int release_array_range(
job_array *pa,
struct batch_request *preq,
char *range_str)
{
tlist_head tl;
int i;
int rc;
array_request_node *rn;
array_request_node *to_free;
char *range = strchr(range_str,'=');
if (range == NULL)
return(PBSE_IVALREQ);
range++; /* move past the '=' */
CLEAR_HEAD(tl);
if (parse_array_request(range,&tl) > 0)
{
/* don't hold the jobs if range error */
return(PBSE_IVALREQ);
}
/* hold just that range from the array */
rn = (array_request_node*)GET_NEXT(tl);
while (rn != NULL)
{
for (i = rn->start; i <= rn->end; i++)
{
if (pa->jobs[i] == NULL)
continue;
/* don't stomp on other memory */
if (i >= pa->ai_qs.array_size)
continue;
if ((rc = release_job(preq,pa->jobs[i])))
return(rc);
}
/* release mem */
to_free = rn;
rn = (array_request_node*)GET_NEXT(rn->request_tokens_link);
free(to_free);
}
return(0);
} /* END release_array_range() */
/*
* hold_array_range()
*
* holds just a specified range from an array
* @param pa - the array to be acted on
* @param range_str - string specifying the range
*/
int hold_array_range(
job_array *pa, /* O */
char *range_str, /* I */
attribute *temphold) /* I */
{
tlist_head tl;
int i;
array_request_node *rn;
array_request_node *to_free;
char *range = strchr(range_str,'=');
if (range == NULL)
return(PBSE_IVALREQ);
range++; /* move past the '=' */
CLEAR_HEAD(tl);
if (parse_array_request(range,&tl) > 0)
{
/* don't hold the jobs if range error */
return(PBSE_IVALREQ);
}
else
{
/* hold just that range from the array */
rn = (array_request_node*)GET_NEXT(tl);
while (rn != NULL)
{
for (i = rn->start; i <= rn->end; i++)
{
if (pa->jobs[i] == NULL)
continue;
/* don't stomp on other memory */
if (i >= pa->ai_qs.array_size)
continue;
hold_job(temphold,pa->jobs[i]);
}
/* release mem */
to_free = rn;
rn = (array_request_node*)GET_NEXT(rn->request_tokens_link);
free(to_free);
}
}
return(0);
} /* END hold_array_range() */
int setup_array_struct(job *pjob)
{
job_array *pa;
/* struct work_task *wt; */
array_request_node *rn;
int bad_token_count;
int array_size;
int rc;
/* setup a link to this job array in the servers all_arrays list */
pa = (job_array *)calloc(1,sizeof(job_array));
pa->ai_qs.struct_version = ARRAY_QS_STRUCT_VERSION;
pa->template_job = pjob;
/*pa->ai_qs.array_size = pjob->ji_wattr[(int)JOB_ATR_job_array_size].at_val.at_long;*/
strcpy(pa->ai_qs.parent_id, pjob->ji_qs.ji_jobid);
strcpy(pa->ai_qs.fileprefix, pjob->ji_qs.ji_fileprefix);
strncpy(pa->ai_qs.owner, pjob->ji_wattr[JOB_ATR_job_owner].at_val.at_str, PBS_MAXUSER + PBS_MAXSERVERNAME + 2);
strncpy(pa->ai_qs.submit_host, get_variable(pjob, pbs_o_host), PBS_MAXSERVERNAME);
pa->ai_qs.num_cloned = 0;
CLEAR_LINK(pa->all_arrays);
CLEAR_HEAD(pa->request_tokens);
append_link(&svr_jobarrays, &pa->all_arrays, (void*)pa);
if (job_save(pjob, SAVEJOB_FULL, 0) != 0)
{
job_purge(pjob);
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pjob != NULL) ? pjob->ji_qs.ji_jobid : "NULL",
"cannot save job");
}
return 1;
}
if ((rc = set_slot_limit(pjob->ji_wattr[JOB_ATR_job_array_request].at_val.at_str, pa)))
{
array_delete(pa);
snprintf(log_buffer,sizeof(log_buffer),
"Array %s requested a slot limit above the max limit %ld, rejecting\n",
pa->ai_qs.parent_id,
server.sv_attr[SRV_ATR_MaxSlotLimit].at_val.at_long);
log_event(PBSEVENT_SYSTEM,
PBS_EVENTCLASS_JOB,
pa->ai_qs.parent_id,
log_buffer);
return(INVALID_SLOT_LIMIT);
}
pa->ai_qs.jobs_running = 0;
pa->ai_qs.num_started = 0;
