/*
* Record the reason that the current candidate job cannot currently run.
* When it is decided that the job will remain queued, place the reason
* string in the comment field of the job structure.
*/
void
schd_comment_job(Job *job, char *reason, int optional)
{
char *id = "schd_comment_job";
char *msg_ptr;
char *old_msg;
/*
* If the 'optional' argument is true, then this comment is optional.
* Do not bother commenting this job if this is not the first time it
* has been seen, and it has been recently modified (hopefully it was
* a comment change). If there is no comment for the job, comment it
* this time.
*/
if (optional &&
(!schd_FirstRun) &&
(job->comment != NULL) &&
!(job->flags & JFLAGS_FIRST_SEEN) &&
(MIN_COMMENT_AGE && ((schd_TimeNow - job->mtime) < MIN_COMMENT_AGE)))
{
return;
}
if (reason == NULL)
reason = "";
old_msg = job->comment;
/* If there is no old message, or they are different, set it. */
if ((old_msg == NULL) || (strcmp(reason, old_msg) != 0))
{
msg_ptr = schd_strdup(reason);
/* Alter PBS' view of the job. */
schd_alterjob(connector, job, ATTR_comment, msg_ptr, NULL);
/* Copy the new comment into the job field. */
if (job->comment)
free(job->comment);
job->comment = msg_ptr;
if (job->comment == NULL)
{
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id,
"schd_strdup(job->comment)");
return;
}
}
return;
}
/*
* This function takes a pointer to a struct batch_status for a job, and
* fills in the appropriate fields of the supplied job struct. It returns
* the number of items that were found.
*/
int
schd_get_jobinfo(Batch_Status *bs, Job *job)
{
int changed = 0;
int istrue;
char tmp_str[120];
char *id = "schd_get_jobinfo";
char *host;
char *p, *tmp_p, *var_p;
AttrList *attr;
memset((void *)job, 0, sizeof(Job));
job->jobid = schd_strdup(bs->name);
if (job->jobid == NULL)
{
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id,
"schd_strdup(bs->name)");
return (-1);
}
changed ++;
for (attr = bs->attribs; attr != NULL; attr = attr->next)
{
/*
* If this is the 'owner' field, chop it into 'owner' and 'host'
* fields, and copy them into the Job struct.
*/
if (!strcmp(attr->name, ATTR_owner))
{
/* Look for the '@' that separates user and hostname. */
strcpy(tmp_str, attr->value);
host = strchr(tmp_str, '@');
if (host)
{
*host = '\0'; /* Replace '@' with NULL (ends username). */
host ++; /* Move to first character of hostname. */
}
job->owner = schd_strdup(tmp_str);
if (job->owner == NULL)
{
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id,
"schd_strdup(job->owner)");
return (-1);
}
changed ++;
continue;
}
/* The group to which to charge the resources for this job. */
if (!strcmp(attr->name, ATTR_egroup))
{
job->group = schd_strdup(attr->value);
if (job->group == NULL)
{
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id,
"schd_strdup(job->group)");
return (-1);
}
changed ++;
continue;
}
/* The comment currently assigned to this job. */
if (!strcmp(attr->name, ATTR_comment))
{
job->comment = schd_strdup(attr->value);
if (job->comment == NULL)
{
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id,
"schd_strdup(job->comment)");
return (-1);
}
changed ++;
continue;
}
/* The host on which this job is running (or was running for
* suspended or checkpointed jobs. */
if (!strcmp(attr->name, ATTR_exechost))
{
job->exechost = schd_strdup(attr->value);
if (job->exechost == NULL)
{
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id,
"schd_strdup(job->exechost)");
return (-1);
}
changed ++;
continue;
}
if (!strcmp(attr->name, ATTR_inter))
{
/* Is this job interactive or not? */
if (schd_val2bool(attr->value, &istrue) == 0)
{
if (istrue)
job->flags |= JFLAGS_INTERACTIVE;
else
job->flags &= ~JFLAGS_INTERACTIVE;
changed ++;
}
else
{
DBPRT(("%s: can't parse %s = %s into boolean\n", id,
attr->name, attr->value));
}
continue;
}
if (!strcmp(attr->name, ATTR_state))
{
/* State is one of 'R', 'Q', 'E', etc. */
job->state = attr->value[0];
changed ++;
continue;
}
if (!strcmp(attr->name, ATTR_queue))
{
job->qname = schd_strdup(attr->value);
if (job->qname == NULL)
{
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id,
"schd_strdup(job->qname)");
return (-1);
}
job->flags |= JFLAGS_QNAME_LOCAL;
changed ++;
continue;
}
if (!strcmp(attr->name, ATTR_v))
{
var_p = schd_strdup(attr->value);
if (var_p == NULL)
{
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id,
"schd_strdup(Variable_List)");
return (-1);
}
p = NULL;
tmp_p = strstr(var_p, "PBS_O_QUEUE");
if (tmp_p)
{
p = strtok(tmp_p, "=");
p = strtok(NULL, ", ");
}
if (p != NULL)
{
job->oqueue = schd_strdup(p);
}
else
{
/* if the originating queue is unknown, default
* to the locally defined "submit" queue.
