int req_rerunjob(
batch_request *preq)
{
int rc = PBSE_NONE;
job *pjob;
int MgrRequired = TRUE;
char log_buf[LOCAL_LOG_BUF_SIZE];
/* check if requestor is admin, job owner, etc */
if (!strcasecmp(preq->rq_ind.rq_rerun, "all"))
{
return(handle_requeue_all(preq));
}
if ((pjob = chk_job_request(preq->rq_ind.rq_rerun, preq)) == 0)
{
/* FAILURE */
/* chk_job_request calls req_reject() */
rc = PBSE_SYSTEM;
return rc; /* This needs to fixed to return an accurate error */
}
mutex_mgr pjob_mutex = mutex_mgr(pjob->ji_mutex, true);
/* the job must be running or completed */
if (pjob->ji_qs.ji_state >= JOB_STATE_EXITING)
{
if (pjob->ji_wattr[JOB_ATR_checkpoint_name].at_flags & ATR_VFLAG_SET)
{
/* allow end-users to rerun checkpointed jobs */
MgrRequired = FALSE;
}
}
else if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
/* job is running */
/* NO-OP */
}
else if (pjob->ji_qs.ji_state == JOB_STATE_QUEUED)
{
// If we are already queued, then there is nothing to do.
rc = PBSE_NONE;
reply_ack(preq);
return(rc);
}
else
{
/* FAILURE - job is in bad state */
rc = PBSE_BADSTATE;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "job %s is in a bad state",
preq->rq_ind.rq_rerun);
req_reject(rc, 0, preq, NULL, log_buf);
return rc;
}
if ((MgrRequired == TRUE) &&
((preq->rq_perm & (ATR_DFLAG_MGWR | ATR_DFLAG_OPWR)) == 0))
{
/* FAILURE */
rc = PBSE_PERM;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"additional permissions required (ATR_DFLAG_MGWR | ATR_DFLAG_OPWR)");
req_reject(rc, 0, preq, NULL, log_buf);
return rc;
}
/* the job must be rerunnable */
if (pjob->ji_wattr[JOB_ATR_rerunable].at_val.at_long == 0)
{
/* NOTE: should force override this constraint? maybe (???) */
/* no, the user is saying that the job will break, and
IEEE Std 1003.1 specifically says rerun is to be rejected
if rerunable==FALSE -garrick */
rc = PBSE_NORERUN;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "job %s not rerunnable",
preq->rq_ind.rq_rerun);
req_reject(rc, 0, preq, NULL, log_buf);
return rc;
}
if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
/* ask MOM to kill off the job if it is running */
int delay = 0;
pbs_queue *pque;
// Apply the user delay first so it takes precedence.
if (pjob->ji_wattr[JOB_ATR_user_kill_delay].at_flags & ATR_VFLAG_SET)
delay = pjob->ji_wattr[JOB_ATR_user_kill_delay].at_val.at_long;
if ((pque = get_jobs_queue(&pjob)) != NULL)
{
mutex_mgr pque_mutex = mutex_mgr(pque->qu_mutex, true);
mutex_mgr server_mutex = mutex_mgr(server.sv_attr_mutex, false);
if (delay == 0)
{
delay = attr_ifelse_long(&pque->qu_attr[QE_ATR_KillDelay],
&server.sv_attr[SRV_ATR_KillDelay],
0);
}
}
else
{
/* why is the pque null. Something went wrong */
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "jobid %s returned a null queue", pjob->ji_qs.ji_jobid);
req_reject(PBSE_UNKQUE, 0, preq, NULL, log_buf);
return(PBSE_UNKQUE);
}
pjob->ji_qs.ji_substate = JOB_SUBSTATE_RERUN;
if (delay != 0)
{
static const char *rerun = "rerun";
char *extra = strdup(rerun);
get_batch_request_id(preq);
/* If a qrerun -f is given requeue the job regardless of the outcome of issue_signal*/
if ((preq->rq_extend) &&
(!strncasecmp(preq->rq_extend, RERUNFORCE, strlen(RERUNFORCE))))
{
std::string extend = RERUNFORCE;
batch_request *dup = new batch_request(*preq);
get_batch_request_id(dup);
rc = issue_signal(&pjob, "SIGTERM", delay_and_send_sig_kill, extra, strdup(dup->rq_id.c_str()));
if (rc == PBSE_NORELYMOM)
{
dup->rq_reply.brp_code = PBSE_NORELYMOM;
pjob_mutex.unlock();
post_rerun(dup);
pjob = svr_find_job(preq->rq_ind.rq_signal.rq_jid, FALSE);
if (pjob == NULL)
{
delete dup;
return(PBSE_NONE);
}
pjob_mutex.set_lock_state(true);
rc = PBSE_NONE;
}
delete dup;
}
else
{
rc = issue_signal(&pjob, "SIGTERM", delay_and_send_sig_kill, extra, strdup(preq->rq_id.c_str()));
if (rc != PBSE_NONE)
{
/* cant send to MOM */
req_reject(rc, 0, preq, NULL, NULL);
}
return(rc);
}
}
else
{
static const char *rerun = "rerun";
char *extra = strdup(rerun);
/* If a qrerun -f is given requeue the job regardless of the outcome of issue_signal*/
if (preq->rq_extend && !strncasecmp(preq->rq_extend, RERUNFORCE, strlen(RERUNFORCE)))