pa->ai_qs.num_failed = 0;
pa->ai_qs.num_successful = 0;
bad_token_count =
parse_array_request(pjob->ji_wattr[JOB_ATR_job_array_request].at_val.at_str,
&(pa->request_tokens));
/* get the number of elements that should be allocated in the array */
rn = (array_request_node *)GET_NEXT(pa->request_tokens);
array_size = 0;
pa->ai_qs.num_jobs = 0;
while (rn != NULL)
{
if (rn->end > array_size)
array_size = rn->end;
/* calculate the actual number of jobs (different from array size) */
pa->ai_qs.num_jobs += rn->end - rn->start + 1;
rn = (array_request_node *)GET_NEXT(rn->request_tokens_link);
}
/* size of array is the biggest index + 1 */
array_size++;
if (server.sv_attr[SRV_ATR_MaxArraySize].at_flags & ATR_VFLAG_SET)
{
int max_array_size = server.sv_attr[SRV_ATR_MaxArraySize].at_val.at_long;
if (max_array_size < pa->ai_qs.num_jobs)
{
array_delete(pa);
return(ARRAY_TOO_LARGE);
}
}
/* initialize the array */
pa->jobs = malloc(array_size * sizeof(job *));
memset(pa->jobs,0,array_size * sizeof(job *));
/* remember array_size */
pa->ai_qs.array_size = array_size;
CLEAR_HEAD(pa->ai_qs.deps);
array_save(pa);
if (bad_token_count > 0)
{
array_delete(pa);
return 2;
}
return 0;
}
int modify_array_range(
job_array *pa, /* I/O */
char *range, /* I */
svrattrl *plist, /* I */
struct batch_request *preq, /* I */
int checkpoint_req) /* I */
{
char id[] = "modify_array_range";
tlist_head tl;
int i, rc;
int mom_relay = 0;
array_request_node *rn;
array_request_node *to_free;
CLEAR_HEAD(tl);
if (parse_array_request(range,&tl) > 0)
{
/* don't hold the jobs if range error */
return(FAILURE);
}
else
{
/* hold just that range from the array */
rn = (array_request_node*)GET_NEXT(tl);
while (rn != NULL)
{
for (i = rn->start; i <= rn->end; i++)
{
if ((i >= pa->ai_qs.array_size) ||
(pa->jobs[i] == NULL))
continue;
rc = modify_job(pa->jobs[i],plist,preq,checkpoint_req, NO_MOM_RELAY);
if (rc == PBSE_RELAYED_TO_MOM)
{
struct batch_request *array_req = NULL;
/* We told modify_job not to call relay_to_mom so we need to contact the mom */
rc = copy_batchrequest(&array_req, preq, 0, i);
if (rc != 0)
{
return(rc);
}
preq->rq_refcount++;
if (mom_relay == 0)
{
preq->rq_refcount++;
}
mom_relay++;
if ((rc = relay_to_mom(
pa->jobs[i],
array_req,
post_modify_arrayreq)))
{
snprintf(log_buffer,sizeof(log_buffer),
"Unable to relay information to mom for job '%s'\n",
pa->jobs[i]->ji_qs.ji_jobid);
log_err(rc,id,log_buffer);
return(rc); /* unable to get to MOM */
}
}
}
/* release mem */
to_free = rn;
rn = (array_request_node*)GET_NEXT(rn->request_tokens_link);
free(to_free);
}
}
if (mom_relay)
{
preq->rq_refcount--;
if (preq->rq_refcount == 0)
{
free_br(preq);
}
return(PBSE_RELAYED_TO_MOM);
}
return(PBSE_NONE);
} /* END modify_array_range() */
/*
* delete_array_range()
*
* deletes a range from a specific array
*
* @param pa - the array whose jobs are deleted
* @param range_str - the user-given range to delete
* @return - the number of jobs skipped, -1 if range error
*/
int delete_array_range(
job_array *pa,
char *range_str)
{
tlist_head tl;
array_request_node *rn;
array_request_node *to_free;
job *pjob;
char *range;
int i;
int num_skipped = 0;
/* get just the numeric range specified, '=' should
* always be there since we put it there in qdel */
range = strchr(range_str,'=');
range++; /* move past the '=' */
CLEAR_HEAD(tl);
if (parse_array_request(range,&tl) > 0)
{
/* don't delete jobs if range error */
return(-1);
}
