END_TEST
START_TEST(remove_checkpoint_test)
{
struct job* test_job = NULL;
remove_checkpoint(NULL); /*TODO: add a kind of assert*/
remove_checkpoint(&test_job);
test_job = job_alloc();
remove_checkpoint(&test_job);
}
void mon_breakpoint_delete_checkpoint(int cp_num)
{
int i;
checkpoint_t *cp = NULL;
if (cp_num == -1) {
/* Add user confirmation here. */
mon_out("Deleting all checkpoints\n");
for (i = 1; i < breakpoint_count; i++) {
if ((cp = find_checkpoint(i))) {
remove_checkpoint(cp);
}
}
} else if (!(cp = find_checkpoint(cp_num))) {
mon_out("#%d not a valid checkpoint\n", cp_num);
return;
} else {
remove_checkpoint(cp);
}
}
static void post_movejob(
struct work_task *pwt)
{
char *id = "post_movejob";
struct batch_request *req;
int newstate;
int newsub;
int stat;
int r;
job *jobp;
req = (struct batch_request *)pwt->wt_parm2;
stat = pwt->wt_aux;
pbs_errno = PBSE_NONE;
if (req->rq_type != PBS_BATCH_MoveJob)
{
sprintf(log_buffer, "bad request type %d\n",
req->rq_type);
log_err(-1, id, log_buffer);
return;
}
jobp = find_job(req->rq_ind.rq_move.rq_jid);
if ((jobp == NULL) || (jobp != (job *)pwt->wt_parm1))
{
sprintf(log_buffer, "job %s not found\n",
req->rq_ind.rq_move.rq_jid);
log_err(-1, id, log_buffer);
}
if (WIFEXITED(stat))
{
r = WEXITSTATUS(stat);
if (r == 0)
{
/* purge server's job structure */
if (jobp->ji_qs.ji_svrflags & JOB_SVFLG_StagedIn)
remove_stagein(jobp);
if (jobp->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_COPIED)
remove_checkpoint(jobp);
strcpy(log_buffer, msg_movejob);
sprintf(log_buffer + strlen(log_buffer), msg_manager,
req->rq_ind.rq_move.rq_destin,
req->rq_user,
req->rq_host);
job_purge(jobp);
}
else
{
r = PBSE_ROUTEREJ;
}
}
else
{
r = PBSE_SYSTEM;
sprintf(log_buffer, msg_badexit, stat);
strcat(log_buffer, id);
log_event(
PBSEVENT_SYSTEM,
PBS_EVENTCLASS_JOB,
jobp->ji_qs.ji_jobid,
log_buffer);
}
if (r)
{
if (jobp != NULL)
{
/* force re-eval of job state out of Transit */
svr_evaljobstate(jobp, &newstate, &newsub, 1);
svr_setjobstate(jobp, newstate, newsub);
}
req_reject(r, 0, req, NULL, NULL);
}
else
{
reply_ack(req);
}
return;
} /* END post_movejob() */
static void post_routejob(
struct work_task *pwt)
{
int newstate;
int newsub;
int r;
int stat = pwt->wt_aux;
char *id = "post_routejob";
job *jobp = (job *)pwt->wt_parm1;
if (WIFEXITED(stat))
{
r = WEXITSTATUS(stat);
}
else
{
r = 2;
sprintf(log_buffer, msg_badexit,
stat);
strcat(log_buffer, id);
log_event(
PBSEVENT_SYSTEM,
PBS_EVENTCLASS_JOB,
jobp->ji_qs.ji_jobid,
log_buffer);
}
switch (r)
{
case 0: /* normal return, job was routed */
if (jobp->ji_qs.ji_svrflags & JOB_SVFLG_StagedIn)
remove_stagein(jobp);
if (jobp->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_COPIED)
remove_checkpoint(jobp);
job_purge(jobp); /* need to remove server job struct */
return;
/*NOTREACHED*/
break;
case 1: /* permanent rejection (or signal) */
if (jobp->ji_qs.ji_substate == JOB_SUBSTATE_ABORT)
{
/* job delete in progress, just set to queued status */
svr_setjobstate(jobp, JOB_STATE_QUEUED, JOB_SUBSTATE_ABORT);
return;
}
add_dest(jobp); /* else mark destination as bad */
/* fall through */
default : /* try routing again */
/* force re-eval of job state out of Transit */
svr_evaljobstate(jobp, &newstate, &newsub, 1);
svr_setjobstate(jobp, newstate, newsub);
if ((r = job_route(jobp)) == PBSE_ROUTEREJ)
job_abt(&jobp, pbse_to_txt(PBSE_ROUTEREJ));
else if (r != 0)
job_abt(&jobp, msg_routexceed);
break;
} /* END switch (r) */
return;
} /* END post_routejob() */
void finish_moving_processing(
job *pjob,
struct batch_request *req,
int status)
{
char log_buf[LOCAL_LOG_BUF_SIZE];
int newstate;
int newsub;
if (req->rq_type != PBS_BATCH_MoveJob)
{
sprintf(log_buf, "bad request type %d\n", req->rq_type);
log_err(-1, __func__, log_buf);
return;
}
if (pjob == NULL)
return;
switch (status)
{
case LOCUTION_SUCCESS:
/* purge server's job structure */
if (pjob->ji_qs.ji_svrflags & JOB_SVFLG_StagedIn)
remove_stagein(&pjob);
if (pjob != NULL)
{
if (pjob->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_COPIED)
remove_checkpoint(&pjob);
}
snprintf(log_buf, sizeof(log_buf), "%s", msg_movejob);
snprintf(log_buf + strlen(log_buf), sizeof(log_buf) - strlen(log_buf), msg_manager,
req->rq_ind.rq_move.rq_destin, req->rq_user, req->rq_host);
if (pjob != NULL)
svr_job_purge(pjob);
reply_ack(req);
break;
default:
status = PBSE_ROUTEREJ;
if (pjob != NULL)
{
/* force re-eval of job state out of Transit */
svr_evaljobstate(*pjob, newstate, newsub, 1);
svr_setjobstate(pjob, newstate, newsub, FALSE);
unlock_ji_mutex(pjob, __func__, "3", LOGLEVEL);
}
req_reject(status, 0, req, NULL, NULL);
break;
} /* END switch (status) */
} /* END finish_moving_processing() */
void finish_routing_processing(
job *pjob,
int status)
{
int newstate;
int newsub;
if (pjob == NULL)
return;
if (LOGLEVEL >= 10)
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, pjob->ji_qs.ji_jobid);
switch (status)
{
case LOCUTION_SUCCESS: /* normal return, job was routed */
if (pjob->ji_qs.ji_svrflags & JOB_SVFLG_StagedIn)
remove_stagein(&pjob);
if (pjob != NULL)
{
if (pjob->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_COPIED)
remove_checkpoint(&pjob);
if (pjob != NULL)
svr_job_purge(pjob); /* need to remove server job struct */
}
break;
case LOCUTION_FAIL: /* permanent rejection (or signal) */
if (pjob->ji_qs.ji_substate == JOB_SUBSTATE_ABORT)
{
/* job delete in progress, just set to queued status */
svr_setjobstate(pjob, JOB_STATE_QUEUED, JOB_SUBSTATE_ABORT, FALSE);
svr_mailowner(pjob, 'a', TRUE, "Couldn't route job to remote server");
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
return;
}
add_dest(pjob); /* else mark destination as bad */
/* fall through */
default: /* try routing again */
svr_mailowner(pjob, 'a', TRUE, "Couldn't route job to remote server");
/* force re-eval of job state out of Transit */
svr_evaljobstate(*pjob, newstate, newsub, 1);
svr_setjobstate(pjob, newstate, newsub, FALSE);
if ((status = job_route(pjob)) == PBSE_ROUTEREJ)
job_abt(&pjob, pbse_to_txt(PBSE_ROUTEREJ));
else if (status != 0)
job_abt(&pjob, msg_routexceed);
else
unlock_ji_mutex(pjob, __func__, "2", LOGLEVEL);
break;
} /* END switch (status) */
return;
} /* END finish_routing_processing() */