int
inIdxList(LS_LONG_INT jobId, struct idxList *idxList)
{
struct idxList *idx;
if (idxList) {
for (idx = idxList; idx; idx = idx->next) {
if (LSB_ARRAY_IDX(jobId) < idx->start ||
LSB_ARRAY_IDX(jobId) > idx->end)
continue;
if (((LSB_ARRAY_IDX(jobId)-idx->start) % idx->step) == 0)
return(TRUE);
}
return(FALSE);
}
return(TRUE);
}
char *
lsb_jobid2str (LS_LONG_INT jobId)
{
static char string[32];
if (LSB_ARRAY_IDX(jobId) == 0) {
sprintf(string, "%d", LSB_ARRAY_JOBID(jobId));
}
else {
sprintf(string, "%d[%d]", LSB_ARRAY_JOBID(jobId),
LSB_ARRAY_IDX(jobId));
}
return(string);
}
static void
lsfdrmaa_job_control( fsd_job_t *self, int action )
{
/*
* XXX: waiting for job state change was removed
* since it is not required for drmaa_control
* to return after change completes.
*/
lsfdrmaa_job_t *lsf_self = (lsfdrmaa_job_t*)self;
LS_LONG_INT job_id;
int signal;
fsd_log_enter(( "({job_id=%s}, action=%d)", self->job_id, action ));
job_id = lsf_self->int_job_id;
switch( action )
{
case DRMAA_CONTROL_SUSPEND:
case DRMAA_CONTROL_HOLD:
signal = SIGSTOP;
break;
case DRMAA_CONTROL_RESUME:
case DRMAA_CONTROL_RELEASE:
signal = SIGCONT;
break;
case DRMAA_CONTROL_TERMINATE:
/* TODO: sending SIGTERM (configurable)? */
signal = SIGKILL;
break;
default:
fsd_exc_raise_fmt(
FSD_ERRNO_INVALID_ARGUMENT,
"job::control: unknown action %d", action );
}
fsd_mutex_lock( &self->session->drm_connection_mutex );
TRY
{
int rc = lsb_signaljob( lsf_self->int_job_id, signal );
fsd_log_debug(( "lsb_signaljob( %d[%d], %d ) = %d",
LSB_ARRAY_JOBID(lsf_self->int_job_id),
LSB_ARRAY_IDX(lsf_self->int_job_id),
signal, rc ));
if( rc < 0 )
fsd_exc_raise_fmt(
FSD_ERRNO_INTERNAL_ERROR,
"job::control: could not send %s to job %s",
fsd_strsignal( signal ), self->job_id
);
}
FINALLY
{
fsd_mutex_unlock( &self->session->drm_connection_mutex );
}
END_TRY
fsd_log_return(( "" ));
}
static void
lsfdrmaa_job_update_status( fsd_job_t *self )
{
lsfdrmaa_job_t *lsf_self = (lsfdrmaa_job_t*)self;
struct jobInfoEnt *volatile job_info = NULL;
bool job_in_queue;
fsd_log_enter(( "({job_id=%s, time_delta=%d})", self->job_id, time(NULL) - self->submit_time ));
do {
fsd_mutex_lock( &self->session->drm_connection_mutex );
TRY
{
int n_records;
int more;
char * username = (lsf_self->int_job_id>0)?"all":NULL;
fsd_log_debug(( "drm connection locked" ));
n_records = lsb_openjobinfo( lsf_self->int_job_id,
NULL, username, NULL, NULL, ALL_JOB );
fsd_log_debug((
"lsb_openjobinfo( %d[%d], NULL, %s, NULL, NULL, ALL_JOB ) =%d",
LSB_ARRAY_JOBID(lsf_self->int_job_id),
LSB_ARRAY_IDX(lsf_self->int_job_id),
username?username:"******",
n_records ));
job_in_queue = n_records > 0;
if(!job_in_queue){
if(!(self->flags & FSD_JOB_CURRENT_SESSION)){
fsd_exc_raise_code( FSD_DRMAA_ERRNO_INVALID_JOB );
}else{/*handling missing job*/
self->on_missing(self);
}
}else{
job_info = lsb_readjobinfo( &more );
fsd_log_debug(( "lsb_readjobinfo(...) =%p: more=%d",
(void*)job_info, more ));
if( job_info == NULL )
fsd_exc_raise_lsf( "lsb_readjobinfo" );
lsf_self->read_job_info( self, job_info );
}
}
FINALLY
{
/* lsfdrmaa_free_job_info( job_info ); */
lsb_closejobinfo();
fsd_log_debug(( "lsb_closejobinfo()" ));
fsd_mutex_unlock( &self->session->drm_connection_mutex );
}
END_TRY
} while( !job_in_queue );
fsd_log_return(( "" ));
}
static void
lsfdrmaa_job_read_job_info( fsd_job_t *self, struct jobInfoEnt *job_info )
{
int status, flags;
fsd_log_enter(( "" ));
{
int i;
fsd_log_debug(( "job status of %s updated from %d[%d]",
self->job_id,
LSB_ARRAY_JOBID(job_info->jobId),
LSB_ARRAY_IDX(job_info->jobId) ));
fsd_log_debug(( "\n status: 0x%x", job_info->status ));
fsd_log_debug(( "\n submitTime: %ld", job_info->submitTime ));
fsd_log_debug(( "\n startTime: %ld", job_info->startTime ));
fsd_log_debug(( "\n endTime: %ld", job_info->startTime ));
fsd_log_debug(( "\n duration: %d", job_info->duration ));
fsd_log_debug(( "\n cpuTime: %f", job_info->cpuTime ));
fsd_log_debug(( "\n cwd: %s", job_info->cwd ));
fsd_log_debug(( "\n fromHost: %s", job_info->fromHost ));
fsd_log_debug(( "\n numExHosts: %d", job_info->numExHosts ));
for( i = 0; i < job_info->numExHosts; i++ )
fsd_log_debug(( "\n exHosts[%d]: %s", i, job_info->exHosts[i] ));
fsd_log_debug(( "\n exitStatus: %d", job_info->exitStatus ));
fsd_log_debug(( "\n execCwd: %s", job_info->execCwd ));
fsd_log_debug(( "\n runRusage.mem: %d", job_info->runRusage.mem ));
fsd_log_debug(( "\n runRusage.swap: %d", job_info->runRusage.swap ));
fsd_log_debug(( "\n runRusage.utime: %d", job_info->runRusage.utime ));
fsd_log_debug(( "\n runRusage.stime: %d", job_info->runRusage.stime ));
fsd_log_debug(( "\n jName: %s", job_info->jName ));
/* fsd_log_debug(( "\n execRusage: %s", job_info->execRusage )); */
}
status = job_info->status;
flags = 0;
if( status & (JOB_STAT_PEND | JOB_STAT_PSUSP) )
flags |= FSD_JOB_QUEUED;
if( status & JOB_STAT_PSUSP )
flags |= FSD_JOB_HOLD;
if( status & (JOB_STAT_RUN | JOB_STAT_USUSP | JOB_STAT_SSUSP) )
flags |= FSD_JOB_RUNNING;
if( status & (JOB_STAT_USUSP | JOB_STAT_SSUSP) )
flags |= FSD_JOB_SUSPENDED;
if( status & (JOB_STAT_DONE | JOB_STAT_EXIT) )
flags |= FSD_JOB_TERMINATED;
if( status & (JOB_STAT_EXIT | JOB_STAT_PERR) )
flags |= FSD_JOB_ABORTED;
self->flags &= ~(FSD_JOB_STATE_MASK | FSD_JOB_ABORTED);
self->flags |= flags;
if( status & (JOB_STAT_WAIT | JOB_STAT_PEND) )
self->state = DRMAA_PS_QUEUED_ACTIVE;
else if( status & JOB_STAT_PSUSP )
self->state = DRMAA_PS_USER_ON_HOLD;
else if( status & JOB_STAT_RUN )
self->state = DRMAA_PS_RUNNING;
else if( status & JOB_STAT_SSUSP )
self->state = DRMAA_PS_SYSTEM_SUSPENDED;
else if( status & JOB_STAT_USUSP )
self->state = DRMAA_PS_USER_SUSPENDED;
else if( status & JOB_STAT_DONE )
self->state = DRMAA_PS_DONE;
else if( status & JOB_STAT_EXIT )
self->state = DRMAA_PS_FAILED;
else if( status & JOB_STAT_PDONE )
self->state = DRMAA_PS_DONE;
else if( status & JOB_STAT_PERR )
self->state = DRMAA_PS_FAILED;
else if( status & JOB_STAT_UNKWN )
self->state = DRMAA_PS_UNDETERMINED;
else
self->state = DRMAA_PS_FAILED;
self->exit_status = job_info->exitStatus & ~0xff;
if( (self->exit_status >> 8) == 0 && (job_info->status & JOB_STAT_EXIT) )
self->exit_status |= 0x01;
self->start_time = job_info->startTime;
self->end_time = job_info->endTime;
self->cpu_usage = job_info->cpuTime;
self->mem_usage = max( self->mem_usage, 1024*job_info->runRusage.mem );
self->vmem_usage = max( self->vmem_usage, 1024*job_info->runRusage.swap );
self->walltime = 60*job_info->duration;
self->n_execution_hosts = job_info->numExHosts;
if( self->execution_hosts == NULL && job_info->exHosts != NULL )
self->execution_hosts
= fsd_explode( (const char*const*)job_info->exHosts, ' ',
job_info->numExHosts );
self->last_update_time = time(NULL);
if( self->state >= DRMAA_PS_DONE )
fsd_cond_broadcast( &self->status_cond );
fsd_log_return(( "" ));
}
void
prtJobStart(struct jobInfoEnt *job, int prtFlag, int jobPid, int tFormat)
{
char prline[MAXLINELEN], tBuff[20];
time_t startTime;
int i = 0;
struct nameList *hostList = NULL;
if (lsbParams[LSB_SHORT_HOSTLIST].paramValue && job->numExHosts > 1
&& strcmp(lsbParams[LSB_SHORT_HOSTLIST].paramValue, "1") == 0) {
hostList = lsb_compressStrList(job->exHosts, job->numExHosts);
if (!hostList) {
exit(99);
}
}
if (tFormat) {
sprintf (tBuff, "%s <%s>", I18N_Job, lsb_jobid2str(job->jobId));
}
else if (LSB_ARRAY_IDX(job->jobId) > 0 )
sprintf (tBuff, " [%d]", LSB_ARRAY_IDX(job->jobId));
else
tBuff[0] = '\0';
if (job->startTime && job->numExHosts) {
if (job->startTime < job->submitTime)
startTime = job->submitTime;
else
startTime = job->startTime;
if ((job->submit.options & SUB_PRE_EXEC)
&& (prtFlag != BJOBS_PRINT)) {
if (prtFlag == BHIST_PRINT_PRE_EXEC)
{
if (tBuff[0] == '\0')
sprintf(prline, "%s: %s",
_i18n_ctime(ls_catd, CTIME_FORMAT_a_b_d_T,
&startTime),
I18N(604, "The pre-exec command is started on")); /* catgets 604 */
else
sprintf(prline, "%s:%s, %s",
_i18n_ctime(ls_catd, CTIME_FORMAT_a_b_d_T,
&startTime),
tBuff,
I18N(605, "the pre-exec command is started on")); /* catgets 605 */
}
else
{
if (tBuff[0] == '\0')
sprintf(prline, "%s: %s",
_i18n_ctime(ls_catd, CTIME_FORMAT_a_b_d_T, &startTime),
I18N(606, "The batch job command is started on")); /*catgets 606 */
else
sprintf(prline, "%s:%s, %s",
_i18n_ctime(ls_catd, CTIME_FORMAT_a_b_d_T, &startTime),
tBuff,
I18N(607, "the batch job command is started on")); /*catgets 607 */
}
} else {
if (jobPid > 0)
{
if (tBuff[0] == '\0')
sprintf(prline, "%s: %s",
_i18n_ctime(ls_catd, CTIME_FORMAT_a_b_d_T, &startTime),
I18N(608, "Started on")); /* catgets 608 */
else
sprintf(prline, "%s:%s %s",
_i18n_ctime(ls_catd, CTIME_FORMAT_a_b_d_T, &startTime),
tBuff,
I18N(609, "started on")); /* catgets 609 */
}
else
{
if (tBuff[0] == '\0')
sprintf(prline, "%s: %s",
_i18n_ctime(ls_catd, CTIME_FORMAT_a_b_d_T, &startTime),
I18N(610, "Dispatched to")); /* catgets 610 */
else
sprintf(prline, "%s: %s %s",
_i18n_ctime(ls_catd, CTIME_FORMAT_a_b_d_T, &startTime),
tBuff,
I18N(611, "dispatched to")); /* catgets 611 */
}
}
prtLineWUF(prline);
if (job->numExHosts > 1) {
sprintf(prline, " %d %s",
job->numExHosts,
I18N(612, "Hosts/Processors")); /* catgets 612 */
prtLineWUF(prline);
}
if (lsbParams[LSB_SHORT_HOSTLIST].paramValue && job->numExHosts > 1
&& strcmp(lsbParams[LSB_SHORT_HOSTLIST].paramValue, "1") == 0) {
for (i = 0; i < hostList->listSize; i++) {
sprintf(prline, " <%d*%s>", hostList->counter[i],
hostList->names[i]);
prtLineWUF(prline);
}
} else {
for (i = 0; i < job->numExHosts; i++) {
sprintf(prline, " <%s>", job->exHosts[i]);
prtLineWUF(prline);
}
}
if (job->execHome && strcmp (job->execHome, "")) {
sprintf(prline, ", %s <%s>",
I18N(615, "Execution Home"), /* catgets 615 */
job->execHome);
prtLineWUF(prline);
}
if (job->execCwd && strcmp (job->execCwd, "")) {
sprintf(prline, ", %s <%s>",
I18N(616, "Execution CWD"), /* catgets 616 */
job->execCwd);
prtLineWUF(prline);
}
if (job->execUsername && strcmp(job->execUsername, "") &&
strcmp(job->user, job->execUsername)) {
sprintf(prline, ", %s <%s>",
I18N(617, "Execution user name"), /* catgets 617 */
job->execUsername);
prtLineWUF(prline);
}
sprintf(prline, ";\n");
prtLineWUF(prline);
}
}
void
prtHeader(struct jobInfoEnt *job, int prt_q, int tFormat)
{
char prline[MAXLINELEN];
if (!tFormat) {
sprintf(prline, "\nJob%s <%s>,", uf_format?"":" Id", lsb_jobid2str(job->jobId));
prtLineWUF(prline);
if (job->submit.options & SUB_JOB_NAME) {
char *jobName, *pos;
jobName = job->submit.jobName;
if ((pos = strchr(jobName, '[')) && LSB_ARRAY_IDX(job->jobId)) {
*pos = '\0';
sprintf(jobName, "%s[%d]", jobName, LSB_ARRAY_IDX(job->jobId));
}
sprintf(prline, " Job Name <%s>,", jobName);
prtLineWUF(prline);
}
}
if (tFormat) {
sprintf(prline, ",");
prtLine(prline);
}
sprintf(prline, " User <%s>,", job->user);
prtLineWUF(prline);
if (lsbMode_ & LSB_MODE_BATCH) {
sprintf(prline, " Project <%s>,", job->submit.projectName);
prtLineWUF(prline);
}
if (job->submit.userGroup && job->submit.userGroup[0] != '\0') {
sprintf(prline, " User Group <%s>,", job->submit.userGroup);
prtLineWUF(prline);
}
if (job->submit.options & SUB_MAIL_USER) {
sprintf(prline, " Mail <%s>,", job->submit.mailUser);
prtLineWUF(prline);
}
if (prt_q) {
sprintf(prline, " Status <%s>, Queue <%s>,",
get_status(job),
job->submit.queue);
prtLineWUF(prline);
}
/* Interactive job */
if (job->submit.options & SUB_INTERACTIVE) {
sprintf(prline, " Interactive");
if (job->submit.options & SUB_PTY) {
strcat(prline, " pseudo-terminal");
if (job->submit.options & (SUB_PTY_SHELL))
strcat(prline, " shell");
}
strcat(prline, " mode,");
prtLineWUF(prline);
}
if (job->jobPriority > 0) {
sprintf(prline, " Job Priority <%d>,", job->jobPriority);
prtLineWUF(prline);
}
if (job->submit.options2 & (SUB2_JOB_CMD_SPOOL)) {
if (tFormat)
sprintf(prline, " Command(Spooled) <%s>", job->submit.command);
else
sprintf(prline, " Command(Spooled) <%s>", job->submit.command);
} else {
if (tFormat)
sprintf(prline, " Command <%s>", job->submit.command);
else
sprintf(prline, " Command <%s>", job->submit.command);
}
prtLineWUF(prline);
if (job->submit.options2 & SUB2_JOB_GROUP) {
sprintf(prline, ", Job Group <%s>", job->submit.job_group);
prtLineWUF(prline);
}
sprintf(prline, "\n");
prtLineWUF(prline);
}
void
prtHeader(struct jobInfoEnt *job, int prt_q, int tFormat)
{
char prline[MAXLINELEN];
if (!tFormat) {
sprintf(prline, "\n%s <%s>,", I18N_Job, lsb_jobid2str(job->jobId));
prtLine(prline);
if (job->submit.options & SUB_JOB_NAME) {
char *jobName, *pos;
jobName = job->submit.jobName;
if ((pos = strchr(jobName, '[')) && LSB_ARRAY_IDX(job->jobId)) {
*pos = '\0';
sprintf(jobName, "%s[%d]", jobName, LSB_ARRAY_IDX(job->jobId));
}
sprintf(prline, (_i18n_msg_get(ls_catd,NL_SETN,552, " Job Name <%s>,")), jobName); /* catgets 552 */
prtLine(prline);
}
}
if (tFormat) {
sprintf(prline, ",");
prtLine(prline);
}
sprintf(prline, " %s <%s>,", I18N_User, job->user);
prtLine(prline);
if (lsbMode_ & LSB_MODE_BATCH) {
sprintf(prline, (_i18n_msg_get(ls_catd,NL_SETN,554, " Project <%s>,")), job->submit.projectName); /* catgets 554 */
prtLine(prline);
}
if (job->submit.options & SUB_MAIL_USER) {
sprintf(prline, (_i18n_msg_get(ls_catd,NL_SETN,556, " Mail <%s>,")), job->submit.mailUser); /* catgets 556 */
prtLine(prline);
}
if (prt_q) {
sprintf(prline, (_i18n_msg_get(ls_catd,NL_SETN,557, " Status <%s>, Queue <%s>,")), /* catgets 557 */
get_status(job),
job->submit.queue);
prtLine(prline);
}
if (job->submit.options & SUB_INTERACTIVE) {
sprintf(prline, (_i18n_msg_get(ls_catd,NL_SETN,558, " Interactive"))); /* catgets 558 */
if (job->submit.options & SUB_PTY) {
strcat(prline, (_i18n_msg_get(ls_catd,NL_SETN,559, " pseudo-terminal"))); /* catgets 559 */
if (job->submit.options & (SUB_PTY_SHELL))
strcat(prline, (_i18n_msg_get(ls_catd,NL_SETN,560, " shell"))); /* catgets 560 */
}
strcat(prline, (_i18n_msg_get(ls_catd,NL_SETN,561, " mode,"))); /* catgets 561 */
prtLine(prline);
}
if ( job->jobPriority > 0 ) {
sprintf(prline, " %s <%d>,", I18N_Job_Priority, job->jobPriority);
prtLine(prline);
}
if (job->submit.options2 & (SUB2_JOB_CMD_SPOOL)) {
if (tFormat)
sprintf(prline, " %s(Spooled) <%s>", I18N_Command, job->submit.command);
else
sprintf(prline, " %s(Spooled) <%s>\n", I18N_Command, job->submit.command);
} else {
if (tFormat)
sprintf(prline, " %s <%s>", I18N_Command, job->submit.command);
else
sprintf(prline, " %s <%s>\n", I18N_Command, job->submit.command);
}
prtLine(prline);
}