Context Navigation

source: trunk/pbs_drmaa/job.c @ 15

Revision 15, 13.4 KB checked in by mmamonski, 13 years ago (diff)
other minor fixies
Property svn:keywords set to `Id`

Rev	Line
[12]	1	/* $Id$ */
[1]	2	/*
	3	* FedStage DRMAA for PBS Pro
	4	* Copyright (C) 2006-2009 FedStage Systems
	5	*
	6	* This program is free software: you can redistribute it and/or modify
	7	* it under the terms of the GNU General Public License as published by
	8	* the Free Software Foundation, either version 3 of the License, or
	9	* (at your option) any later version.
	10	*
	11	* This program is distributed in the hope that it will be useful,
	12	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	13	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
	14	* GNU General Public License for more details.
	15	*
	16	* You should have received a copy of the GNU General Public License
	17	* along with this program. If not, see <http://www.gnu.org/licenses/>.
	18	*/
	19
	20	#ifdef HAVE_CONFIG_H
	21	# include <config.h>
	22	#endif
	23
	24	#include <signal.h>
	25	#include <stdlib.h>
	26	#include <string.h>
	27	#include <unistd.h>
	28
	29	#include <drmaa_utils/drmaa.h>
	30	#include <drmaa_utils/drmaa_util.h>
	31	#include <pbs_error.h>
	32	#include <pbs_ifl.h>
	33
	34	#include <pbs_drmaa/job.h>
[7]	35	#include <pbs_drmaa/log_reader.h>
[1]	36	#include <pbs_drmaa/pbs_attrib.h>
	37	#include <pbs_drmaa/session.h>
	38	#include <pbs_drmaa/util.h>
	39
	40	#ifndef lint
	41	static char rcsid[]
	42	# ifdef __GNUC__
	43	__attribute__ ((unused))
	44	# endif
[12]	45	= "$Id$";
[1]	46	#endif
	47
	48
	49	static void
	50	pbsdrmaa_job_control( fsd_job_t *self, int action );
	51
	52	static void
	53	pbsdrmaa_job_update_status( fsd_job_t *self );
	54
	55	static void
	56	pbsdrmaa_job_on_missing( fsd_job_t *self );
	57
[7]	58	void
	59	pbsdrmaa_job_on_missing_standard( fsd_job_t *self );
	60
	61	void
	62	pbsdrmaa_job_on_missing_log_based( fsd_job_t *self );
	63
[1]	64	static void
	65	pbsdrmaa_job_update( fsd_job_t self, struct batch_status );
	66
	67
	68	fsd_job_t *
	69	pbsdrmaa_job_new( char *job_id )
	70	{
	71	pbsdrmaa_job_t self = (pbsdrmaa_job_t)fsd_job_new( job_id );
	72	fsd_realloc( self, 1, pbsdrmaa_job_t );
	73	self->super.control = pbsdrmaa_job_control;
	74	self->super.update_status = pbsdrmaa_job_update_status;
	75	self->super.on_missing = pbsdrmaa_job_on_missing;
	76	self->update = pbsdrmaa_job_update;
	77	return (fsd_job_t*)self;
	78	}
	79
	80
	81	static void
	82	pbsdrmaa_job_control( fsd_job_t *self, int action )
	83	{
	84	volatile bool conn_lock = false;
	85	pbsdrmaa_session_t session = (pbsdrmaa_session_t)self->session;
	86	const char *job_id = self->job_id;
	87	const char *apicall = NULL;
	88	int rc = PBSE_NONE;
	89
	90	fsd_log_enter(( "({job_id=%s}, action=%d)",
	91	self->job_id, action ));
	92
	93	TRY
	94	{
	95	int try_count;
	96	const int max_tries = 3;
	97
	98	conn_lock = fsd_mutex_lock( &self->session->drm_connection_mutex );
	99
	100	/TODO reconnect /
	101	for( try_count=0; try_count < max_tries; try_count++ )
	102	{
	103	switch( action )
	104	{
	105	/*
	106	* We cannot know whether we did suspend job
	107	* in other way than remembering this inside DRMAA session.
	108	*/
	109	case DRMAA_CONTROL_SUSPEND:
	110	apicall = "pbs_sigjob";
	111	rc = pbs_sigjob( session->pbs_conn, (char*)job_id,
	112	"SIGSTOP", NULL );
[10]	113	fsd_log_info(("pbs_sigjob(%s, SIGSTOP) =%d", job_id, rc));
[1]	114	if( rc == PBSE_NONE )
	115	self->flags \|= FSD_JOB_SUSPENDED;
	116	break;
	117	case DRMAA_CONTROL_RESUME:
	118	apicall = "pbs_sigjob";
	119	rc = pbs_sigjob( session->pbs_conn, (char*)job_id,
	120	"SIGCONT", NULL );
[10]	121	fsd_log_info(("pbs_sigjob(%s, SIGCONT) =%d", job_id, rc));
[1]	122	if( rc == PBSE_NONE )
	123	self->flags &= ~FSD_JOB_SUSPENDED;
	124	break;
	125	case DRMAA_CONTROL_HOLD:
	126	apicall = "pbs_holdjob";
	127	rc = pbs_holdjob( session->pbs_conn, (char*)job_id,
	128	USER_HOLD, NULL );
[10]	129	fsd_log_info(("pbs_sigjob(%s, SIGHOLD) =%d", job_id, rc));
[1]	130	if( rc == PBSE_NONE )
	131	self->flags \|= FSD_JOB_HOLD;
	132	break;
	133	case DRMAA_CONTROL_RELEASE:
	134	apicall = "pbs_rlsjob";
	135	rc = pbs_rlsjob( session->pbs_conn, (char*)job_id,
	136	USER_HOLD, NULL );
[10]	137	fsd_log_info(("pbs_rlsjob(%s) =%d", job_id, rc));
[1]	138	if( rc == PBSE_NONE )
	139	self->flags &= FSD_JOB_HOLD;
	140	break;
	141	case DRMAA_CONTROL_TERMINATE:
	142	apicall = "pbs_deljob";
	143	rc = pbs_deljob( session->pbs_conn, (char*)job_id, NULL );
[10]	144	fsd_log_info(("pbs_deljob(%s) =%d", job_id, rc));
[1]	145	/* Torque:
	146	* deldelay=N -- delay between SIGTERM and SIGKILL (default 0) */
	147	if( rc == PBSE_NONE )
	148	{
	149	self->flags &= FSD_JOB_TERMINATED_MASK;
	150	if( (self->flags & FSD_JOB_TERMINATED) == 0 )
	151	self->flags \|= FSD_JOB_TERMINATED \| FSD_JOB_ABORTED;
	152	}
	153	break;
	154	}
	155
	156	if( rc == PBSE_NONE )
	157	break;
	158	else if( rc == PBSE_INTERNAL )
	159	{
	160	/*
	161	* In PBS Pro pbs_sigjob raises internal server error (PBSE_INTERNAL)
	162	* when job just changed its state to running.
	163	*/
	164	fsd_log_debug(( "repeating request (%d of %d)",
	165	try_count+2, max_tries ));
	166	sleep( 1 );
	167	}
	168	else
	169	pbsdrmaa_exc_raise_pbs( apicall );
	170	} /* end for */
	171	}
	172	FINALLY
	173	{
	174	if( conn_lock )
	175	conn_lock = fsd_mutex_unlock( &self->session->drm_connection_mutex );
	176	}
	177	END_TRY
	178
	179	fsd_log_return((""));
	180	}
	181
	182
	183	void
	184	pbsdrmaa_job_update_status( fsd_job_t *self )
	185	{
	186	volatile bool conn_lock = false;
	187	struct batch_status *volatile status = NULL;
	188	pbsdrmaa_session_t session = (pbsdrmaa_session_t)self->session;
	189
	190	fsd_log_enter(( "({job_id=%s})", self->job_id ));
	191	TRY
	192	{
	193	conn_lock = fsd_mutex_lock( &self->session->drm_connection_mutex );
	194	retry:
	195	#ifdef PBS_PROFESSIONAL
	196	status = pbs_statjob( session->pbs_conn, self->job_id, NULL, NULL );
	197	#else
	198	status = pbs_statjob( session->pbs_conn, self->job_id, session->status_attrl, NULL );
	199	#endif
[9]	200	fsd_log_info(( "pbs_statjob(fd=%d, job_id=%s, attribs={...}) =%p",
[1]	201	session->pbs_conn, self->job_id, (void*)status ));
	202	if( status == NULL )
	203	{
	204	#ifndef PBS_PROFESSIONAL
	205	fsd_log_error(("pbs_statjob error: %d, %s, %s", pbs_errno, pbse_to_txt(pbs_errno), pbs_strerror(pbs_errno)));
	206	#else
[15]	207	# ifndef PBS_PROFESSIONAL_NO_LOG
[1]	208	fsd_log_error(("pbs_statjob error: %d, %s", pbs_errno, pbse_to_txt(pbs_errno)));
[15]	209	# else
[13]	210	fsd_log_error(("pbs_statjob error: %d", pbs_errno));
[15]	211	# endif
[1]	212	#endif
	213	switch( pbs_errno )
	214	{
	215	case PBSE_UNKJOBID:
	216	break;
	217	case PBSE_PROTOCOL:
	218	case PBSE_EXPIRED:
[3]	219	if ( session->pbs_conn >= 0 )
	220	pbs_disconnect( session->pbs_conn );
[1]	221	sleep(1);
	222	session->pbs_conn = pbs_connect( session->super.contact );
	223	if( session->pbs_conn < 0 )
	224	pbsdrmaa_exc_raise_pbs( "pbs_connect" );
	225	else
	226	{
	227	fsd_log_error(("retry:"));
	228	goto retry;
	229	}
	230	default:
	231	pbsdrmaa_exc_raise_pbs( "pbs_statjob" );
	232	break;
	233	case 0: /* ? */
	234	fsd_exc_raise_code( FSD_ERRNO_INTERNAL_ERROR );
	235	break;
	236	}
	237	}
	238
	239	conn_lock = fsd_mutex_unlock( &self->session->drm_connection_mutex );
	240
	241	if( status != NULL )
	242	{
	243	((pbsdrmaa_job_t*)self)->update( self, status );
	244	}
[7]	245	else if( self->state < DRMAA_PS_DONE )
[1]	246	self->on_missing( self );
	247	}
	248	FINALLY
	249	{
	250	if( conn_lock )
	251	conn_lock = fsd_mutex_unlock( &self->session->drm_connection_mutex );
	252	if( status != NULL )
	253	pbs_statfree( status );
	254	}
	255	END_TRY
	256
	257	fsd_log_return((""));
	258	}
	259
	260
	261	void
	262	pbsdrmaa_job_update( fsd_job_t self, struct batch_status status )
	263	{
	264	struct attrl *attribs = status->attribs;
	265	struct attrl *i = NULL;
	266	char pbs_state = 0;
	267	int exit_status = -2;
	268	const char *cpu_usage = NULL;
	269	const char *mem_usage = NULL;
	270	const char *vmem_usage = NULL;
	271	const char *walltime = NULL;
	272	long unsigned int modify_time = 0;
	273
	274	fsd_log_enter(( "({job_id=%s})", self->job_id ));
	275	#ifdef DEBUGGING
	276	pbsdrmaa_dump_attrl( attribs, NULL );
	277	#endif
	278	fsd_assert( !strcmp( self->job_id, status->name ) );
	279
	280	for( i = attribs; i != NULL; i = i->next )
	281	{
	282	int attr;
	283	attr = pbsdrmaa_pbs_attrib_by_name( i->name );
	284	switch( attr )
	285	{
	286	case PBSDRMAA_ATTR_JOB_STATE:
	287	pbs_state = i->value[0];
	288	break;
	289	case PBSDRMAA_ATTR_EXIT_STATUS:
	290	exit_status = atoi( i->value );
	291	break;
	292	case PBSDRMAA_ATTR_RESOURCES_USED:
	293	if( !strcmp( i->resource, "cput" ) )
	294	cpu_usage = i->value;
	295	else if( !strcmp( i->resource, "mem" ) )
	296	mem_usage = i->value;
	297	else if( !strcmp( i->resource, "vmem" ) )
	298	vmem_usage = i->value;
	299	else if( !strcmp( i->resource, "walltime" ) )
	300	walltime = i->value;
	301	break;
	302	case PBSDRMAA_ATTR_QUEUE:
	303	if (!self->queue)
	304	self->queue = fsd_strdup(i->value);
	305	break;
	306	case PBSDRMAA_ATTR_ACCOUNT_NAME:
	307	if (!self->project)
	308	self->project = fsd_strdup(i->value);
	309	break;
	310	case PBSDRMAA_ATTR_EXECUTION_HOST:
	311	if (!self->execution_hosts) {
	312	fsd_log_debug(("execution_hosts = %s", i->value));
	313	self->execution_hosts = fsd_strdup(i->value);
	314	}
	315	break;
	316	case PBSDRMAA_ATTR_START_TIME:
	317	{
	318	long unsigned int start_time;
	319	if (self->start_time == 0 && sscanf(i->value, "%lu", &start_time) == 1)
	320	self->start_time = start_time;
	321	break;
	322	}
	323	case PBSDRMAA_ATTR_MTIME:
	324	if (sscanf(i->value, "%lu", &modify_time) != 1)
	325	modify_time = 0;
	326	break;
	327	}
	328	}
	329
	330	if( pbs_state )
	331	fsd_log_debug(( "pbs_state: %c", pbs_state ));
	332
	333	if( exit_status != -2 )
	334	{
	335	fsd_log_debug(( "exit_status: %d", exit_status ));
	336	self->exit_status = exit_status;
	337	}
	338	if(pbs_state){
	339	switch( pbs_state )
	340	{
	341	case 'C': /* Job is completed after having run. */
	342	self->flags &= FSD_JOB_TERMINATED_MASK;
	343	self->flags \|= FSD_JOB_TERMINATED;
	344	if (exit_status != -2) { /* has exit code */
	345	if( self->exit_status == 0)
	346	self->state = DRMAA_PS_DONE;
	347	else
	348	self->state = DRMAA_PS_FAILED;
	349	} else {
	350	self->state = DRMAA_PS_FAILED;
	351	self->exit_status = -1;
	352	}
	353	self->end_time = modify_time; /* take last modify time as end time */
	354	break;
	355	case 'E': /* Job is exiting after having run. - MM: ignore exiting state (transient state) - outputs might have not been transfered yet,
	356	MM2: mark job as running if current job status is undetermined - fix "ps after job was ripped" */
	357	if (self->state == DRMAA_PS_UNDETERMINED)
	358	self->state = DRMAA_PS_RUNNING;
	359	break;
	360	case 'H': /* Job is held. */
	361	self->state = DRMAA_PS_USER_ON_HOLD;
	362	self->flags \|= FSD_JOB_HOLD;
	363	break;
	364	case 'Q': /* Job is queued, eligible to run or routed. */
	365	case 'W': /* Job is waiting for its execution time to be reached. */
	366	self->state = DRMAA_PS_QUEUED_ACTIVE;
	367	self->flags &= ~FSD_JOB_HOLD;
	368	break;
	369	case 'R': /* Job is running. */
	370	case 'T': /* Job is being moved to new location (?). */
	371	{
	372	if( self->flags & FSD_JOB_SUSPENDED )
	373	self->state = DRMAA_PS_USER_SUSPENDED;
	374	else
	375	self->state = DRMAA_PS_RUNNING;
	376	break;
	377	}
	378	case 'S': /* (Unicos only) job is suspend. */
	379	self->state = DRMAA_PS_SYSTEM_SUSPENDED;
	380	break;
	381	case 0: default:
	382	self->state = DRMAA_PS_UNDETERMINED;
	383	break;
	384	}
	385	}
	386	fsd_log_debug(( "job_ps: %s", drmaa_job_ps_to_str(self->state) ));
	387
	388	{
	389	int hours, minutes, seconds;
	390	long mem;
[12]	391	if( cpu_usage && sscanf( cpu_usage, "%d:%d:%d", &hours, &minutes, &seconds ) == 3 )
[1]	392	{
	393	self->cpu_usage = 60( 60hours + minutes ) + seconds;
	394	fsd_log_debug(( "cpu_usage: %s=%lds", cpu_usage, self->cpu_usage ));
	395	}
	396	if( mem_usage && sscanf( mem_usage, "%ldkb", &mem ) == 1 )
	397	{
	398	self->mem_usage = 1024*mem;
	399	fsd_log_debug(( "mem_usage: %s=%ldB", mem_usage, self->mem_usage ));
	400	}
	401	if( vmem_usage && sscanf( vmem_usage, "%ldkb", &mem ) == 1 )
	402	{
	403	self->vmem_usage = 1024*mem;
	404	fsd_log_debug(( "vmem_usage: %s=%ldB", vmem_usage, self->vmem_usage ));
	405	}
[12]	406	if( walltime && sscanf( walltime, "%d:%d:%d", &hours, &minutes, &seconds ) == 3 )
[1]	407	{
	408	self->walltime = 60( 60hours + minutes ) + seconds;
	409	fsd_log_debug(( "walltime: %s=%lds", walltime, self->walltime ));
	410	}
	411	}
	412	}
	413
	414	void
	415	pbsdrmaa_job_on_missing( fsd_job_t *self )
	416	{
[7]	417	pbsdrmaa_session_t pbssession = (pbsdrmaa_session_t)self->session;
[12]	418
[7]	419	if( pbssession->pbs_home == NULL )
	420	pbsdrmaa_job_on_missing_standard( self );
	421	else
	422	pbsdrmaa_job_on_missing_log_based( self );
	423	}
	424
	425	void
	426	pbsdrmaa_job_on_missing_standard( fsd_job_t *self )
	427	{
[1]	428	fsd_drmaa_session_t *session = self->session;
[7]	429
[1]	430	unsigned missing_mask = 0;
	431
	432	fsd_log_enter(( "({job_id=%s})", self->job_id ));
[12]	433	fsd_log_warning(( "Job %s missing from DRM queue", self->job_id ));
[1]	434
	435	switch( session->missing_jobs )
[7]	436	{
[1]	437	case FSD_REVEAL_MISSING_JOBS: missing_mask = 0; break;
	438	case FSD_IGNORE_MISSING_JOBS: missing_mask = 0x73; break;
	439	case FSD_IGNORE_QUEUED_MISSING_JOBS: missing_mask = 0x13; break;
[7]	440	}
[1]	441	fsd_log_debug(( "last job_ps: %s (0x%02x); mask: 0x%02x",
	442	drmaa_job_ps_to_str(self->state), self->state, missing_mask ));
	443
	444	if( self->state < DRMAA_PS_DONE
	445	&& (self->state & ~missing_mask) )
	446	fsd_exc_raise_fmt(
	447	FSD_ERRNO_INTERNAL_ERROR,
	448	"self %s missing from queue", self->job_id
	449	);
	450
	451	if( (self->flags & FSD_JOB_TERMINATED_MASK) == 0 )
[7]	452	{
[1]	453	self->flags &= FSD_JOB_TERMINATED_MASK;
	454	self->flags \|= FSD_JOB_TERMINATED;
[7]	455	}
[1]	456
	457	if( (self->flags & FSD_JOB_ABORTED) == 0
	458	&& session->missing_jobs == FSD_IGNORE_MISSING_JOBS )
[7]	459	{ /* assume everthing was ok */
[1]	460	self->state = DRMAA_PS_DONE;
	461	self->exit_status = 0;
[7]	462	}
[1]	463	else
[7]	464	{ /* job aborted */
[1]	465	self->state = DRMAA_PS_FAILED;
	466	self->exit_status = -1;
[7]	467	}
[1]	468
[12]	469	fsd_cond_broadcast( &self->status_cond);
	470
[1]	471	fsd_log_return(( "; job_ps=%s, exit_status=%d",
	472	drmaa_job_ps_to_str(self->state), self->exit_status ));
[7]	473	}
	474
	475	void
	476	pbsdrmaa_job_on_missing_log_based( fsd_job_t *self )
	477	{
	478	fsd_drmaa_session_t *session = self->session;
	479	pbsdrmaa_log_reader_t *log_reader = NULL;
	480
	481	fsd_log_enter(( "({job_id=%s})", self->job_id ));
[12]	482	fsd_log_info(( "Job %s missing from DRM queue", self->job_id ));
[7]	483
	484	TRY
	485	{
	486	log_reader = pbsdrmaa_log_reader_new( session, self);
	487	log_reader->read_log( log_reader );
[1]	488	}
[7]	489	FINALLY
	490	{
	491	pbsdrmaa_log_reader_destroy( log_reader );
	492	}
	493	END_TRY
	494
	495	fsd_log_return(( "; job_ps=%s, exit_status=%d",
	496	drmaa_job_ps_to_str(self->state), self->exit_status ));
[1]	497	}

Note: See TracBrowser for help on using the repository browser.

Download in other formats: