void
checkFinish (void)
{
struct jobCard *jobCard, *nextJob;
for (jobCard = jobQueHead->forw; (jobCard != jobQueHead);
jobCard = nextJob) {
nextJob = jobCard->forw;
if (!(IS_FINISH(jobCard->jobSpecs.jStatus))
&& !(IS_POST_FINISH(jobCard->jobSpecs.jStatus) ) ) {
if ( (jobsig(jobCard, 0, FALSE) < 0)
|| ( (jobCard->jobSpecs.jAttrib & JOB_FORCE_KILL)
&& (jobCard->jobSpecs.termTime
< time(0)-MAX(6,jobTerminateInterval*3)) ) ) {
jobGone (jobCard);
}
}
if (jobCard->jobSpecs.actPid) {
if (killpg(jobCard->jobSpecs.actPid, SIGCONT) == 0)
continue;
if (kill(jobCard->jobSpecs.actPid, SIGCONT) == 0)
continue;
if (jobCard->cleanupPid > 0 &&
kill(jobCard->cleanupPid, SIGCONT) == 0)
continue;
sigActEnd(jobCard);
continue;
}
if (IS_FINISH(jobCard->jobSpecs.jStatus)
|| IS_POST_FINISH(jobCard->jobSpecs.jStatus)
|| (jobCard->jobSpecs.jStatus & JOB_STAT_PEND)) {
job_finish (jobCard, TRUE);
}
}
}
int
status_job (mbdReqType reqType,
struct jobCard *jp, int newStatus, sbdReplyType err)
{
static char fname[] = "status_job()";
static int seq = 1;
static char lastHost[MAXHOSTNAMELEN];
int reply;
char *request_buf;
char *reply_buf = NULL;
XDR xdrs;
struct LSFHeader hdr;
int cc;
struct statusReq statusReq;
int flags;
int i;
int len;
struct lsfAuth *auth = NULL;
if ((logclass & LC_TRACE) && (logclass & LC_SIGNAL))
ls_syslog (LOG_DEBUG, "%s: Entering ... regType %d jobId %s",
fname, reqType, lsb_jobid2str (jp->jobSpecs.jobId));
if (newStatus == JOB_STAT_EXIT)
{
jp->userJobSucc = FALSE;
}
if (MASK_STATUS (newStatus) == JOB_STAT_DONE)
{
jp->userJobSucc = TRUE;
}
if (IS_POST_FINISH (newStatus))
{
if (jp->userJobSucc != TRUE)
{
return 0;
}
}
if (masterHost == NULL)
return -1;
if (jp->notReported < 0)
{
jp->notReported = -INFINIT_INT;
return (0);
}
statusReq.jobId = jp->jobSpecs.jobId;
statusReq.actPid = jp->jobSpecs.actPid;
statusReq.jobPid = jp->jobSpecs.jobPid;
statusReq.jobPGid = jp->jobSpecs.jobPGid;
statusReq.newStatus = newStatus;
statusReq.reason = jp->jobSpecs.reasons;
statusReq.subreasons = jp->jobSpecs.subreasons;
statusReq.sbdReply = err;
statusReq.lsfRusage = jp->lsfRusage;
statusReq.execUid = jp->jobSpecs.execUid;
statusReq.numExecHosts = 0;
statusReq.execHosts = NULL;
statusReq.exitStatus = jp->w_status;
statusReq.execCwd = jp->jobSpecs.execCwd;
statusReq.execHome = jp->jobSpecs.execHome;
statusReq.execUsername = jp->execUsername;
statusReq.queuePostCmd = "";
statusReq.queuePreCmd = "";
statusReq.msgId = jp->delieveredMsgId;
if (IS_FINISH (newStatus))
{
if (jp->maxRusage.mem > jp->runRusage.mem)
jp->runRusage.mem = jp->maxRusage.mem;
if (jp->maxRusage.swap > jp->runRusage.swap)
jp->runRusage.swap = jp->maxRusage.swap;
if (jp->maxRusage.stime > jp->runRusage.stime)
jp->runRusage.stime = jp->maxRusage.stime;
if (jp->maxRusage.utime > jp->runRusage.utime)
jp->runRusage.utime = jp->maxRusage.utime;
}
statusReq.runRusage.mem = jp->runRusage.mem;
statusReq.runRusage.swap = jp->runRusage.swap;
statusReq.runRusage.utime = jp->runRusage.utime;
statusReq.runRusage.stime = jp->runRusage.stime;
statusReq.runRusage.npids = jp->runRusage.npids;
statusReq.runRusage.pidInfo = jp->runRusage.pidInfo;
statusReq.runRusage.npgids = jp->runRusage.npgids;
statusReq.runRusage.pgid = jp->runRusage.pgid;
statusReq.actStatus = jp->actStatus;
statusReq.sigValue = jp->jobSpecs.actValue;
statusReq.seq = seq;
seq++;
if (seq >= MAX_SEQ_NUM)
seq = 1;
len = 1024 + ALIGNWORD_ (sizeof (struct statusReq));
len += ALIGNWORD_ (strlen (statusReq.execHome)) + 4 +
ALIGNWORD_ (strlen (statusReq.execCwd)) + 4 +
ALIGNWORD_ (strlen (statusReq.execUsername)) + 4;
for (i = 0; i < statusReq.runRusage.npids; i++)
