i < job_buffer_ptr->record_count; i++) {
if (job_ptr->job_id == job_id)
return job_ptr->job_state;
}
if (quiet_flag == -1)
printf("Could not find job %u", job_id);
return (uint16_t) NO_VAL;
}
/*
* scontrol_print_job - print the specified job's information
* IN job_id - job's id or NULL to print information about all jobs
*/
extern void
scontrol_print_job (char * job_id_str)
{
int error_code = SLURM_SUCCESS, i, print_cnt = 0;
uint32_t job_id = 0;
uint16_t array_id = (uint16_t) NO_VAL;
job_info_msg_t * job_buffer_ptr = NULL;
job_info_t *job_ptr = NULL;
char *end_ptr = NULL;
if (job_id_str) {
job_id = (uint32_t) strtol (job_id_str, &end_ptr, 10);
if (end_ptr[0] == '_')
array_id = strtol( end_ptr + 1, &end_ptr, 10 );
}
error_code = scontrol_load_job(&job_buffer_ptr, job_id);
if (error_code) {
exit_code = 1;
if (quiet_flag != 1)
slurm_perror ("slurm_load_jobs error");
return;
}
if (quiet_flag == -1) {
char time_str[32];
slurm_make_time_str ((time_t *)&job_buffer_ptr->last_update,
time_str, sizeof(time_str));
printf ("last_update_time=%s, records=%d\n",
time_str, job_buffer_ptr->record_count);
}
job_ptr = job_buffer_ptr->job_array ;
for (i = 0, job_ptr = job_buffer_ptr->job_array;
i < job_buffer_ptr->record_count; i++, job_ptr++) {
if ((array_id != (uint16_t) NO_VAL) &&
(array_id != job_ptr->array_task_id))
continue;
slurm_print_job_info(stdout, job_ptr, one_liner);
print_cnt++;
}
if (print_cnt == 0) {
if (job_id_str) {
exit_code = 1;
if (quiet_flag != 1) {
if (array_id == (uint16_t) NO_VAL) {
printf("Job %u not found\n", job_id);
} else {
printf("Job %u_%u not found\n",
job_id, array_id);
}
}
} else if (quiet_flag != 1)
printf ("No jobs in the system\n");
}
}
/* Translate a job name to relevant job IDs
* NOTE: xfree the return value to avoid memory leak */
static char *_job_name2id(char *job_name, uint32_t job_uid)
{
int i, rc;
job_info_msg_t *resp;
slurm_job_info_t *job_ptr;
char *job_id_str = NULL, *sep = "";
xassert(job_name);
rc = scontrol_load_job(&resp, 0);
if (rc == SLURM_SUCCESS) {
if (resp->record_count == 0) {
error("JobName %s not found", job_name);
slurm_free_job_info_msg(resp);
return job_id_str;
}
for (i = 0, job_ptr = resp->job_array; i < resp->record_count;
i++, job_ptr++) {
if ((job_uid != NO_VAL) &&
(job_uid != job_ptr->user_id))
continue;
if (!job_ptr->name || xstrcmp(job_name, job_ptr->name))
continue;
if (job_ptr->array_task_id != NO_VAL) {
xstrfmtcat(job_id_str, "%s%u_%u", sep,
job_ptr->array_job_id,
job_ptr->array_task_id);
} else {
xstrfmtcat(job_id_str, "%s%u", sep,
job_ptr->job_id);
}
sep = ",";
}
if (!job_id_str) {
if (job_uid == NO_VAL) {
error("No jobs with name \'%s\'", job_name);
} else {
