Context Navigation

source: trunk/slurm_drmaa/job.c @ 29

Revision 29, 18.9 KB checked in by mmamonski, 12 years ago (diff)
bump version, avoid segault on missing jobs
Property svn:executable set to ``* Property svn:keywords set to `Id`

Rev	Line
[5]	1	/* $Id$ */
[1]	2	/*
	3	* PSNC DRMAA for SLURM
[13]	4	* Copyright (C) 2011 Poznan Supercomputing and Networking Center
[1]	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	#include <string.h>
	20	#include <stdlib.h>
	21	#include <unistd.h>
	22	#include <signal.h>
	23
	24	#include <drmaa_utils/common.h>
	25	#include <drmaa_utils/conf.h>
	26	#include <drmaa_utils/datetime.h>
	27	#include <drmaa_utils/drmaa.h>
	28	#include <drmaa_utils/drmaa_util.h>
	29	#include <drmaa_utils/environ.h>
	30	#include <drmaa_utils/template.h>
	31
	32	#include <slurm_drmaa/job.h>
	33	#include <slurm_drmaa/session.h>
	34	#include <slurm_drmaa/util.h>
	35
	36	#include <slurm/slurm.h>
	37	#include <stdint.h>
	38
	39	static void
	40	slurmdrmaa_job_control( fsd_job_t *self, int action )
	41	{
	42	slurmdrmaa_job_t slurm_self = (slurmdrmaa_job_t)self;
	43	job_desc_msg_t job_desc;
	44
	45	fsd_log_enter(( "({job_id=%s}, action=%d)", self->job_id, action ));
	46
	47	fsd_mutex_lock( &self->session->drm_connection_mutex );
	48	TRY
	49	{
	50	switch( action )
	51	{
	52	case DRMAA_CONTROL_SUSPEND:
	53	if(slurm_suspend(fsd_atoi(self->job_id)) == -1) {
	54	fsd_exc_raise_fmt( FSD_ERRNO_INTERNAL_ERROR,"slurm_suspend error: %s,job_id: %s",slurm_strerror(slurm_get_errno()),self->job_id);
	55	}
[7]	56	slurm_self->user_suspended = true;
[1]	57	break;
	58	case DRMAA_CONTROL_HOLD:
	59	/* change priority to 0*/
	60	slurm_init_job_desc_msg(&job_desc);
	61	slurm_self->old_priority = job_desc.priority;
	62	job_desc.job_id = atoi(self->job_id);
	63	job_desc.priority = 0;
[13]	64	job_desc.alloc_sid = 0;
[1]	65	if(slurm_update_job(&job_desc) == -1) {
	66	fsd_exc_raise_fmt( FSD_ERRNO_INTERNAL_ERROR,"slurm_update_job error: %s,job_id: %s",slurm_strerror(slurm_get_errno()),self->job_id);
	67	}
	68	break;
	69	case DRMAA_CONTROL_RESUME:
	70	if(slurm_resume(fsd_atoi(self->job_id)) == -1) {
	71	fsd_exc_raise_fmt( FSD_ERRNO_INTERNAL_ERROR,"slurm_resume error: %s,job_id: %s",slurm_strerror(slurm_get_errno()),self->job_id);
	72	}
[7]	73	slurm_self->user_suspended = false;
[1]	74	break;
	75	case DRMAA_CONTROL_RELEASE:
	76	/* change priority back*/
	77	slurm_init_job_desc_msg(&job_desc);
	78	job_desc.priority = 1;
	79	job_desc.job_id = atoi(self->job_id);
	80	if(slurm_update_job(&job_desc) == -1) {
	81	fsd_exc_raise_fmt( FSD_ERRNO_INTERNAL_ERROR,"slurm_update_job error: %s,job_id: %s",slurm_strerror(slurm_get_errno()),self->job_id);
	82	}
	83	break;
	84	case DRMAA_CONTROL_TERMINATE:
	85	if(slurm_kill_job(fsd_atoi(self->job_id),SIGKILL,0) == -1) {
	86	fsd_exc_raise_fmt( FSD_ERRNO_INTERNAL_ERROR,"slurm_terminate_job error: %s,job_id: %s",slurm_strerror(slurm_get_errno()),self->job_id);
	87	}
	88	break;
	89	default:
	90	fsd_exc_raise_fmt(
	91	FSD_ERRNO_INVALID_ARGUMENT,