rn = (array_request_node*)GET_NEXT(tl);
while (rn != NULL)
{
for (i = rn->start; i <= rn->end; i++)
{
if (pa->jobs[i] == NULL)
continue;
/* don't stomp on other memory */
if (i >= pa->ai_qs.array_size)
continue;
pjob = pa->jobs[i];
if (pjob->ji_qs.ji_state >= JOB_STATE_EXITING)
{
/* invalid state for request, skip */
continue;
}
if (attempt_delete((void *)pjob) == FALSE)
num_skipped++;
}
to_free = rn;
rn = (array_request_node*)GET_NEXT(rn->request_tokens_link);
/* release mem */
free(to_free);
}
return(num_skipped);
}
int setup_array_struct(
job *pjob)
{
job_array *pa;
array_request_node *rn;
int bad_token_count;
int array_size;
int rc;
char log_buf[LOCAL_LOG_BUF_SIZE];
long max_array_size;
pa = (job_array *)calloc(1,sizeof(job_array));
pa->ai_qs.struct_version = ARRAY_QS_STRUCT_VERSION;
strcpy(pa->ai_qs.parent_id, pjob->ji_qs.ji_jobid);
strcpy(pa->ai_qs.fileprefix, pjob->ji_qs.ji_fileprefix);
snprintf(pa->ai_qs.owner, sizeof(pa->ai_qs.owner), "%s", pjob->ji_wattr[JOB_ATR_job_owner].at_val.at_str);
snprintf(pa->ai_qs.submit_host, sizeof(pa->ai_qs.submit_host), "%s", get_variable(pjob, pbs_o_host));
pa->ai_qs.num_cloned = 0;
CLEAR_HEAD(pa->request_tokens);
pa->ai_mutex = calloc(1, sizeof(pthread_mutex_t));
pthread_mutex_init(pa->ai_mutex,NULL);
lock_ai_mutex(pa, __func__, NULL, LOGLEVEL);
if (job_save(pjob, SAVEJOB_FULL, 0) != 0)
{
/* the array is deleted in svr_job_purge */
unlock_ai_mutex(pa, __func__, "1", LOGLEVEL);
svr_job_purge(pjob);
/* Does job array need to be removed? */
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pjob != NULL) ? pjob->ji_qs.ji_jobid : "NULL",
"cannot save job");
}
return(1);
}
if ((rc = set_slot_limit(pjob->ji_wattr[JOB_ATR_job_array_request].at_val.at_str, pa)))
{
long max_limit = 0;
get_svr_attr_l(SRV_ATR_MaxSlotLimit, &max_limit);
array_delete(pa);
snprintf(log_buf,sizeof(log_buf),
"Array %s requested a slot limit above the max limit %ld, rejecting\n",
pa->ai_qs.parent_id,
max_limit);
log_event(PBSEVENT_SYSTEM,PBS_EVENTCLASS_JOB,pa->ai_qs.parent_id,log_buf);
return(INVALID_SLOT_LIMIT);
}
pa->ai_qs.jobs_running = 0;
pa->ai_qs.num_started = 0;
pa->ai_qs.num_failed = 0;
pa->ai_qs.num_successful = 0;
bad_token_count = parse_array_request(
pjob->ji_wattr[JOB_ATR_job_array_request].at_val.at_str,
&(pa->request_tokens));
/* get the number of elements that should be allocated in the array */
rn = (array_request_node *)GET_NEXT(pa->request_tokens);
array_size = 0;
pa->ai_qs.num_jobs = 0;
while (rn != NULL)
{
if (rn->end > array_size)
array_size = rn->end;
/* calculate the actual number of jobs (different from array size) */
pa->ai_qs.num_jobs += rn->end - rn->start + 1;
rn = (array_request_node *)GET_NEXT(rn->request_tokens_link);
}
/* size of array is the biggest index + 1 */
array_size++;
if (get_svr_attr_l(SRV_ATR_MaxArraySize, &max_array_size) == PBSE_NONE)
{
if (max_array_size < pa->ai_qs.num_jobs)
{
array_delete(pa);
return(ARRAY_TOO_LARGE);
}
}
/* initialize the array */
pa->job_ids = calloc(array_size, sizeof(char *));
if (pa->job_ids == NULL)
{
sprintf(log_buf, "Failed to alloc job_ids: job %s", pjob->ji_qs.ji_jobid);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, log_buf);
return(PBSE_MEM_MALLOC);
}
/* remember array_size */
pa->ai_qs.array_size = array_size;
CLEAR_HEAD(pa->ai_qs.deps);
array_save(pa);
if (bad_token_count > 0)
{
array_delete(pa);
return 2;
}
pjob->ji_arraystruct = pa;
insert_array(pa);
