コード例 #1
0
ファイル: sbd.policy.c プロジェクト: spandon/openlava
static int
cleanupMigJob(struct jobCard *jp)
{
    static char fname[] = "cleanupMigJob()";
    int pid;


    unlockHosts (jp, jp->jobSpecs.numToHosts);

    if (!jp->jobSpecs.postCmd || jp->jobSpecs.postCmd[0] == '\0')
        return 0;


    if ((pid = fork()) < 0) {
	ls_syslog(LOG_ERR, I18N_JOB_FAIL_S_M, fname,
	    lsb_jobid2str(jp->jobSpecs.jobId), "fork");
        lsb_merr2(_i18n_msg_get(ls_catd , NL_SETN, 700,
	    "Unable to fork a child to run the queue's post-exec command for job <%s>.  Please run <%s> manually if necessary.\n"), /* catgets 700 */
	    lsb_jobid2str(jp->jobSpecs.jobId),
	    jp->jobSpecs.postCmd);
	return (pid);
    }

    if (pid)
	return (pid);



    closeBatchSocket();
    putEnv(LS_EXEC_T, "END");

    if (postJobSetup(jp) == -1) {
	ls_syslog(LOG_ERR, I18N_JOB_FAIL_S, fname,
	    lsb_jobid2str(jp->jobSpecs.jobId), "postJobSetup");
	lsb_merr2(_i18n_msg_get(ls_catd , NL_SETN, 701,
	    "Unable to setup the environment for job <%s> to run the queue's post exec.  Please run <%s> manually if necessary.\n"), /* catgets 701 */
	    lsb_jobid2str(jp->jobSpecs.jobId),
	    jp->jobSpecs.postCmd);
	exit(-1);
    }

    runQPost(jp);
    exit(0);
}
コード例 #2
0
ファイル: sbd.serv.c プロジェクト: CraigNoble/openlava
void
do_newjob(XDR *xdrs, int chfd, struct LSFHeader *reqHdr)
{
    static char        fname[] = "do_newjob()";
    char               reply_buf[MSGSIZE];
    XDR                xdrs2;
    struct jobSpecs    jobSpecs;
    struct jobReply    jobReply;
    struct jobCard     *jp;
    sbdReplyType       reply;
    struct LSFHeader   replyHdr;
    char               *replyStruct;
    struct lsfAuth     *auth = NULL;

    memset(&jobReply, 0, sizeof(struct jobReply));

    if (!xdr_jobSpecs(xdrs, &jobSpecs, reqHdr)) {
	reply = ERR_BAD_REQ;
	ls_syslog(LOG_ERR, I18N_FUNC_FAIL, fname, "xdr_jobSpecs");
	goto sendReply;
    }


    for (jp = jobQueHead->forw; (jp != jobQueHead); jp = jp->forw) {
        if (jp->jobSpecs.jobId == jobSpecs.jobId) {

	    jobReply.jobId = jp->jobSpecs.jobId;
	    jobReply.jobPid = jp->jobSpecs.jobPid;
	    jobReply.jobPGid = jp->jobSpecs.jobPGid;
	    jobReply.jStatus = jp->jobSpecs.jStatus;
	    reply = ERR_NO_ERROR;
	    goto sendReply;
	}
    }

    jp = calloc(1, sizeof(struct jobCard));
    if (jp == NULL) {
	ls_syslog(LOG_ERR, I18N_JOB_FAIL_S_M, fname,
                  lsb_jobid2str(jobSpecs.jobId), "calloc");
	reply = ERR_MEM;
	goto sendReply;
    }

    memcpy((char *) &jp->jobSpecs, (char *) &jobSpecs,
	   sizeof(struct jobSpecs));

    jp->jobSpecs.jStatus &= ~JOB_STAT_MIG;
    jp->jobSpecs.startTime = now;
    jp->jobSpecs.reasons = 0;
    jp->jobSpecs.subreasons = 0;
    /* Initialize the core number
     */
    jp->core_num = -1;