	92	"job::control: unknown action %d", action );
	93	}
	94
	95	fsd_log_debug(("job::control: successful"));
	96	}
	97	FINALLY
	98	{
	99	fsd_mutex_unlock( &self->session->drm_connection_mutex );
	100	}
	101	END_TRY
	102
	103	fsd_log_return(( "" ));
	104	}
	105
	106
	107	static void
	108	slurmdrmaa_job_update_status( fsd_job_t *self )
	109	{
	110	job_info_msg_t *job_info = NULL;
[7]	111	slurmdrmaa_job_t * slurm_self = (slurmdrmaa_job_t *) self;
[1]	112	fsd_log_enter(( "({job_id=%s})", self->job_id ));
	113
	114	fsd_mutex_lock( &self->session->drm_connection_mutex );
	115	TRY
	116	{
	117	if ( slurm_load_job( &job_info, fsd_atoi(self->job_id), SHOW_ALL) ) {
[27]	118	int _slurm_errno = slurm_get_errno();
	119
	120	if (_slurm_errno == ESLURM_INVALID_JOB_ID) {
	121	self->on_missing(self);
	122	} else {
[29]	123	fsd_exc_raise_fmt(FSD_ERRNO_INTERNAL_ERROR,"slurm_load_jobs error: %s,job_id: %s", slurm_strerror(slurm_get_errno()), self->job_id);
[27]	124	}
	125	}
[29]	126	if (job_info) {
[27]	127	fsd_log_debug(("state = %d, state_reason = %d", job_info->job_array[0].job_state, job_info->job_array[0].state_reason));
[29]	128
	129	switch(job_info->job_array[0].job_state & JOB_STATE_BASE)
	130	{
[1]	131
[29]	132	case JOB_PENDING:
	133	switch(job_info->job_array[0].state_reason)
	134	{
	135	#if SLURM_VERSION_NUMBER >= SLURM_VERSION_NUM(2,2,0)
	136	case WAIT_HELD_USER: /* job is held by user */
	137	fsd_log_debug(("interpreting as DRMAA_PS_USER_ON_HOLD"));
	138	self->state = DRMAA_PS_USER_ON_HOLD;
	139	break;
	140	#endif
	141	case WAIT_HELD: /* job is held by administrator */
	142	fsd_log_debug(("interpreting as DRMAA_PS_SYSTEM_ON_HOLD"));
	143	self->state = DRMAA_PS_SYSTEM_ON_HOLD;
	144	break;
	145	default:
	146	fsd_log_debug(("interpreting as DRMAA_PS_QUEUED_ACTIVE"));
	147	self->state = DRMAA_PS_QUEUED_ACTIVE;
	148	}
	149	break;
	150	case JOB_RUNNING:
	151	fsd_log_debug(("interpreting as DRMAA_PS_RUNNING"));
	152	self->state = DRMAA_PS_RUNNING;
	153	break;
	154	case JOB_SUSPENDED:
	155	if(slurm_self->user_suspended == true) {
	156	fsd_log_debug(("interpreting as DRMAA_PS_USER_SUSPENDED"));
	157	self->state = DRMAA_PS_USER_SUSPENDED;
	158	} else {
	159	fsd_log_debug(("interpreting as DRMAA_PS_SYSTEM_SUSPENDED"));
	160	self->state = DRMAA_PS_SYSTEM_SUSPENDED;
	161	}
	162	break;
	163	case JOB_COMPLETE:
	164	fsd_log_debug(("interpreting as DRMAA_PS_DONE"));
	165	self->state = DRMAA_PS_DONE;
	166	self->exit_status = job_info->job_array[0].exit_code;
	167	fsd_log_debug(("exit_status = %d -> %d",self->exit_status, WEXITSTATUS(self->exit_status)));
	168	break;
	169	case JOB_CANCELLED:
	170	fsd_log_debug(("interpreting as DRMAA_PS_FAILED (aborted)"));
	171	self->state = DRMAA_PS_FAILED;
	172	self->exit_status = -1;
	173	case JOB_FAILED:
	174	case JOB_TIMEOUT:
	175	case JOB_NODE_FAIL:
	176	#if SLURM_VERSION_NUMBER >= SLURM_VERSION_NUM(2,3,0)
	177	case JOB_PREEMPTED:
	178	#endif
	179	fsd_log_debug(("interpreting as DRMAA_PS_FAILED"));
	180	self->state = DRMAA_PS_FAILED;
	181	self->exit_status = job_info->job_array[0].exit_code;
	182	fsd_log_debug(("exit_status = %d -> %d",self->exit_status, WEXITSTATUS(self->exit_status)));
	183	break;
	184	default: /unknown state /
	185	fsd_log_error(("Unknown job state: %d. Please send bug report: http://apps.man.poznan.pl/trac/slurm-drmaa", job_info->job_array[0].job_state));
	186	}
[1]	187
[29]	188	if (job_info->job_array[0].job_state & JOB_STATE_FLAGS & JOB_COMPLETING) {
	189	fsd_log_debug(("Epilog completing"));
	190	}
[27]	191
[29]	192	if (job_info->job_array[0].job_state & JOB_STATE_FLAGS & JOB_CONFIGURING) {
	193	fsd_log_debug(("Nodes booting"));
	194	}
[27]	195
[29]	196	if (self->exit_status == -1) /* input,output,error path failure etc*/
	197	self->state = DRMAA_PS_FAILED;
[1]	198
[29]	199	self->last_update_time = time(NULL);
	200
	201	if( self->state >= DRMAA_PS_DONE ) {
	202	fsd_log_debug(("exit_status = %d, WEXITSTATUS(exit_status) = %d", self->exit_status, WEXITSTATUS(self->exit_status)));
	203	fsd_cond_broadcast( &self->status_cond );
	204	}
[27]	205	}
[1]	206	}
	207	FINALLY
	208	{
	209	if(job_info != NULL)
	210	slurm_free_job_info_msg (job_info);
	211
	212	fsd_mutex_unlock( &self->session->drm_connection_mutex );
	213	}
	214	END_TRY
	215
	216	fsd_log_return(( "" ));
	217	}
	218
[27]	219	static void
	220	slurmdrmaa_job_on_missing( fsd_job_t *self )
	221	{
	222
	223	fsd_log_enter(( "({job_id=%s})", self->job_id ));
	224	fsd_log_warning(( "Job %s missing from DRM queue", self->job_id ));
	225
	226	fsd_log_info(( "job_on_missing: last job_ps: %s (0x%02x)", drmaa_job_ps_to_str(self->state), self->state));
	227
	228	if( self->state >= DRMAA_PS_RUNNING ) { /if the job ever entered running state assume finished /
	229	self->state = DRMAA_PS_DONE;
	230	self->exit_status = 0;
	231	}
	232	else {
	233	self->state = DRMAA_PS_FAILED; /* otherwise failed */
	234	self->exit_status = -1;
	235	}
	236
	237	fsd_log_info(("job_on_missing evaluation result: state=%d exit_status=%d", self->state, self->exit_status));
	238
	239	fsd_cond_broadcast( &self->status_cond);
	240	fsd_cond_broadcast( &self->session->wait_condition );
	241
	242	fsd_log_return(( "; job_ps=%s, exit_status=%d", drmaa_job_ps_to_str(self->state), self->exit_status ));
	243	}
	244
[1]	245	fsd_job_t *
	246	slurmdrmaa_job_new( char *job_id )
	247	{
	248	slurmdrmaa_job_t *self = NULL;
	249	self = (slurmdrmaa_job_t*)fsd_job_new( job_id );
	250
	251	fsd_realloc( self, 1, slurmdrmaa_job_t );
	252
	253	self->super.control = slurmdrmaa_job_control;
	254	self->super.update_status = slurmdrmaa_job_update_status;
[27]	255	self->super.on_missing = slurmdrmaa_job_on_missing;
[1]	256	self->old_priority = UINT32_MAX;
[7]	257	self->user_suspended = true;
[1]	258	return (fsd_job_t*)self;
	259	}
	260
	261
	262	void
	263	slurmdrmaa_job_create_req(
	264	fsd_drmaa_session_t *session,
	265	const fsd_template_t *jt,
	266	fsd_environ_t **envp,
	267	job_desc_msg_t * job_desc,
	268	int n_job /* ~job_step */
	269	)
	270	{