error("No jobs with user ID %u and name \'%s\'",
job_uid, job_name);
}
}
} else {
error("Could not load state information: %m");
}
return job_id_str;
}
/*
* scontrol_print_completing - print jobs in completing state and
* associated nodes in COMPLETING or DOWN state
*/
extern void
scontrol_print_completing (void)
{
int error_code, i;
job_info_msg_t *job_info_msg;
job_info_t *job_info;
node_info_msg_t *node_info_msg;
uint16_t show_flags = 0;
error_code = scontrol_load_job (&job_info_msg, 0);
if (error_code) {
exit_code = 1;
if (quiet_flag != 1)
slurm_perror ("slurm_load_jobs error");
return;
}
/* Must load all nodes including hidden for cross-index
* from job's node_inx to node table to work */
/*if (all_flag) Always set this flag */
show_flags |= SHOW_ALL;
if (federation_flag)
show_flags |= SHOW_FEDERATION;
if (local_flag)
show_flags |= SHOW_LOCAL;
error_code = scontrol_load_nodes(&node_info_msg, show_flags);
if (error_code) {
exit_code = 1;
if (quiet_flag != 1)
slurm_perror ("slurm_load_nodes error");
return;
}
/* Scan the jobs for completing state */
job_info = job_info_msg->job_array;
for (i = 0; i < job_info_msg->record_count; i++) {
if (job_info[i].job_state & JOB_COMPLETING)
scontrol_print_completing_job(&job_info[i],
node_info_msg);
}
slurm_free_node_info_msg(node_info_msg);
}
/*
* scontrol_hold - perform some job hold/release operation
* IN op - hold/release operation
* IN job_str - a job ID or job name
* RET 0 if no slurm error, errno otherwise. parsing error prints
* error message and returns 0
*/
extern int
scontrol_hold(char *op, char *job_str)
{
static uint32_t last_job_id = NO_VAL;
static job_info_msg_t *jobs = NULL;
job_array_resp_msg_t *resp = NULL;
int i, rc = SLURM_SUCCESS, rc2;
int j;
job_desc_msg_t job_msg;
uint32_t job_id = 0;
char *job_name = NULL;
char *job_id_str = NULL;
slurm_job_info_t *job_ptr;
if (job_str && !xstrncasecmp(job_str, "JobID=", 6))
job_str += 6;
if (job_str && !xstrncasecmp(job_str, "Job=", 4))
job_str += 4;
slurm_init_job_desc_msg (&job_msg);
if ((xstrncasecmp(op, "holdu", 5) == 0) ||
(xstrncasecmp(op, "uhold", 5) == 0)) {
job_msg.priority = 0;
job_msg.alloc_sid = ALLOC_SID_USER_HOLD;
} else if (xstrncasecmp(op, "hold", 4) == 0) {
job_msg.priority = 0;
job_msg.alloc_sid = 0;
} else
job_msg.priority = INFINITE;
if (_is_job_id(job_str)) {
while ((job_msg.job_id_str = _next_job_id())) {
rc2 = slurm_update_job2(&job_msg, &resp);
if (rc2 != SLURM_SUCCESS) {
rc2 = slurm_get_errno();
rc = MAX(rc, rc2);
exit_code = 1;
if (quiet_flag != 1) {
fprintf(stderr, "%s for job %s\n",
slurm_strerror(rc2),
job_msg.job_id_str);
}
} else if (resp) {
for (i = 0; i < resp->job_array_count; i++) {