unlock_ai_mutex(pa, __func__, "1", LOGLEVEL);
return(PBSE_NONE);
} /* END setup_array_struct() */
int modify_array_range(
job_array *pa, /* I/O */
char *range, /* I */
svrattrl *plist, /* I */
struct batch_request *preq, /* I */
int checkpoint_req) /* I */
{
char log_buf[LOCAL_LOG_BUF_SIZE];
tlist_head tl;
int i;
int rc;
int mom_relay = 0;
job *pjob;
array_request_node *rn;
array_request_node *to_free;
CLEAR_HEAD(tl);
if (parse_array_request(range,&tl) > 0)
{
/* don't hold the jobs if range error */
return(FAILURE);
}
else
{
/* hold just that range from the array */
rn = (array_request_node*)GET_NEXT(tl);
while (rn != NULL)
{
for (i = rn->start; i <= rn->end; i++)
{
if ((i >= pa->ai_qs.array_size) ||
(pa->job_ids[i] == NULL))
continue;
if ((pjob = svr_find_job(pa->job_ids[i], FALSE)) == NULL)
{
free(pa->job_ids[i]);
pa->job_ids[i] = NULL;
}
else
{
pthread_mutex_unlock(pa->ai_mutex);
rc = modify_job((void **)&pjob, plist, preq, checkpoint_req, NO_MOM_RELAY);
pa = get_jobs_array(&pjob);
if (pjob != NULL)
{
if (rc == PBSE_RELAYED_TO_MOM)
{
struct batch_request *array_req = NULL;
/* We told modify_job not to call relay_to_mom so we need to contact the mom */
if ((rc = copy_batchrequest(&array_req, preq, 0, i)) != PBSE_NONE)
{
return(rc);
}
preq->rq_refcount++;
if (mom_relay == 0)
{
preq->rq_refcount++;
}
mom_relay++;
/* The array_req is freed in relay_to_mom (failure)
* or in issue_Drequest (success) */
if ((rc = relay_to_mom(&pjob, array_req, NULL)))
{
snprintf(log_buf,sizeof(log_buf),
"Unable to relay information to mom for job '%s'\n",
pjob->ji_qs.ji_jobid);
log_err(rc, __func__, log_buf);
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
return(rc); /* unable to get to MOM */
}
else
{
unlock_ji_mutex(pjob, __func__, "2", LOGLEVEL);
post_modify_arrayreq(array_req);
}
}
else
unlock_ji_mutex(pjob, __func__, "2", LOGLEVEL);
}
else
pa->job_ids[i] = NULL;
}
}
/* release mem */
to_free = rn;
rn = (array_request_node*)GET_NEXT(rn->request_tokens_link);
free(to_free);
}
}
if (mom_relay)
{
preq->rq_refcount--;
if (preq->rq_refcount == 0)
{
free_br(preq);
}
return(PBSE_RELAYED_TO_MOM);
}
return(PBSE_NONE);
} /* END modify_array_range() */
int release_array_range(
job_array *pa,
struct batch_request *preq,
char *range_str)
{
tlist_head tl;
int i;
int rc;
job *pjob;
array_request_node *rn;
array_request_node *to_free;
char *range = strchr(range_str,'=');
if (range == NULL)
return(PBSE_IVALREQ);
range++; /* move past the '=' */
CLEAR_HEAD(tl);
if (parse_array_request(range,&tl) > 0)
{
/* don't hold the jobs if range error */
return(PBSE_IVALREQ);
}
/* hold just that range from the array */
rn = (array_request_node*)GET_NEXT(tl);
while (rn != NULL)
{
for (i = rn->start; i <= rn->end; i++)
{
/* don't stomp on other memory */
if (i >= pa->ai_qs.array_size)
continue;
if (pa->job_ids[i] == NULL)
continue;
if ((pjob = svr_find_job(pa->job_ids[i], FALSE)) == NULL)
{
free(pa->job_ids[i]);
pa->job_ids[i] = NULL;
}
else
{
if ((rc = release_job(preq,pjob)))
{
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
return(rc);
}
unlock_ji_mutex(pjob, __func__, "2", LOGLEVEL);
}
}
/* release mem */
to_free = rn;
rn = (array_request_node*)GET_NEXT(rn->request_tokens_link);
free(to_free);
}
return(PBSE_NONE);
} /* END release_array_range() */
/*
* delete_array_range()
*
* deletes a range from a specific array
*
* @param pa - the array whose jobs are deleted
* @param range_str - the user-given range to delete
