void
fsd_job_get_termination_status( fsd_job_t *self,
int *status, fsd_iter_t **rusage_out )
{
fsd_iter_t* volatile rusage = NULL;
TRY
{
if( rusage_out )
{
rusage = fsd_iter_new( NULL, 0 );
rusage->append( rusage, fsd_asprintf(
"submission_time=%ld", (long)self->submit_time ) );
if (self->start_time)
rusage->append( rusage, fsd_asprintf(
"start_time=%ld", (long)self->start_time ) );
if (self->end_time)
rusage->append( rusage, fsd_asprintf(
"end_time=%ld", (long)self->end_time ) );
rusage->append( rusage, fsd_asprintf(
"cpu=%ld", self->cpu_usage ) );
rusage->append( rusage, fsd_asprintf(
"mem=%ld", self->mem_usage ) );
rusage->append( rusage, fsd_asprintf(
"vmem=%ld", self->vmem_usage ) );
rusage->append( rusage, fsd_asprintf(
"walltime=%ld", self->walltime ) );
rusage->append( rusage, fsd_asprintf(
"hosts=%s", self->execution_hosts ) );
if (self->queue) {
rusage->append( rusage, fsd_asprintf("queue=%s", self->queue ) );
}
if (self->project) {
rusage->append( rusage, fsd_asprintf("project=%s", self->project ) );
}
}
}
EXCEPT_DEFAULT
{
if( rusage )
rusage->destroy( rusage );
if( rusage_out )
*rusage_out = NULL;
fsd_exc_reraise();
}
ELSE
{
if( status )
*status = self->exit_status;
if( rusage_out )
*rusage_out = rusage;
}
END_TRY
}
void
slurmdrmaa_job_create_req(
fsd_drmaa_session_t *session,
const fsd_template_t *jt,
fsd_environ_t **envp,
job_desc_msg_t * job_desc,
int n_job /* ~job_step */
)
{
fsd_expand_drmaa_ph_t *volatile expand = NULL;
TRY
{
expand = fsd_expand_drmaa_ph_new( NULL, NULL, fsd_asprintf("%d",n_job) );
slurmdrmaa_job_create( session, jt, envp, expand, job_desc, n_job);
}
EXCEPT_DEFAULT
{
fsd_exc_reraise();
}
FINALLY
{
if( expand )
expand->destroy( expand );
}
END_TRY
}
void
slurmdrmaa_job_create(
fsd_drmaa_session_t *session,
const fsd_template_t *jt,
fsd_environ_t **envp,
fsd_expand_drmaa_ph_t *expand,
job_desc_msg_t * job_desc,
int n_job
)
{
const char *input_path_orig = NULL;
const char *output_path_orig = NULL;
const char *error_path_orig = NULL;
char *volatile input_path = NULL;
char *volatile output_path = NULL;
char *volatile error_path = NULL;
bool input_host = false;
bool output_host = false;
bool error_host = false;
bool join_files = false;
const char *value;
const char *const *vector;
const char *job_category = "default";
job_desc->user_id = getuid();
job_desc->group_id = getgid();
job_desc->env_size = 0;
/* job name */
value = jt->get_attr( jt, DRMAA_JOB_NAME );
if( value )
{
job_desc->name = fsd_strdup(value);
fsd_log_debug(("# job_name = %s",job_desc->name));
}
/* job state at submit */
value = jt->get_attr( jt, DRMAA_JS_STATE );
if( value )
{
if( 0 == strcmp( value, DRMAA_SUBMISSION_STATE_ACTIVE ) )
{}
else if( 0 == strcmp( value, DRMAA_SUBMISSION_STATE_HOLD ) )
{
job_desc->priority = 0;
fsd_log_debug(("# hold = user"));
}
else
{
fsd_exc_raise_msg(FSD_DRMAA_ERRNO_INVALID_ATTRIBUTE_VALUE, "invalid value of drmaa_js_state attribute" );
}
}
TRY
{
const char *command = NULL;
char *command_expanded = NULL;
char *temp_script_old = NULL;
char *temp_script = "";
const char *const *i;
int j;
/* remote command */
command = jt->get_attr( jt, DRMAA_REMOTE_COMMAND );
if( command == NULL )
fsd_exc_raise_msg(
FSD_DRMAA_ERRNO_CONFLICTING_ATTRIBUTE_VALUES,
"drmaa_remote_command not set for job template"
);
command_expanded = expand->expand( expand, fsd_strdup(command), FSD_DRMAA_PH_HD | FSD_DRMAA_PH_WD );
temp_script = fsd_asprintf("#!/bin/bash\n%s",command_expanded);
fsd_free(command_expanded);
/* arguments list */
vector = jt->get_v_attr( jt, DRMAA_V_ARGV );
if( vector )
{
for( i = vector, j = 2; *i; i++, j++ )
{
char *arg_expanded = expand->expand( expand, fsd_strdup(*i), FSD_DRMAA_PH_HD | FSD_DRMAA_PH_WD );
temp_script_old = fsd_strdup(temp_script);
if (strcmp(temp_script, "") != 0) {
fsd_free(temp_script);
}
/* add too script */
temp_script = fsd_asprintf("%s '%s'", temp_script_old, arg_expanded);
fsd_free(temp_script_old);
fsd_free(arg_expanded);
}
}
job_desc->script = fsd_asprintf("%s\n", temp_script);
fsd_log_debug(("# Script:\n%s", job_desc->script));
fsd_free(temp_script);
}
END_TRY
/* start time */
value = jt->get_attr( jt, DRMAA_START_TIME );
if( value )
{
job_desc->begin_time = fsd_datetime_parse( value );
fsd_log_debug(( "\n drmaa_start_time: %s -> %ld", value, (long)job_desc->begin_time));
}
/* propagate all environment variables from submission host */
{
extern char **environ;
char **i;
unsigned j = 0;
for ( i = environ; *i; i++) {
job_desc->env_size++;
}
fsd_log_debug(("environ env_size = %d",job_desc->env_size));
fsd_calloc(job_desc->environment, job_desc->env_size+1, char *);
for ( i = environ; *i; i++,j++ ) {
job_desc->environment[j] = fsd_strdup(*i);
}
}
/* environment */
vector = jt->get_v_attr( jt, DRMAA_V_ENV );
if( vector )
{
const char *const *i;
unsigned j = 0;
unsigned env_offset = job_desc->env_size;
for( i = vector; *i; i++ )
{
job_desc->env_size++;
}
fsd_log_debug(("jt env_size = %d",job_desc->env_size));
fsd_log_debug(("# environment ="));
fsd_realloc(job_desc->environment, job_desc->env_size+1, char *);
for( i = vector; *i; i++,j++ )
{
job_desc->environment[j + env_offset] = fsd_strdup(*i);
fsd_log_debug((" %s", job_desc->environment[j+ env_offset]));
}
}
/* wall clock time hard limit */
value = jt->get_attr( jt, DRMAA_WCT_HLIMIT );
if (value)
{
job_desc->time_limit = slurmdrmaa_datetime_parse( value );
fsd_log_debug(("# wct_hlimit = %s -> %ld",value, (long int)slurmdrmaa_datetime_parse( value )));
}
/*expand->set(expand, FSD_DRMAA_PH_INCR,fsd_asprintf("%d", n_job));*/ /* set current value */
/* TODO: test drmaa_ph_incr */
/* job working directory */
value = jt->get_attr( jt, DRMAA_WD );
if( value )
{
char *cwd_expanded = expand->expand( expand, fsd_strdup(value), FSD_DRMAA_PH_HD | FSD_DRMAA_PH_INCR );
expand->set( expand, FSD_DRMAA_PH_WD, fsd_strdup(cwd_expanded));
fsd_log_debug(("# work_dir = %s",cwd_expanded));
job_desc->work_dir = fsd_strdup(cwd_expanded);
fsd_free(cwd_expanded);
}
else
{
char cwdbuf[4096] = "";
if ((getcwd(cwdbuf, 4095)) == NULL) {
char errbuf[256] = "InternalError";
(void)strerror_r(errno, errbuf, 256); /*on error the default message would be returned */
fsd_log_error(("getcwd failed: %s", errbuf));
job_desc->work_dir = fsd_strdup(".");
} else {
job_desc->work_dir = fsd_strdup(cwdbuf);
}
fsd_log_debug(("work_dir(default:CWD) %s", job_desc->work_dir));
}
TRY
{
/* input path */
input_path_orig = jt->get_attr( jt, DRMAA_INPUT_PATH );
if( input_path_orig )
{
input_path = internal_map_file( expand, input_path_orig, &input_host,"input" );
fsd_log_debug(( "\n drmaa_input_path: %s -> %s", input_path_orig, input_path ));
}
/* output path */
output_path_orig = jt->get_attr( jt, DRMAA_OUTPUT_PATH );
if( output_path_orig )
{
output_path = internal_map_file( expand, output_path_orig, &output_host,"output" );
fsd_log_debug(( "\n drmaa_output_path: %s -> %s", output_path_orig, output_path ));
}
/* error path */
error_path_orig = jt->get_attr( jt, DRMAA_ERROR_PATH );
if( error_path_orig )
{
error_path = internal_map_file( expand, error_path_orig, &error_host,"error" );
fsd_log_debug(( "\n drmaa_error_path: %s -> %s", error_path_orig, error_path ));
}
/* join files */
value = jt->get_attr( jt, DRMAA_JOIN_FILES );
if( value )
{
if( (value[0] == 'y' || value[0] == 'Y') && value[1] == '\0' )
join_files = true;
else if( (value[0] == 'n' || value[0] == 'N') && value[1] == '\0' )
join_files = false;
else
fsd_exc_raise_msg(
FSD_DRMAA_ERRNO_INVALID_ATTRIBUTE_VALUE,
"invalid value of drmaa_join_files attribute" );
}
if( join_files )
{
if( output_path == NULL )
fsd_exc_raise_msg(FSD_DRMAA_ERRNO_CONFLICTING_ATTRIBUTE_VALUES, "drmaa_join_files is set and output file is not given" );
if( error_path!=NULL && 0 != strcmp( output_path, error_path ) )
fsd_log_warning(( "Error file was given but will be ignored since drmaa_join_files was set." ));
if (error_path)
fsd_free(error_path);
error_path = fsd_strdup(output_path);
}
else
{
if( error_path == NULL && output_path )
error_path = fsd_strdup( "/dev/null" );
if( output_path == NULL && error_path )
output_path = fsd_strdup( "/dev/null" );
}
/* email addresses to send notifications */
vector = jt->get_v_attr( jt, DRMAA_V_EMAIL );
if( vector && vector[0] )
{
/* only to one email address message may be send */
job_desc->mail_user = fsd_strdup(vector[0]);
job_desc->mail_type = MAIL_JOB_BEGIN | MAIL_JOB_END | MAIL_JOB_FAIL;
fsd_log_debug(("# mail_user = %s\n",vector[0]));
fsd_log_debug(("# mail_type = %o\n",job_desc->mail_type));
if( vector[1] != NULL )
{
fsd_log_error(( "SLURM only supports one e-mail notification address" ));
fsd_exc_raise_msg(FSD_DRMAA_ERRNO_INVALID_ATTRIBUTE_VALUE,"SLURM only supports one e-mail notification address");
}
}
/* block email */
value = jt->get_attr( jt, DRMAA_BLOCK_EMAIL );
if( value )
{
bool block;
if( strcmp(value, "0") == 0 )
{
block = true;
fsd_log_debug(("# block_email = true"));
fsd_log_debug(("# mail_user delated"));
fsd_free(job_desc->mail_user);
job_desc->mail_user = NULL;
}
else if( strcmp(value, "1") == 0 )
block = false;
else
fsd_exc_raise_msg(FSD_DRMAA_ERRNO_INVALID_ATTRIBUTE_VALUE,"invalid value of drmaa_block_email attribute" );
if( block && output_path == NULL )
{
fsd_log_debug(( "output path not set and we want to block e-mail, set to /dev/null" ));
output_path = fsd_strdup( "/dev/null" );
}
}
if( input_path )
{
job_desc->std_in = fsd_strdup(input_path);
fsd_log_debug(("# input = %s", input_path));
}
if( output_path )
{
job_desc->std_out = fsd_strdup(output_path);
fsd_log_debug(("# output = %s", output_path));
}
if( error_path )
{
job_desc->std_err = fsd_strdup(error_path);
fsd_log_debug(("# error = %s", error_path));
}
}
FINALLY
{
fsd_free( input_path );
fsd_free( output_path );
fsd_free( error_path );
input_path = NULL;
output_path = NULL;
error_path = NULL;
}
END_TRY
/* job category */
value = jt->get_attr( jt, DRMAA_JOB_CATEGORY );
if( value )
job_category = value;
{
fsd_conf_option_t *category_value = NULL;
category_value = fsd_conf_dict_get( session->job_categories, job_category );
if( category_value != NULL )
{
if( category_value->type != FSD_CONF_STRING )
fsd_exc_raise_fmt(
FSD_ERRNO_INTERNAL_ERROR,
"configuration error: job category should be string"
);
fsd_log_debug(("# Job category %s : %s\n",value,category_value->val.string));
slurmdrmaa_parse_native(job_desc,category_value->val.string);
}
else
{
if( value != NULL )
fsd_exc_raise_fmt(
FSD_DRMAA_ERRNO_INVALID_ATTRIBUTE_VALUE,
"invalid job category: %s", job_category
);
}
}
/* set defaults for constraints - ref: slurm.h */
fsd_log_debug(("# Setting defaults for tasks and processors" ));
job_desc->num_tasks = 1;
job_desc->min_cpus = 0;
job_desc->cpus_per_task = 0;
job_desc->pn_min_cpus = 0;
/* native specification */
value = jt->get_attr( jt, DRMAA_NATIVE_SPECIFICATION );
if( value )
{
fsd_log_debug(("# Native specification: %s\n", value));
slurmdrmaa_parse_native(job_desc, value);
}
}
fsd_iter_t *
slurmdrmaa_session_run_bulk(
fsd_drmaa_session_t *self,
const fsd_template_t *jt,
int start, int end, int incr )
{
int ret = 0;
unsigned i = 0;
int job_id = 0;
int task_id = 0;
fsd_job_t *volatile job = NULL;
volatile unsigned n_jobs = (end - start) / incr + 1;
char ** volatile job_ids = fsd_calloc( job_ids, n_jobs + 1, char* );
volatile bool connection_lock = false;
fsd_environ_t *volatile env = NULL;
job_desc_msg_t job_desc;
submit_response_msg_t *submit_response = NULL;
slurmdrmaa_init_job_desc( &job_desc );
TRY
{
connection_lock = fsd_mutex_lock( &self->drm_connection_mutex );
slurmdrmaa_job_create_req( self, jt, (fsd_environ_t**)&env , &job_desc, 0 );
/* Create job array spec if more than 1 task */
if(n_jobs > 1)
{
fsd_calloc(job_desc.array_inx, ARRAY_INX_MAXLEN, char*);
ret = snprintf(job_desc.array_inx, ARRAY_INX_MAXLEN, "%d-%d:%d", start, end, incr );
if (ret < 0 || ret >= ARRAY_INX_MAXLEN) {
fsd_exc_raise_fmt(FSD_ERRNO_INTERNAL_ERROR, "snprintf: not enough memory");
}
fsd_log_debug(("array job '%s' prepared", job_desc.array_inx));
}
/* Submit the batch job */
if(slurm_submit_batch_job(&job_desc, &submit_response) != SLURM_SUCCESS){
fsd_exc_raise_fmt(
FSD_ERRNO_INTERNAL_ERROR,"slurm_submit_batch_job: %s",slurm_strerror(slurm_get_errno()));
}
connection_lock = fsd_mutex_unlock( &self->drm_connection_mutex );
/* Watch each job in the array */
for (i = 0; i < n_jobs; ++i) {
job_id = (int) submit_response->job_id;
task_id = start + i*incr;
if (n_jobs > 1) {
/* Array job */
if (!working_cluster_rec)
job_ids[i] = fsd_asprintf("%d_%d", job_id, task_id); /* .0*/
else
job_ids[i] = fsd_asprintf("%d_%d.%s", job_id, task_id, working_cluster_rec->name);
} else {
/* Single job */
if (!working_cluster_rec)
job_ids[i] = fsd_asprintf("%d", job_id); /* .0*/
else
job_ids[i] = fsd_asprintf("%d.%s", job_id, working_cluster_rec->name);
}
fsd_log_debug(("job %s submitted", job_ids[i]));
job = slurmdrmaa_job_new( fsd_strdup(job_ids[i]) );
job->session = self;
job->submit_time = time(NULL);
self->jobs->add( self->jobs, job );
job->release( job );
job = NULL;
}
if (working_cluster_rec)
slurmdb_destroy_cluster_rec(working_cluster_rec);
working_cluster_rec = NULL;
}