if ((resp->error_code[i]==SLURM_SUCCESS)
&& (resp->job_array_count == 1))
continue;
exit_code = 1;
if (quiet_flag == 1)
continue;
fprintf(stderr, "%s: %s\n",
resp->job_array_id[i],
slurm_strerror(resp->
error_code[i]));
}
slurm_free_job_array_resp(resp);
resp = NULL;
}
job_msg.job_id_str = _next_job_id();
}
return rc;
} else if (job_str) {
if (!xstrncasecmp(job_str, "Name=", 5)) {
job_str += 5;
job_id = 0;
job_name = job_str;
last_job_id = NO_VAL;
} else {
exit_code = 1;
rc = ESLURM_INVALID_JOB_ID;
slurm_seterrno(rc);
if (quiet_flag != 1) {
fprintf(stderr, "%s for job %s\n",
slurm_strerror(rc), job_str);
}
return rc;
}
} else {
last_job_id = NO_VAL; /* Refresh cache on next call */
return 0;
}
if (last_job_id != job_id) {
if (scontrol_load_job(&jobs, job_id)) {
if (quiet_flag == -1)
slurm_perror ("slurm_load_job error");
return 1;
}
last_job_id = job_id;
}
/* set current user, needed e.g., for AllowGroups checks */
for (i = 0, job_ptr = jobs->job_array; i < jobs->record_count;
i++, job_ptr++) {
if (xstrcmp(job_name, job_ptr->name))
continue;
if (!IS_JOB_PENDING(job_ptr)) {
if (job_ptr->array_task_id != NO_VAL)
continue;
slurm_seterrno(ESLURM_JOB_NOT_PENDING);
rc = MAX(rc, ESLURM_JOB_NOT_PENDING);
}
if (job_ptr->array_task_str) {
xstrfmtcat(job_id_str, "%u_%s",
job_ptr->array_job_id,
job_ptr->array_task_str);
} else if (job_ptr->array_task_id != NO_VAL) {
xstrfmtcat(job_id_str, "%u_%u",
job_ptr->array_job_id,
job_ptr->array_task_id);
} else {
xstrfmtcat(job_id_str, "%u", job_ptr->job_id);
}
job_msg.job_id_str = job_id_str;
rc2 = slurm_update_job2(&job_msg, &resp);
if (rc2 != SLURM_SUCCESS) {
rc2 = slurm_get_errno();
rc = MAX(rc, rc2);
exit_code = 1;
if (quiet_flag != 1) {
fprintf(stderr, "%s for job %s\n",
slurm_strerror(rc2),
job_msg.job_id_str);
}
} else if (resp) {
for (j = 0; j < resp->job_array_count; j++) {
if ((resp->error_code[j] == SLURM_SUCCESS) &&
(resp->job_array_count == 1))
continue;
exit_code = 1;
if (quiet_flag == 1)
continue;
fprintf(stderr, "%s: %s\n",
resp->job_array_id[j],
slurm_strerror(resp->error_code[j]));
}
slurm_free_job_array_resp(resp);
resp = NULL;
}
xfree(job_id_str);
}
return rc;
}
/*
* scontrol_print_job - print the specified job's information
* IN job_id - job's id or NULL to print information about all jobs
*/
extern void scontrol_print_job(char * job_id_str)
{
int error_code = SLURM_SUCCESS, i, print_cnt = 0;
uint32_t job_id = 0;
uint32_t array_id = NO_VAL, pack_job_offset = NO_VAL;
job_info_msg_t * job_buffer_ptr = NULL;
job_info_t *job_ptr = NULL;
char *end_ptr = NULL;
if (job_id_str) {
char *tmp_job_ptr = job_id_str;
/*
* Check that the input is a valid job id (i.e. 123 or 123_456).