* @return - the number of jobs skipped, -1 if range error
*/
int delete_array_range(
job_array *pa,
char *range_str)
{
tlist_head tl;
array_request_node *rn;
array_request_node *to_free;
job *pjob;
char *range;
int i;
int num_skipped = 0;
int deleted;
/* get just the numeric range specified, '=' should
* always be there since we put it there in qdel */
range = strchr(range_str,'=');
range++; /* move past the '=' */
CLEAR_HEAD(tl);
if (parse_array_request(range,&tl) > 0)
{
/* don't delete jobs if range error */
return(-1);
}
rn = (array_request_node*)GET_NEXT(tl);
while (rn != NULL)
{
for (i = rn->start; i <= rn->end; i++)
{
if (pa->job_ids[i] == NULL)
continue;
/* don't stomp on other memory */
if (i >= pa->ai_qs.array_size)
continue;
if ((pjob = svr_find_job(pa->job_ids[i], FALSE)) == NULL)
{
free(pa->job_ids[i]);
pa->job_ids[i] = NULL;
}
else
{
if (pjob->ji_qs.ji_state >= JOB_STATE_EXITING)
{
/* invalid state for request, skip */
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
continue;
}
pthread_mutex_unlock(pa->ai_mutex);
deleted = attempt_delete(pjob);
if (deleted == FALSE)
{
/* if the job was deleted, this mutex would be taked care of elsewhere. When it fails,
* release it here */
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
num_skipped++;
}
pthread_mutex_lock(pa->ai_mutex);
}
}
to_free = rn;
rn = (array_request_node*)GET_NEXT(rn->request_tokens_link);
/* release mem */
free(to_free);
}
return(num_skipped);
}
/*
* delete_array_range()
*
* deletes a range from a specific array
*
* @param pa - the array whose jobs are deleted
* @param range_str - the user-given range to delete
* @return - the number of jobs skipped, -1 if range error
*/
int delete_array_range(
job_array *pa,
char *range_str)
{
tlist_head tl;
array_request_node *rn;
array_request_node *to_free;
job *pjob;
char *range;
int i;
int num_skipped = 0;
int num_deleted = 0;
int deleted;
int running;
/* get just the numeric range specified, '=' should
* always be there since we put it there in qdel */
if((range = strchr(range_str,'=')) == NULL)
return(-1);
range++; /* move past the '=' */
CLEAR_HEAD(tl);
if (parse_array_request(range,&tl) > 0)
{
/* don't delete jobs if range error */
return(-1);
}
rn = (array_request_node*)GET_NEXT(tl);
while (rn != NULL)
{
for (i = rn->start; i <= rn->end; i++)
{
if (pa->job_ids[i] == NULL)
continue;
/* don't stomp on other memory */
if (i >= pa->ai_qs.array_size)
continue;
if ((pjob = svr_find_job(pa->job_ids[i], FALSE)) == NULL)
{
free(pa->job_ids[i]);
pa->job_ids[i] = NULL;
}
else
{
mutex_mgr pjob_mutex = mutex_mgr(pjob->ji_mutex, true);
if (pjob->ji_qs.ji_state >= JOB_STATE_EXITING)
{
/* invalid state for request, skip */
continue;
}
running = (pjob->ji_qs.ji_state == JOB_STATE_RUNNING);
pthread_mutex_unlock(pa->ai_mutex);
deleted = attempt_delete(pjob);
/* we come out of attempt_delete unlocked */
pjob_mutex.set_lock_on_exit(false);
if (deleted == FALSE)
{
/* if the job was deleted, this mutex would be taked care of elsewhere. When it fails,
* release it here */
num_skipped++;
}
else if (running == FALSE)
{
/* running jobs will increase the deleted count when their obit is reported */
num_deleted++;
}
pthread_mutex_lock(pa->ai_mutex);
}
}
to_free = rn;
rn = (array_request_node*)GET_NEXT(rn->request_tokens_link);
/* release mem */
free(to_free);
}
pa->ai_qs.num_failed += num_deleted;
return(num_skipped);
} /* END delete_array_range() */