*/
job->oqueue = schd_strdup(schd_SubmitQueue->queue->qname);
}
free(var_p);
changed ++;
continue;
}
if (!strcmp(attr->name, ATTR_l))
{
if (!strcmp(attr->resource, "arch"))
{
job->arch = schd_strdup(attr->value);
changed ++;
}
else if (!strcmp(attr->resource, "mem"))
{
job->memory = schd_val2byte(attr->value);
changed ++;
}
else if (!strcmp(attr->resource, "ncpus"))
{
job->ncpus = atoi(attr->value);
changed ++;
}
else if (!strcmp(attr->resource, "walltime"))
{
job->walltime = schd_val2sec(attr->value);
changed ++;
}
/* That's all for requested resources. */
continue;
}
if (!strcmp(attr->name, ATTR_used))
{
if (!strcmp(attr->resource, "walltime"))
{
job->walltime_used = schd_val2sec(attr->value);
changed ++;
}
/* No other interesting cases. */
continue;
}
/* Creation time attribute. */
if (!strcmp(attr->name, ATTR_ctime))
{
/* How long ago was it put in the queue ? */
job->time_queued = schd_TimeNow - atoi(attr->value);
continue;
}
/* Modified time attribute. */
if (!strcmp(attr->name, ATTR_mtime))
{
/* When was the job last modified? */
job->mtime = atoi(attr->value);
continue;
}
/* Job Substate attribute. */
if (!strcmp(attr->name, ATTR_substate))
{
if (atoi(attr->value) == 43 /* JOB_SUBSTATE_SUSPEND */)
job->flags |= JFLAGS_SUSPENDED;
continue;
}
/*
* When was the job last eligible to run? When a user-hold is
* released, this value is updated to the current time. This
* prevents users from gaining higher priority from holding their
* jobs.
*/
if (!strcmp(attr->name, ATTR_etime))
{
job->eligible = schd_TimeNow - atoi(attr->value);
continue;
}
}
if (job->memory < 1)
{
job->memory = get_default_mem(job->oqueue);
schd_alterjob(connector, job, ATTR_l, schd_byte2val(job->memory), "mem");
changed++;
}
/*
* If this job is in the "Running" or "Suspended" state, compute how
* many seconds remain until it is completed.
*/
if (job->state == 'R' || job->state == 'S')
{
job->time_left = job->walltime - job->walltime_used;
}
/*
* If this job was enqueued since the last time we ran, set the job
* flag to indicate that we have not yet seen this job. This makes it
* a candidate for additional processing. There may be some inaccuracy,
* since the time_t has resolution of 1 second. Attempt to err on the
* side of caution.
*/
if ((job->state == 'Q') && (job->time_queued != UNSPECIFIED))
{
if (job->time_queued <= (schd_TimeNow - schd_TimeLast))
{
job->flags |= JFLAGS_FIRST_SEEN;
}
}
/*
* If this job was previously running and is now queued, then we
* need to (a) flag it as having been checkpointed, and (b) move
* it back to the submit queue, if its not already there.