    if (jp->jobSpecs.jAttrib & Q_ATTRIB_EXCLUSIVE) {
	if (lockHosts (jp) < 0) {
	    ls_syslog(LOG_ERR, I18N_JOB_FAIL_S, fname,
                      lsb_jobid2str(jp->jobSpecs.jobId), "lockHosts");
            unlockHosts (jp, jp->jobSpecs.numToHosts);
	    reply = ERR_LOCK_FAIL;
	    freeWeek(jp->week);
	    FREEUP(jp);
	    goto sendReply;
        }
    }
    jp->runTime = 0;
    if (initJobCard(jp, &jobSpecs, (int *)&reply) < 0) {

	if (jp->jobSpecs.jAttrib & Q_ATTRIB_EXCLUSIVE) {
	    unlockHosts (jp, jp->jobSpecs.numToHosts);
	}
	FREEUP(jp);
	goto sendReply;
    }

    jp->execJobFlag = 0;

    if (jp->runTime < 0) {
        jp->runTime = 0;
    }
    jp->execGid = 0;
    jp->execUsername[0] = '\0';
    jp->jobSpecs.execUid = -1;
    jp->jobSpecs.execUsername[0] = '\0';

    if (jp->jobSpecs.jobSpoolDir[0] != '\0') {
        char *tmp;

        if ((tmp = getUnixSpoolDir (jp->jobSpecs.jobSpoolDir)) == NULL) {

            jp->jobSpecs.jobSpoolDir[0] = '\0';
        }
    }

    if ((logclass & LC_TRACE) && jp->jobSpecs.jobSpoolDir[0] != 0) {
        ls_syslog(LOG_DEBUG,
                  "%s: the SpoolDir for  job <%s>  is %s \n",
                  fname, lsb_jobid2str(jp->jobSpecs.jobId),
                  jp->jobSpecs.jobSpoolDir);
    }
    if (jp->jobSpecs.options & SUB_PRE_EXEC)
	SBD_SET_STATE(jp, (JOB_STAT_RUN | JOB_STAT_PRE_EXEC))
        else
            SBD_SET_STATE(jp, JOB_STAT_RUN);

    reply = job_exec(jp, chfd);

    if (reply != ERR_NO_ERROR) {
	ls_syslog(LOG_ERR, I18N_JOB_FAIL_S, fname,
                  lsb_jobid2str(jp->jobSpecs.jobId), "job_exec");
	if (jp->jobSpecs.jAttrib & Q_ATTRIB_EXCLUSIVE) {
            unlockHosts (jp, jp->jobSpecs.numToHosts);
	}
	deallocJobCard(jp);
    } else {
	jobReply.jobId = jp->jobSpecs.jobId;
	jobReply.jobPid = jp->jobSpecs.jobPid;
	jobReply.jobPGid = jp->jobSpecs.jobPGid;
	jobReply.jStatus = jp->jobSpecs.jStatus;
    }


sendReply:
    xdr_lsffree(xdr_jobSpecs, (char *)&jobSpecs, reqHdr);
    xdrmem_create(&xdrs2, reply_buf, MSGSIZE, XDR_ENCODE);
    initLSFHeader_(&replyHdr);
    replyHdr.opCode = reply;
    replyStruct = (reply == ERR_NO_ERROR) ? (char *) &jobReply : (char *) NULL;
    if (!xdr_encodeMsg(&xdrs2, replyStruct, &replyHdr, xdr_jobReply, 0, auth)) {
	ls_syslog(LOG_ERR, I18N_FUNC_FAIL, fname, "xdr_jobReply");
	lsb_merr(_i18n_msg_get(ls_catd , NL_SETN, 5804,
			       "Fatal error: xdr_jobReply() failed; sbatchd relifing")); /* catgets 5804 */
	relife();
    }

    if (chanWrite_(chfd, reply_buf, XDR_GETPOS(&xdrs2)) <= 0) {
	ls_syslog(LOG_ERR, _i18n_msg_get(ls_catd , NL_SETN, 5805,
					 "%s: Sending jobReply (len=%d) to master failed: %m"), /* catgets 5805 */
		  fname, XDR_GETPOS(&xdrs2));
    }

    xdr_destroy(&xdrs2);


    if (reply == ERR_NO_ERROR && !daemonParams[LSB_BSUBI_OLD].paramValue &&
	PURE_INTERACTIVE(&jp->jobSpecs)) {
  	if (status_job (BATCH_STATUS_JOB, jp, jp->jobSpecs.jStatus,
		        ERR_NO_ERROR) < 0) {
            jp->notReported++;
	}
    }

}