*/
while (*tmp_job_ptr) {
if (!isdigit(*tmp_job_ptr) &&
(*tmp_job_ptr != '_') && (*tmp_job_ptr != '+')) {
exit_code = 1;
slurm_seterrno(ESLURM_INVALID_JOB_ID);
if (quiet_flag != 1)
slurm_perror("scontrol_print_job error");
return;
}
++tmp_job_ptr;
}
job_id = (uint32_t) strtol (job_id_str, &end_ptr, 10);
if (end_ptr[0] == '_')
array_id = strtol(end_ptr + 1, &end_ptr, 10);
if (end_ptr[0] == '+')
pack_job_offset = strtol(end_ptr + 1, &end_ptr, 10);
}
error_code = scontrol_load_job(&job_buffer_ptr, job_id);
if (error_code) {
exit_code = 1;
if (quiet_flag != 1)
slurm_perror ("slurm_load_jobs error");
return;
}
if (quiet_flag == -1) {
char time_str[32];
slurm_make_time_str ((time_t *)&job_buffer_ptr->last_update,
time_str, sizeof(time_str));
printf ("last_update_time=%s, records=%d\n",
time_str, job_buffer_ptr->record_count);
}
for (i = 0, job_ptr = job_buffer_ptr->job_array;
i < job_buffer_ptr->record_count; i++, job_ptr++) {
char *save_array_str = NULL;
uint32_t save_task_id = 0;
if (!_pack_id_match(job_ptr, pack_job_offset))
continue;
if (!_task_id_in_job(job_ptr, array_id))
continue;
if ((array_id != NO_VAL) && job_ptr->array_task_str) {
save_array_str = job_ptr->array_task_str;
job_ptr->array_task_str = NULL;
save_task_id = job_ptr->array_task_id;
job_ptr->array_task_id = array_id;
}
slurm_print_job_info(stdout, job_ptr, one_liner);
if (save_array_str) {
job_ptr->array_task_str = save_array_str;
job_ptr->array_task_id = save_task_id;
}
print_cnt++;
}
if (print_cnt == 0) {
if (job_id_str) {
exit_code = 1;
if (quiet_flag != 1) {
if (array_id != NO_VAL) {
printf("Job %u_%u not found\n",
job_id, array_id);
} else if (pack_job_offset != NO_VAL) {
printf("Job %u+%u not found\n",
job_id, pack_job_offset);
} else {
printf("Job %u not found\n", job_id);
}
}
} else if (quiet_flag != 1)
printf ("No jobs in the system\n");
}
}
/*
* scontrol_hold - perform some job hold/release operation
* IN op - suspend/resume operation
* IN job_id_str - a job id
* RET 0 if no slurm error, errno otherwise. parsing error prints
* error message and returns 0
*/
extern int
scontrol_hold(char *op, char *job_id_str)
{
int i, rc = SLURM_SUCCESS;
char *next_str;
job_desc_msg_t job_msg;
uint32_t job_id;
uint32_t array_id;
job_info_msg_t *resp;
slurm_job_info_t *job_ptr;
if (job_id_str) {
job_id = (uint32_t) strtol(job_id_str, &next_str, 10);
if (next_str[0] == '_')
array_id = strtol(next_str+1, &next_str, 10);
else
array_id = NO_VAL;
if ((job_id == 0) || (next_str[0] != '\0')) {
fprintf(stderr, "Invalid job id specified\n");
return 1;
}
} else {
fprintf(stderr, "Invalid job id specified\n");
return 1;
}
if (scontrol_load_job(&resp, job_id)) {
if (quiet_flag == -1)
slurm_perror ("slurm_load_job error");
return 1;
}
slurm_init_job_desc_msg (&job_msg);
job_msg.job_id = job_id;
/* set current user, needed e.g., for AllowGroups checks */
job_msg.user_id = getuid();
if ((strncasecmp(op, "holdu", 5) == 0) ||
(strncasecmp(op, "uhold", 5) == 0)) {
job_msg.priority = 0;
job_msg.alloc_sid = ALLOC_SID_USER_HOLD;
} else if (strncasecmp(op, "hold", 4) == 0) {
job_msg.priority = 0;
job_msg.alloc_sid = 0;
} else
job_msg.priority = INFINITE;
for (i = 0, job_ptr = resp->job_array; i < resp->record_count;
i++, job_ptr++) {
if ((array_id != NO_VAL) &&
(job_ptr->array_task_id != array_id))
continue;
if (!IS_JOB_PENDING(job_ptr)) {
if ((array_id == NO_VAL) &&
(job_ptr->array_task_id != NO_VAL))
continue;
slurm_seterrno(ESLURM_JOB_NOT_PENDING);
return ESLURM_JOB_NOT_PENDING;
}
job_msg.job_id = job_ptr->job_id;
if (slurm_update_job(&job_msg))
rc = slurm_get_errno();
}
return rc;
}