	271	fsd_expand_drmaa_ph_t *volatile expand = NULL;
	272
	273	TRY
	274	{
	275	expand = fsd_expand_drmaa_ph_new( NULL, NULL, fsd_asprintf("%d",n_job) );
	276	slurmdrmaa_job_create( session, jt, envp, expand, job_desc, n_job);
	277	}
	278	EXCEPT_DEFAULT
	279	{
	280	fsd_exc_reraise();
	281	}
	282	FINALLY
	283	{
	284	if( expand )
	285	expand->destroy( expand );
	286	}
	287	END_TRY
	288	}
	289
	290	static char *
	291	internal_map_file( fsd_expand_drmaa_ph_t expand, const char path,
	292	bool host_given, const char name )
	293	{
	294	const char *p;
	295
	296	for( p = path; *p != ':'; p++ )
	297	if( *p == '\0' )
	298	fsd_exc_raise_fmt( FSD_DRMAA_ERRNO_INVALID_ATTRIBUTE_FORMAT,
	299	"invalid format of drmaa_%s_path: missing colon", name );
	300	if( host_given )
	301	*host_given = ( p != path );
	302
	303	p++;
	304
	305	return expand->expand( expand, fsd_strdup(p), FSD_DRMAA_PH_HD \| FSD_DRMAA_PH_WD \| FSD_DRMAA_PH_INCR );
	306	}
	307
	308	void
	309	slurmdrmaa_job_create(
	310	fsd_drmaa_session_t *session,
	311	const fsd_template_t *jt,
	312	fsd_environ_t **envp,
	313	fsd_expand_drmaa_ph_t *expand,
	314	job_desc_msg_t * job_desc,
	315	int n_job
	316	)
	317	{
	318	const char *input_path_orig = NULL;
	319	const char *output_path_orig = NULL;
	320	const char *error_path_orig = NULL;
	321	char *volatile input_path = NULL;
	322	char *volatile output_path = NULL;
	323	char *volatile error_path = NULL;
	324	bool input_host = false;
	325	bool output_host = false;
	326	bool error_host = false;
	327	bool join_files = false;
	328	const char *value;
	329	const char const vector;
	330	const char *job_category = "default";
	331
	332	slurmdrmaa_init_job_desc( job_desc );
	333
	334	slurm_init_job_desc_msg( job_desc );
	335
	336	job_desc->user_id = getuid();
	337	job_desc->group_id = getgid();
	338
	339	job_desc->env_size = 0;
	340
	341	/* job name */
	342	value = jt->get_attr( jt, DRMAA_JOB_NAME );
	343	if( value )
	344	{
	345	job_desc->name = fsd_strdup(value);
	346	fsd_log_debug(("# job_name = %s",job_desc->name));
	347	}
	348
	349	/* job state at submit */
	350	value = jt->get_attr( jt, DRMAA_JS_STATE );
	351	if( value )
	352	{
	353	if( 0 == strcmp( value, DRMAA_SUBMISSION_STATE_ACTIVE ) )
	354	{}
	355	else if( 0 == strcmp( value, DRMAA_SUBMISSION_STATE_HOLD ) )
	356	{
	357	job_desc->priority = 0;
	358	fsd_log_debug(("# hold = user"));
	359	}
	360	else
	361	{
	362	fsd_exc_raise_msg(FSD_DRMAA_ERRNO_INVALID_ATTRIBUTE_VALUE, "invalid value of drmaa_js_state attribute" );
	363	}
	364	}
	365
	366	TRY
	367	{
	368	const char *command = NULL;
	369	char *command_expanded = NULL;
	370	char *temp_script_old = NULL;
	371	char *temp_script = "";
	372	const char const i;
	373	int j;
	374
	375	/* remote command */
	376	command = jt->get_attr( jt, DRMAA_REMOTE_COMMAND );
	377	if( command == NULL )
	378	fsd_exc_raise_msg(
	379	FSD_DRMAA_ERRNO_CONFLICTING_ATTRIBUTE_VALUES,
	380	"drmaa_remote_command not set for job template"
	381	);
	382