*/
if (job->exechost && job->state == 'Q')
{
job->flags |= JFLAGS_CHKPTD;
if (strcmp(job->qname, schd_SubmitQueue->queue->qname))
{
sprintf(log_buffer, "moving Q'd job %s back to SUBMIT Q",
job->jobid);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
pbs_movejob(connector, job->jobid, schd_SubmitQueue->queue->qname,
NULL);
}
}
/*
* if this job is currently Suspended (a substate of 'R'unning), then
* pretend its queued, so that the scheduling logic will work.
*/
if (job->state == 'S')
{
job->state = 'Q';
job->flags |= JFLAGS_SUSPENDED;
}
/* if this job is suspended, checkpointed, or otherwise "queued"
* on an exection queue, update the internal representation of
* to pretend it is really on the submit queue.
*/
if ((job->flags & JFLAGS_SUSPENDED) || (job->flags & JFLAGS_CHKPTD))
{
free(job->qname);
job->qname = schd_strdup(schd_SubmitQueue->queue->qname);
}
/*
* If this job came from the EXPRESS queue, set the flag so that it
* will be treated with the highest of priority.
*/
if (!strcmp(job->oqueue, schd_EXPRESS_Q_NAME))
job->flags |= JFLAGS_PRIORITY;
/*
* If the 'etime' attribute wasn't found, set it to the time the job has
* been queued. Most jobs will be eligible to run their entire lifetime.
* The exception is a job that has been held - if it was a user hold,
* the release will reset the etime to the latest value.
* If not eligible time was given, use the job's creation time.
*/
if (!job->eligible)
job->eligible = job->time_queued;
/* if this job has waited too long, and its queue is NOT over its
* shares, then bump it up in priority.
*/
if (job->eligible > schd_MAX_WAIT_TIME && job->sort_order <= 100)
job->flags |= JFLAGS_WAITING;
return (changed);
}
static int
bump_rsrc_requests(Job *job, int cpu_req, size_t mem_req)
{
char *id = "bump_rsrc_requests";
char *val;
char buf[64];
int bumped = 0;
/*
* If a job gives the "wrong" value for the memory request (for the
* number of nodes required to fulfill the request), then bump the
* memory request to the amount of memory the assigned nodes would
* consume.
*/
if ((mem_req == 0) || (mem_req != (job->nodes * MB_PER_NODE)))
{
/* Make a printable version of the requested memory. */
strcpy(buf, schd_byte2val(mem_req));
/* Compute the "right" memory request, based on the nodes. */
val = schd_byte2val(job->nodes * MB_PER_NODE);
if (val == NULL)
return (1);
if (schd_alterjob(connector, job, ATTR_l, val, "mem"))
{
DBPRT(("%s: Failed to set job %s \"mem\" attr to %s\n", id,
job->jobid, val));
return (1);
}
bumped++;
}
/*
* If a job gives the "wrong" value for the CPU request (for the
* number of nodes required to fulfill the request), then bump the
* CPU request to the number of CPUs the assigned nodes would
* consume.
*/
if ((cpu_req == 0) || (cpu_req != (job->nodes * PE_PER_NODE)))
{
/* Compute the "right" memory request, based on the nodes. */
sprintf(buf, "%d", (job->nodes * PE_PER_NODE));
if (schd_alterjob(connector, job, ATTR_l, buf, "ncpus"))
{
DBPRT(("%s: Failed to set job %s \"ncpus\" attr to %s\n", id,
job->jobid, buf));
return (1);
}
bumped++;
}
if (bumped)
{
strncpy(buf, schd_byte2val(job->nodes * MB_PER_NODE), sizeof(buf) - 1);
sprintf(log_buffer, "%s cpu/mem (%d/%s) bumped to %d/%s (%d nodes)",
job->jobid, cpu_req, schd_byte2val(mem_req),
job->nodes * PE_PER_NODE, buf, job->nodes);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
DBPRT(("%s: %s\n", id, log_buffer));
}
return (0);
}