len += ALIGNWORD_ (sizeof (struct pidInfo)) + 4;
for (i = 0; i < statusReq.runRusage.npgids; i++)
len += ALIGNWORD_ (sizeof (int)) + 4;
if (logclass & (LC_TRACE | LC_COMM))
ls_syslog (LOG_DEBUG, "%s: The length of the job message is: <%d>", fname,
len);
if ((request_buf = malloc (len)) == NULL)
{
ls_syslog (LOG_ERR, I18N_FUNC_FAIL_M, fname, "malloc");
return (-1);
}
xdrmem_create (&xdrs, request_buf, len, XDR_ENCODE);
initLSFHeader_ (&hdr);
hdr.opCode = reqType;
if (!xdr_encodeMsg (&xdrs, (char *) &statusReq, &hdr, xdr_statusReq, 0,
auth))
{
ls_syslog (LOG_ERR, I18N_JOB_FAIL_S_M,
fname, lsb_jobid2str (jp->jobSpecs.jobId), "xdr_statusReq");
lsb_merr2 (I18N_FUNC_FAIL, fname, "xdr_statusReq");
xdr_destroy (&xdrs);
FREEUP (request_buf);
relife ();
}
flags = CALL_SERVER_NO_HANDSHAKE;
if (statusChan >= 0)
flags |= CALL_SERVER_USE_SOCKET;
if (reqType == BATCH_RUSAGE_JOB)
flags |= CALL_SERVER_NO_WAIT_REPLY;
if (logclass & LC_COMM)
ls_syslog (LOG_DEBUG1, "%s: before call_server statusChan=%d flags=%d",
fname, statusChan, flags);
cc = call_server (masterHost,
mbd_port,
request_buf,
XDR_GETPOS (&xdrs),
&reply_buf,
&hdr,
connTimeout, readTimeout, &statusChan, NULL, NULL, flags);
if (cc < 0)
{
statusChan = -1;
if (!equalHost_ (masterHost, lastHost))
{
if (errno != EINTR)
ls_syslog (LOG_DEBUG,
"%s: Failed to reach mbatchd on host <%s> for job <%s>: %s",
fname, masterHost, lsb_jobid2str (jp->jobSpecs.jobId),
lsb_sysmsg ());
strcpy (lastHost, masterHost);
}
xdr_destroy (&xdrs);
FREEUP (request_buf);
failcnt++;
return (-1);
}
else if (cc == 0)
{
}
failcnt = 0;
lastHost[0] = '\0';
xdr_destroy (&xdrs);
FREEUP (request_buf);
if (cc)
free (reply_buf);
if (flags & CALL_SERVER_NO_WAIT_REPLY)
{
struct timeval timeval;
timeval.tv_sec = 0;
timeval.tv_usec = 0;
if (rd_select_ (chanSock_ (statusChan), &timeval) == 0)
{
jp->needReportRU = FALSE;
jp->lastStatusMbdTime = now;
return 0;
}
CLOSECD (statusChan);
if (logclass & LC_COMM)
ls_syslog (LOG_DEBUG1,
"%s: Job <%s> rd_select() failed, assume connection broken",
fname, lsb_jobid2str (jp->jobSpecs.jobId));
return (-1);
}
reply = hdr.opCode;
switch (reply)
{
case LSBE_NO_ERROR:
case LSBE_LOCK_JOB:
jp->needReportRU = FALSE;
jp->lastStatusMbdTime = now;
if (reply == LSBE_LOCK_JOB)
{
if (IS_SUSP (jp->jobSpecs.jStatus))
jp->jobSpecs.reasons |= SUSP_MBD_LOCK;
else
ls_syslog (LOG_ERR, _i18n_msg_get (ls_catd, NL_SETN, 5204, "%s: Job <%s> is in status <%x> and mbatchd wants to lock it, ignored."), /* catgets 5204 */
fname,
lsb_jobid2str (jp->jobSpecs.jobId),
jp->jobSpecs.jStatus);
}
return (0);
case LSBE_NO_JOB:
if (!IS_POST_FINISH (jp->jobSpecs.jStatus))
{
ls_syslog (LOG_ERR, _i18n_msg_get (ls_catd, NL_SETN, 5205, "%s: Job <%s> is forgotten by mbatchd on host <%s>, ignored."), fname, lsb_jobid2str (jp->jobSpecs.jobId), masterHost); /* catgets 5205 */
}
jp->notReported = -INFINIT_INT;
return (0);
case LSBE_STOP_JOB:
if (jobsig (jp, SIGSTOP, TRUE) < 0)
SET_STATE (jp->jobSpecs.jStatus, JOB_STAT_EXIT);
else
{
SET_STATE (jp->jobSpecs.jStatus, JOB_STAT_USUSP);
jp->jobSpecs.reasons |= SUSP_USER_STOP;
}
return (-1);
case LSBE_SBATCHD:
ls_syslog (LOG_ERR, _i18n_msg_get (ls_catd, NL_SETN, 5206, "%s: mbatchd on host <%s> doesn't think I'm configured as a batch server when I report the status for job <%s>"), /* catgets 5206 */
fname, masterHost, lsb_jobid2str (jp->jobSpecs.jobId));
return (-1);
default:
ls_syslog (LOG_ERR, _i18n_msg_get (ls_catd, NL_SETN, 5207, "%s: Illegal reply code <%d> from mbatchd on host <%s> for job <%s>"), /* catgets 5207 */
fname,
reply, masterHost, lsb_jobid2str (jp->jobSpecs.jobId));
return (-1);
}
}
