END_TEST
START_TEST(job_set_wait_test)
{
struct job test_job;
struct pbs_attribute test_attribute;
int result = -1;
memset(&test_job, 0, sizeof(test_job));
memset(&test_attribute, 0, sizeof(test_attribute));
result = job_set_wait(NULL, &test_job, 0);
fail_unless(result == PBSE_BAD_PARAMETER, "NULL input attribute pointer fail");
result = job_set_wait(&test_attribute, NULL, 0);
fail_unless(result == PBSE_BAD_PARAMETER, "NULL input job pointer fail");
result = job_set_wait(&test_attribute, &test_job, 0);
fail_unless(result == PBSE_NONE, "job_set_wait fail");
}
static void post_stagein(
struct work_task *pwt)
{
int code;
int newstate;
int newsub;
job *pjob;
struct batch_request *preq;
attribute *pwait;
preq = pwt->wt_parm1;
code = preq->rq_reply.brp_code;
pjob = find_job(preq->rq_extra);
free(preq->rq_extra);
if (pjob != NULL)
{
if (code != 0)
{
/* stage in failed - hold job */
free_nodes(pjob);
pwait = &pjob->ji_wattr[(int)JOB_ATR_exectime];
if ((pwait->at_flags & ATR_VFLAG_SET) == 0)
{
pwait->at_val.at_long = time_now + PBS_STAGEFAIL_WAIT;
pwait->at_flags |= ATR_VFLAG_SET;
job_set_wait(pwait, pjob, 0);
}
svr_setjobstate(pjob, JOB_STATE_WAITING, JOB_SUBSTATE_STAGEFAIL);
if (preq->rq_reply.brp_choice == BATCH_REPLY_CHOICE_Text)
{
/* set job comment */
/* NYI */
svr_mailowner(
pjob,
MAIL_STAGEIN,
MAIL_FORCE,
preq->rq_reply.brp_un.brp_txt.brp_str);
}
}
else
{
/* stage in was successful */
pjob->ji_qs.ji_svrflags |= JOB_SVFLG_StagedIn;
if (pjob->ji_qs.ji_substate == JOB_SUBSTATE_STAGEGO)
{
if (is_checkpoint_restart(pjob))
{
/* need to copy checkpoint file to mom before running */
svr_send_checkpoint(
pjob,
preq,
JOB_STATE_RUNNING,
JOB_SUBSTATE_CHKPTGO);
}
else
{
/* continue to start job running */
svr_strtjob2(pjob, NULL);
}
}
else
{
svr_evaljobstate(pjob, &newstate, &newsub, 0);
svr_setjobstate(pjob, newstate, newsub);
}
}
} /* END if (pjob != NULL) */
release_req(pwt); /* close connection and release request */
return;
} /* END post_stagein() */
static void post_checkpointsend(
struct work_task *pwt)
{
int code;
job *pjob;
struct batch_request *preq;
attribute *pwait;
preq = pwt->wt_parm1;
code = preq->rq_reply.brp_code;
pjob = find_job(preq->rq_extra);
free(preq->rq_extra);
if (pjob != NULL)
{
if (code != 0)
{
/* copy failed - hold job */
free_nodes(pjob);
pwait = &pjob->ji_wattr[(int)JOB_ATR_exectime];
if ((pwait->at_flags & ATR_VFLAG_SET) == 0)
{
pwait->at_val.at_long = time_now + PBS_STAGEFAIL_WAIT;
pwait->at_flags |= ATR_VFLAG_SET;
job_set_wait(pwait, pjob, 0);
}
svr_setjobstate(pjob, JOB_STATE_WAITING, JOB_SUBSTATE_STAGEFAIL);
if (preq->rq_reply.brp_choice == BATCH_REPLY_CHOICE_Text)
{
sprintf(log_buffer, "Failed to copy checkpoint file to mom - %s",
preq->rq_reply.brp_un.brp_txt.brp_str);
log_event(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
pjob->ji_qs.ji_jobid,
log_buffer);
/* NYI */
svr_mailowner(
pjob,
MAIL_CHKPTCOPY,
MAIL_FORCE,
preq->rq_reply.brp_un.brp_txt.brp_str);
}
}
else
{
/* checkpoint copy was successful */
pjob->ji_qs.ji_svrflags |= JOB_SVFLG_CHECKPOINT_COPIED;
/* set restart_name attribute to the checkpoint_name we just copied */
job_attr_def[(int)JOB_ATR_restart_name].at_set(
&pjob->ji_wattr[(int)JOB_ATR_restart_name],
&pjob->ji_wattr[(int)JOB_ATR_checkpoint_name],
SET);
pjob->ji_modified = 1;
job_save(pjob, SAVEJOB_FULL);
/* continue to start job running */
svr_strtjob2(pjob, NULL);
}
} /* END if (pjob != NULL) */
release_req(pwt); /* close connection and release request */
return;
} /* END post_checkpointsend() */