{
std::string extend = RERUNFORCE;
rc = issue_signal(&pjob, "SIGKILL", post_rerun, extra, strdup(extend.c_str()));
if (rc == PBSE_NORELYMOM)
rc = PBSE_NONE;
}
else
rc = issue_signal(&pjob, "SIGKILL", post_rerun, extra, NULL);
}
}
else
{
if (pjob->ji_wattr[JOB_ATR_hold].at_val.at_long == HOLD_n)
{
svr_setjobstate(pjob, JOB_STATE_QUEUED, JOB_SUBSTATE_QUEUED, FALSE);
}
else
{
svr_setjobstate(pjob, JOB_STATE_HELD, JOB_SUBSTATE_HELD, FALSE);
}
/* reset some job attributes */
pjob->ji_wattr[JOB_ATR_comp_time].at_flags &= ~ATR_VFLAG_SET;
pjob->ji_wattr[JOB_ATR_reported].at_flags &= ~ATR_VFLAG_SET;
set_statechar(pjob);
rc = -1;
}
/* finalize_rerunjob will return with pjob->ji_mutex unlocked */
pjob_mutex.set_unlock_on_exit(false);
return finalize_rerunjob(preq,pjob,rc);
}
int req_rerunjob(
struct batch_request *preq)
{
int rc = PBSE_NONE;
job *pjob;
int MgrRequired = TRUE;
char log_buf[LOCAL_LOG_BUF_SIZE];
/* check if requestor is admin, job owner, etc */
if ((pjob = chk_job_request(preq->rq_ind.rq_rerun, preq)) == 0)
{
/* FAILURE */
/* chk_job_request calls req_reject() */
rc = PBSE_SYSTEM;
return rc; /* This needs to fixed to return an accurate error */
}
/* the job must be running or completed */
if (pjob->ji_qs.ji_state >= JOB_STATE_EXITING)
{
if (pjob->ji_wattr[JOB_ATR_checkpoint_name].at_flags & ATR_VFLAG_SET)
{
/* allow end-users to rerun checkpointed jobs */
MgrRequired = FALSE;
}
}
else if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
/* job is running */
/* NO-OP */
}
else
{
/* FAILURE - job is in bad state */
rc = PBSE_BADSTATE;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "job %s is in a bad state",
preq->rq_ind.rq_rerun);
req_reject(rc, 0, preq, NULL, log_buf);
unlock_ji_mutex(pjob, __func__, "2", LOGLEVEL);
return rc;
}
if ((MgrRequired == TRUE) &&
((preq->rq_perm & (ATR_DFLAG_MGWR | ATR_DFLAG_OPWR)) == 0))
{
/* FAILURE */
rc = PBSE_PERM;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"additional permissions required (ATR_DFLAG_MGWR | ATR_DFLAG_OPWR)");
req_reject(rc, 0, preq, NULL, log_buf);
unlock_ji_mutex(pjob, __func__, "3", LOGLEVEL);
return rc;
}
/* the job must be rerunnable */
if (pjob->ji_wattr[JOB_ATR_rerunable].at_val.at_long == 0)
{
/* NOTE: should force override this constraint? maybe (???) */
/* no, the user is saying that the job will break, and
IEEE Std 1003.1 specifically says rerun is to be rejected
if rerunable==FALSE -garrick */
rc = PBSE_NORERUN;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "job %s not rerunnable",
preq->rq_ind.rq_rerun);
req_reject(rc, 0, preq, NULL, log_buf);
unlock_ji_mutex(pjob, __func__, "4", LOGLEVEL);
return rc;
}
if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
/* ask MOM to kill off the job if it is running */
int delay = 0;
pbs_queue *pque;
if ((pque = get_jobs_queue(&pjob)) != NULL)
{
mutex_mgr pque_mutex = mutex_mgr(pque->qu_mutex, true);
pthread_mutex_lock(server.sv_attr_mutex);
delay = attr_ifelse_long(&pque->qu_attr[QE_ATR_KillDelay],
&server.sv_attr[SRV_ATR_KillDelay],
0);
pthread_mutex_unlock(server.sv_attr_mutex);
}
else if (pjob == NULL)
{
rc = PBSE_NORERUN;
unlock_ji_mutex(pjob, __func__, "4", LOGLEVEL);
return rc;
}
if(delay != 0)
{
static const char *rerun = "rerun";
char *extra = strdup(rerun);
get_batch_request_id(preq);
if ((rc = issue_signal(&pjob, "SIGTERM", delay_and_send_sig_kill, extra, strdup(preq->rq_id))))
{
/* cant send to MOM */
req_reject(rc, 0, preq, NULL, NULL);
}
unlock_ji_mutex(pjob, __func__, "4", LOGLEVEL);
return rc;
}
else
{
static const char *rerun = "rerun";
char *extra = strdup(rerun);
rc = issue_signal(&pjob, "SIGKILL", post_rerun, extra, NULL);
}
}
else
{
if (pjob->ji_wattr[JOB_ATR_hold].at_val.at_long == HOLD_n)
{
svr_setjobstate(pjob, JOB_STATE_QUEUED, JOB_SUBSTATE_QUEUED, FALSE);
}
else
{
svr_setjobstate(pjob, JOB_STATE_HELD, JOB_SUBSTATE_HELD, FALSE);
}
/* reset some job attributes */
pjob->ji_wattr[JOB_ATR_comp_time].at_flags &= ~ATR_VFLAG_SET;
pjob->ji_wattr[JOB_ATR_reported].at_flags &= ~ATR_VFLAG_SET;
set_statechar(pjob);
rc = -1;
}
return finalize_rerunjob(preq,pjob,rc);
}