	383	command_expanded = expand->expand( expand, fsd_strdup(command), FSD_DRMAA_PH_HD \| FSD_DRMAA_PH_WD );
	384
	385	temp_script = fsd_asprintf("#!/bin/bash\n%s",command_expanded);
	386	fsd_free(command_expanded);
	387
	388	/* arguments list */
	389	vector = jt->get_v_attr( jt, DRMAA_V_ARGV );
	390
	391	if( vector )
	392	{
	393	for( i = vector, j = 2; *i; i++, j++ )
	394	{
	395	char arg_expanded = expand->expand( expand, fsd_strdup(i), FSD_DRMAA_PH_HD \| FSD_DRMAA_PH_WD );
	396
	397	temp_script_old = fsd_strdup(temp_script);
	398
	399	if (strcmp(temp_script, "") != 0) {
	400	fsd_free(temp_script);
	401	}
	402	/* add too script */
	403	temp_script = fsd_asprintf("%s '%s'", temp_script_old, arg_expanded);
	404	fsd_free(temp_script_old);
	405	fsd_free(arg_expanded);
	406	}
	407	}
	408
	409	job_desc->script = fsd_asprintf("%s\n", temp_script);
	410	fsd_log_debug(("# Script:\n%s", job_desc->script));
	411	fsd_free(temp_script);
	412	}
	413	END_TRY
	414
	415
	416	/* start time */
	417	value = jt->get_attr( jt, DRMAA_START_TIME );
	418	if( value )
	419	{
	420	job_desc->begin_time = fsd_datetime_parse( value );
	421	fsd_log_debug(( "\n drmaa_start_time: %s -> %ld", value, (long)job_desc->begin_time));
	422	}
	423
[27]	424	/* propagate all environment variables from submission host */
	425	{
[17]	426	extern char **environ;
[27]	427	char **i;
	428	unsigned j = 0;
[17]	429
[27]	430	for ( i = environ; *i; i++) {
	431	job_desc->env_size++;
	432	}
[17]	433
	434	fsd_log_debug(("environ env_size = %d",job_desc->env_size));
	435	fsd_calloc(job_desc->environment, job_desc->env_size+1, char *);
	436
[27]	437	for ( i = environ; *i; i++,j++ ) {
	438	job_desc->environment[j] = fsd_strdup(*i);
	439	}
	440	}
[17]	441
[1]	442	/* environment */
[17]	443
[1]	444	vector = jt->get_v_attr( jt, DRMAA_V_ENV );
	445	if( vector )
	446	{
	447	const char const i;
	448	unsigned j = 0;
[17]	449	unsigned env_offset = job_desc->env_size;
[1]	450
	451	for( i = vector; *i; i++ )
	452	{
	453	job_desc->env_size++;
	454	}
[17]	455	fsd_log_debug(("jt env_size = %d",job_desc->env_size));
[1]	456
	457	fsd_log_debug(("# environment ="));
[17]	458	fsd_realloc(job_desc->environment, job_desc->env_size+1, char *);
[1]	459
	460	for( i = vector; *i; i++,j++ )
	461	{
[17]	462	job_desc->environment[j + env_offset] = fsd_strdup(*i);
	463	fsd_log_debug((" %s", job_desc->environment[j+ env_offset]));
[1]	464	}
	465	}
	466
	467	/* wall clock time hard limit */
	468	value = jt->get_attr( jt, DRMAA_WCT_HLIMIT );
	469	if (value)
	470	{
	471	job_desc->time_limit = slurmdrmaa_datetime_parse( value );
[17]	472	fsd_log_debug(("# wct_hlimit = %s -> %ld",value, (long int)slurmdrmaa_datetime_parse( value )));
[1]	473	}
	474
	475
	476	/expand->set(expand, FSD_DRMAA_PH_INCR,fsd_asprintf("%d", n_job));/ /* set current value */
	477	/* TODO: test drmaa_ph_incr */
	478	/* job working directory */
	479	value = jt->get_attr( jt, DRMAA_WD );
	480	if( value )
	481	{
	482	char *cwd_expanded = expand->expand( expand, fsd_strdup(value), FSD_DRMAA_PH_HD \| FSD_DRMAA_PH_INCR );
	483
	484	expand->set( expand, FSD_DRMAA_PH_WD, fsd_strdup(cwd_expanded));
	485
	486	fsd_log_debug(("# work_dir = %s",cwd_expanded));
	487	job_desc->work_dir = fsd_strdup(cwd_expanded);
	488	fsd_free(cwd_expanded);
	489	}
	490	else
	491	{
	492	char cwdbuf[4096] = "";
	493
	494	if ((getcwd(cwdbuf, 4095)) == NULL) {
	495	char errbuf[256] = "InternalError";
	496	(void)strerror_r(errno, errbuf, 256); /on error the default message would be returned /
	497	fsd_log_error(("getcwd failed: %s", errbuf));
	498	job_desc->work_dir = fsd_strdup(".");
	499	} else {
	500	job_desc->work_dir = fsd_strdup(cwdbuf);
	501	}
	502
	503	fsd_log_debug(("work_dir(default:CWD) %s", job_desc->work_dir));
	504	}
	505
	506	TRY
	507	{
	508	/* input path */
	509	input_path_orig = jt->get_attr( jt, DRMAA_INPUT_PATH );
	510	if( input_path_orig )
	511	{
	512	input_path = internal_map_file( expand, input_path_orig, &input_host,"input" );
	513	fsd_log_debug(( "\n drmaa_input_path: %s -> %s", input_path_orig, input_path ));
	514	}
	515
	516	/* output path */
	517	output_path_orig = jt->get_attr( jt, DRMAA_OUTPUT_PATH );
	518	if( output_path_orig )
	519	{
	520	output_path = internal_map_file( expand, output_path_orig, &output_host,"output" );
	521	fsd_log_debug(( "\n drmaa_output_path: %s -> %s", output_path_orig, output_path ));
	522	}
	523
	524	/* error path */
	525	error_path_orig = jt->get_attr( jt, DRMAA_ERROR_PATH );
	526	if( error_path_orig )
	527	{
	528	error_path = internal_map_file( expand, error_path_orig, &error_host,"error" );
	529	fsd_log_debug(( "\n drmaa_error_path: %s -> %s", error_path_orig, error_path ));
	530	}
	531
	532	/* join files */
	533	value = jt->get_attr( jt, DRMAA_JOIN_FILES );
	534	if( value )
	535	{
	536	if( (value[0] == 'y' \|\| value[0] == 'Y') && value[1] == '\0' )
	537	join_files = true;
	538	else if( (value[0] == 'n' \|\| value[0] == 'N') && value[1] == '\0' )
	539	join_files = false;
	540	else
	541	fsd_exc_raise_msg(
	542	FSD_DRMAA_ERRNO_INVALID_ATTRIBUTE_VALUE,
	543	"invalid value of drmaa_join_files attribute" );
	544	}
	545
	546	if( join_files )
	547	{
	548	if( output_path == NULL )
	549	fsd_exc_raise_msg(FSD_DRMAA_ERRNO_CONFLICTING_ATTRIBUTE_VALUES, "drmaa_join_files is set and output file is not given" );
	550	if( error_path!=NULL && 0 != strcmp( output_path, error_path ) )
	551	fsd_log_warning(( "Error file was given but will be ignored since drmaa_join_files was set." ));
	552
	553	if (error_path)
	554	fsd_free(error_path);
	555
	556	error_path = fsd_strdup(output_path);
	557	}
	558	else
	559	{
	560	if( error_path == NULL && output_path )
	561	error_path = fsd_strdup( "/dev/null" );
	562	if( output_path == NULL && error_path )
	563	output_path = fsd_strdup( "/dev/null" );
	564	}
	565
	566
	567	/* email addresses to send notifications */
	568	vector = jt->get_v_attr( jt, DRMAA_V_EMAIL );
	569	if( vector && vector[0] )
	570	{
	571	/* only to one email address message may be send */
	572	job_desc->mail_user = fsd_strdup(vector[0]);
[25]	573	job_desc->mail_type = MAIL_JOB_BEGIN \| MAIL_JOB_END \| MAIL_JOB_FAIL;
[1]	574	fsd_log_debug(("# mail_user = %s\n",vector[0]));
[25]	575	fsd_log_debug(("# mail_type = %o\n",job_desc->mail_type));
[1]	576	if( vector[1] != NULL )
	577	{
[25]	578	fsd_log_error(( "SLURM only supports one e-mail notification address" ));
	579	fsd_exc_raise_msg(FSD_DRMAA_ERRNO_INVALID_ATTRIBUTE_VALUE,"SLURM only supports one e-mail notification address");
[1]	580	}
	581	}
	582
	583	/* block email */
	584	value = jt->get_attr( jt, DRMAA_BLOCK_EMAIL );
	585	if( value )
	586	{
	587	bool block;
	588	if( strcmp(value, "0") == 0 )
	589	{
	590	block = true;
	591	fsd_log_debug(("# block_email = true"));
	592	fsd_log_debug(("# mail_user delated"));
	593	fsd_free(job_desc->mail_user);
	594	job_desc->mail_user = NULL;
	595	}
	596	else if( strcmp(value, "1") == 0 )
	597	block = false;
	598	else
	599	fsd_exc_raise_msg(FSD_DRMAA_ERRNO_INVALID_ATTRIBUTE_VALUE,"invalid value of drmaa_block_email attribute" );
	600
	601	if( block && output_path == NULL )
	602	{
	603	fsd_log_debug(( "output path not set and we want to block e-mail, set to /dev/null" ));
	604	output_path = fsd_strdup( "/dev/null" );
	605	}
	606	}
	607
	608	if( input_path )
	609	{
	610	job_desc->std_in = fsd_strdup(input_path);
	611	fsd_log_debug(("# input = %s", input_path));
	612	}
	613
	614	if( output_path )
	615	{
	616	job_desc->std_out = fsd_strdup(output_path);
	617	fsd_log_debug(("# output = %s", output_path));
	618	}
	619
	620	if( error_path )
	621	{
	622	job_desc->std_err = fsd_strdup(error_path);
	623	fsd_log_debug(("# error = %s", error_path));
	624	}
	625	}
	626	FINALLY
	627	{
	628	fsd_free( input_path );
	629	fsd_free( output_path );
	630	fsd_free( error_path );
	631	input_path = NULL;
	632	output_path = NULL;
	633	error_path = NULL;
	634	}
	635	END_TRY
	636
	637	/* native specification */
	638	value = jt->get_attr( jt, DRMAA_NATIVE_SPECIFICATION );
	639	if( value )
	640	{
	641	fsd_log_debug(("# Native specification: %s\n", value));
	642	slurmdrmaa_parse_native(job_desc, value);
	643	}
	644
	645	/* job category */
	646	value = jt->get_attr( jt, DRMAA_JOB_CATEGORY );
	647	if( value )
	648	job_category = value;
	649
	650	{
	651	fsd_conf_option_t *category_value = NULL;
	652	category_value = fsd_conf_dict_get( session->job_categories, job_category );
	653
	654	if( category_value != NULL )
	655	{
	656	if( category_value->type != FSD_CONF_STRING )
	657	fsd_exc_raise_fmt(
	658	FSD_ERRNO_INTERNAL_ERROR,
	659	"configuration error: job category should be string"
	660	);
	661
	662	fsd_log_debug(("# Job category %s : %s\n",value,category_value->val.string));
	663	slurmdrmaa_parse_native(job_desc,category_value->val.string);
	664	}
	665	else
	666	{
	667	if( value != NULL )
	668	fsd_exc_raise_fmt(
	669	FSD_DRMAA_ERRNO_INVALID_ATTRIBUTE_VALUE,
	670	"invalid job category: %s", job_category
	671	);
	672	}
	673	}
	674
	675	}
	676
[27]	677

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

Download in other formats: