Newer
Older
/*****************************************************************************\
* node_scheduler.c - select and allocated nodes to jobs
* Note: there is a global node table (node_record_table_ptr)
*****************************************************************************
* Copyright (C) 2002-2006 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
* Written by Morris Jette <jette1@llnl.gov>
* UCRL-CODE-226842.
*
* This file is part of SLURM, a resource management program.
* For details, see <http://www.llnl.gov/linux/slurm/>.
*
* SLURM is free software; you can redistribute it and/or modify it under
* the terms of the GNU General Public License as published by the Free
* Software Foundation; either version 2 of the License, or (at your option)
* any later version.
*
* In addition, as a special exception, the copyright holders give permission
* to link the code of portions of this program with the OpenSSL library under
* certain conditions as described in each individual source file, and
* distribute linked combinations including the two. You must obey the GNU
* General Public License in all respects for all of the code used other than
* OpenSSL. If you modify file(s) with this exception, you may extend this
* exception to your version of the file(s), but you are not obligated to do
* so. If you do not wish to do so, delete this exception statement from your
* version. If you delete this exception statement from all source files in
* the program, then also delete it here.
*
* SLURM is distributed in the hope that it will be useful, but WITHOUT ANY
* WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
* FOR A PARTICULAR PURPOSE. See the GNU General Public License for more
* details.
*
* You should have received a copy of the GNU General Public License along
* with SLURM; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
\*****************************************************************************/
#ifdef HAVE_CONFIG_H
# include "config.h"
#ifdef HAVE_SYS_SYSLOG_H
# include <sys/syslog.h>
#endif
#include <errno.h>
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>
#include <syslog.h>
#include <unistd.h>
#include <slurm/slurm_errno.h>
#include "src/common/hostlist.h"
#include "src/common/list.h"
#include "src/common/node_select.h"
#include "src/common/xassert.h"
#include "src/common/xmalloc.h"
#include "src/common/xstring.h"
#include "src/slurmctld/agent.h"
#include "src/slurmctld/node_scheduler.h"
#include "src/slurmctld/sched_plugin.h"
#include "src/slurmctld/slurmctld.h"
#define MAX_FEATURES 32 /* max exclusive features "[fs1|fs2]"=2 */
#define MAX_RETRIES 10
struct node_set { /* set of nodes with same configuration */
uint32_t cpus_per_node; /* NOTE: This is the minimum count,
* if FastSchedule==0 then individual
* nodes within the same configuration
* line (in slurm.conf) can actually
* have different CPU counts */
bitstr_t *feature_bits;
bitstr_t *my_bitmap;
static int _add_node_set_info(struct node_set *node_set_ptr,
const int mem_cnt, int cr_enabled,
struct job_record *job);
static int _build_feature_list(struct job_record *job_ptr);
static int _build_node_list(struct job_record *job_ptr,
struct node_set **node_set_pptr,
int *node_set_size);
static void _feature_list_delete(void *x);
static void _filter_nodes_in_set(struct node_set *node_set_ptr,
struct job_details *detail_ptr);
static int _match_feature(char *seek, char *available);
static int _nodes_in_sets(bitstr_t *req_bitmap,
struct node_set * node_set_ptr,
int node_set_size);
#ifndef HAVE_BG
static int _pick_best_load(struct job_record *job_ptr, bitstr_t * bitmap,
uint32_t min_nodes, uint32_t max_nodes,
uint32_t req_nodes, bool test_only);
static int _job_count_bitmap(bitstr_t * bitmap, bitstr_t * jobmap,
int job_cnt);
static int _pick_best_nodes(struct node_set *node_set_ptr,
int node_set_size, bitstr_t ** select_bitmap,
struct job_record *job_ptr,
struct part_record *part_ptr,
uint32_t min_nodes, uint32_t max_nodes,
uint32_t req_nodes);
static void _print_feature_list(uint32_t job_id, List feature_list);
static bitstr_t *_valid_features(struct job_details *detail_ptr,
char *available);
/*
* allocate_nodes - change state of specified nodes to NODE_STATE_ALLOCATED
* IN job_ptr - job being allocated resources
* globals: node_record_count - number of nodes in the system
* node_record_table_ptr - pointer to global node table
* last_node_update - last update time of node table
extern void allocate_nodes(struct job_record *job_ptr)
last_node_update = time(NULL);
for (i = 0; i < node_record_count; i++) {
if (bit_test(job_ptr->node_bitmap, i))
make_node_alloc(&node_record_table_ptr[i], job_ptr);
return;
}
* count_cpus - report how many cpus are associated with the identified nodes
* IN bitmap - map of nodes to tally
* RET cpu count
* globals: node_record_count - number of nodes configured
* node_record_table_ptr - pointer to global node table
int i, sum;
sum = 0;
for (i = 0; i < node_record_count; i++) {
if (bit_test(bitmap, i) != 1)
continue;
if (slurmctld_conf.fast_schedule)
sum += node_record_table_ptr[i].config_ptr->cpus;
else
sum += node_record_table_ptr[i].cpus;
return sum;
}
/*
* deallocate_nodes - for a given job, deallocate its nodes and make

Moe Jette
committed
* their state NODE_STATE_COMPLETING
* IN job_ptr - pointer to terminating job (already in some COMPLETING state)
* IN timeout - true if job exhausted time limit, send REQUEST_KILL_TIMELIMIT
* RPC instead of REQUEST_TERMINATE_JOB
* IN suspended - true if job was already suspended (node's job_run_cnt
* already decremented);
* globals: node_record_count - number of nodes in the system
* node_record_table_ptr - pointer to global node table
*/
extern void deallocate_nodes(struct job_record *job_ptr, bool timeout,
bool suspended)
int i;
kill_job_msg_t *kill_job = NULL;
agent_arg_t *agent_args = NULL;
int down_node_cnt = 0;
uint16_t base_state;
xassert(job_ptr);
xassert(job_ptr->details);
if (select_g_job_fini(job_ptr) != SLURM_SUCCESS)
error("select_g_job_fini(%u): %m", job_ptr->job_id);
agent_args = xmalloc(sizeof(agent_arg_t));

Moe Jette
committed
if (timeout)
agent_args->msg_type = REQUEST_KILL_TIMELIMIT;
else
agent_args->msg_type = REQUEST_TERMINATE_JOB;
agent_args->retry = 0; /* re_kill_job() resends as needed */
kill_job = xmalloc(sizeof(kill_job_msg_t));
last_node_update = time(NULL);
kill_job->job_id = job_ptr->job_id;
kill_job->job_uid = job_ptr->user_id;
kill_job->nodes = xstrdup(job_ptr->nodes);
kill_job->time = time(NULL);
kill_job->select_jobinfo = select_g_copy_jobinfo(
job_ptr->select_jobinfo);
for (i = 0; i < node_record_count; i++) {
struct node_record *node_ptr = &node_record_table_ptr[i];
if (bit_test(job_ptr->node_bitmap, i) == 0)
base_state = node_ptr->node_state & NODE_STATE_BASE;
if (base_state == NODE_STATE_DOWN) {
/* Issue the KILL RPC, but don't verify response */
down_node_cnt++;
bit_clear(job_ptr->node_bitmap, i);
job_ptr->node_cnt--;
make_node_comp(node_ptr, job_ptr, suspended);
#ifdef HAVE_FRONT_END /* Operate only on front-end */
if (agent_args->node_count > 0)
continue;
#endif
hostlist_push(agent_args->hostlist, node_ptr->name);
agent_args->node_count++;
if ((agent_args->node_count - down_node_cnt) == 0) {
job_ptr->job_state &= (~JOB_COMPLETING);
delete_step_records(job_ptr, 1);
slurm_sched_schedule();
}
if (agent_args->node_count == 0) {
error("Job %u allocated no nodes to be killed on",
job_ptr->job_id);
xfree(kill_job->nodes);
select_g_free_jobinfo(&kill_job->select_jobinfo);
xfree(kill_job);
xfree(agent_args);
return;
agent_queue_request(agent_args);
* _match_feature - determine if the desired feature is one of those available
* IN seek - desired feature
* IN available - comma separated list of available features
* RET 1 if found, 0 otherwise
static int _match_feature(char *seek, char *available)
char *tmp_available = NULL, *str_ptr3 = NULL, *str_ptr4 = NULL;
int found;
if (seek == NULL)
return 1; /* nothing to look for */
if (available == NULL)
return SLURM_SUCCESS; /* nothing to find */
tmp_available = xstrdup(available);
found = 0;
str_ptr3 = (char *) strtok_r(tmp_available, ",", &str_ptr4);
while (str_ptr3) {
if (strcmp(seek, str_ptr3) == 0) { /* we have a match */
found = 1;
break;
}
str_ptr3 = (char *) strtok_r(NULL, ",", &str_ptr4);
xfree(tmp_available);
return found;
}
#ifndef HAVE_BG
/*
* _pick_best_load - Given a specification of scheduling requirements,
* identify the nodes which "best" satisfy the request.
* "best" is defined as the least loaded nodes
* IN job_ptr - pointer to job being scheduled
* IN/OUT bitmap - usable nodes are set on input, nodes not required to
* satisfy the request are cleared, other left set
* IN min_nodes - minimum count of nodes
* IN max_nodes - maximum count of nodes (0==don't care)
* IN req_nodes - requested (or desired) count of nodes
* RET zero on success, EINVAL otherwise
* globals: node_record_count - count of nodes configured
* node_record_table_ptr - pointer to global node table
* NOTE: bitmap must be a superset of req_nodes at the time that
* _pick_best_load is called
*/
static int
_pick_best_load(struct job_record *job_ptr, bitstr_t * bitmap,
uint32_t min_nodes, uint32_t max_nodes,
uint32_t req_nodes, bool test_only)
bitstr_t *basemap;
int i, error_code = EINVAL, node_cnt = 0, prev_cnt = 0, set_cnt;
basemap = bit_copy(bitmap);
if (basemap == NULL)
fatal("bit_copy malloc failure");
set_cnt = bit_set_count(bitmap);
if ((set_cnt < min_nodes) ||
((req_nodes > min_nodes) && (set_cnt < req_nodes)))
return error_code; /* not usable */
for (i=0; node_cnt<set_cnt; i++) {
node_cnt = _job_count_bitmap(basemap, bitmap, i);
if ((node_cnt == 0) || (node_cnt == prev_cnt))
continue; /* nothing new to test */
if ((node_cnt < min_nodes) ||
((req_nodes > min_nodes) && (node_cnt < req_nodes)))
continue; /* need more nodes */
error_code = select_g_job_test(job_ptr, bitmap,
req_nodes, test_only);
if (!error_code)
break;
prev_cnt = node_cnt;
FREE_NULL_BITMAP(basemap);
return error_code;
}
/*
* Set the bits in 'jobmap' that correspond to bits in the 'bitmap'
* that are running 'job_cnt' jobs or less, and clear the rest.
static int
_job_count_bitmap(bitstr_t * bitmap, bitstr_t * jobmap, int job_cnt)
int i, count = 0;
bitoff_t size = bit_size(bitmap);
for (i = 0; i < size; i++) {
if (bit_test(bitmap, i) &&
(node_record_table_ptr[i].run_job_cnt <= job_cnt)) {
bit_set(jobmap, i);
count++;
} else {
bit_clear(jobmap, i);
}
return count;
#endif
/*
* Decide if a job can share nodes with other jobs based on the
* following three input parameters:
*
* IN user_flag - may be 0 (do not share nodes), 1 (node sharing allowed),
* or any other number means "don't care"
* IN part_max_share - current partition's node sharing policy
* IN cons_res_flag - 1 if the consumable resources flag is enable, 0 otherwise
*
* RET - 1 if nodes can be shared, 0 if nodes cannot be shared
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
*
*
* The followed table details the node SHARED state for the various scenarios
*
* part= part= part= part=
* cons_res user_request EXCLUS NO YES FORCE
* -------- ------------ ------ ----- ----- -----
* no default/exclus whole whole whole share/O
* no share=yes whole whole share/O share/O
* yes default whole share share/O share/O
* yes exclusive whole whole whole share/O
* yes share=yes whole share share/O share/O
*
* whole = whole node is allocated exclusively to the user
* share = nodes may be shared but the resources are not overcommitted
* share/O = nodes are shared and the resources can be overcommitted
*
* part->max_share:
* &SHARED_FORCE = FORCE
* 0 = EXCLUSIVE
* 1 = NO
* > 1 = YES
*
* job_ptr->details->shared:
* (uint16_t)NO_VAL = default
* 0 = exclusive
* 1 = share=yes
*
* Here are the desired scheduler actions to take:
* IF cons_res enabled, THEN 'shared' ensures that p_i_bitmap is used AND
* _pick_best_load IS NOT called
* IF cons_res NOT enabled, THEN 'shared' ensures that share_bitmap is used AND
* _pick_best_load IS called
*/
static int
_resolve_shared_status(uint16_t user_flag, uint16_t part_max_share,
int cons_res_flag)
{
/* no sharing if part=EXCLUSIVE */
if (part_max_share == 0)
return 0;
/* sharing if part=FORCE */
if (part_max_share & SHARED_FORCE)
return 1;
if (cons_res_flag) {
/* sharing unless user requested exclusive */
if (user_flag == 0)
return 0;
return 1;
} else {
/* no sharing if part=NO */
if (part_max_share == 1)
return 0;
/* share if the user requested it */
if (user_flag == 1)
return 1;
}
return 0;
}
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
/*
* If the job has required feature counts, then accumulate those
* required resources using multiple calls to _pick_best_nodes()
* and adding those selected nodes to the job's required node list.
* Upon completion, return job's requirements to match the values
* which were in effect upon calling this function.
* Input and output are the same as _pick_best_nodes().
*/
static int
_get_req_features(struct node_set *node_set_ptr, int node_set_size,
bitstr_t ** select_bitmap, struct job_record *job_ptr,
struct part_record *part_ptr,
uint32_t min_nodes, uint32_t max_nodes, uint32_t req_nodes)
{
uint32_t saved_min_nodes, saved_job_min_nodes;
bitstr_t *saved_req_node_bitmap = NULL;
uint32_t saved_num_procs, saved_req_nodes;
int tmp_node_set_size;
struct node_set *tmp_node_set_ptr;
int error_code = SLURM_SUCCESS, i;
bitstr_t *feature_bitmap, *accumulate_bitmap = NULL;
/* save job and request state */
saved_min_nodes = min_nodes;
saved_req_nodes = req_nodes;
saved_job_min_nodes = job_ptr->details->min_nodes;
if (job_ptr->details->req_node_bitmap)
saved_req_node_bitmap = bit_copy(job_ptr->details->req_node_bitmap);
job_ptr->details->req_node_bitmap = NULL;
saved_num_procs = job_ptr->num_procs;
job_ptr->num_procs = 1;
tmp_node_set_ptr = xmalloc(sizeof(struct node_set) * node_set_size);
/* Accumulate nodes with required feature counts.
* Ignored if job_ptr->details->req_node_layout is set (by wiki2).
* Selected nodes become part of job's required node list. */
if (job_ptr->details->feature_list &&
(job_ptr->details->req_node_layout == NULL)) {
ListIterator feat_iter;
struct feature_record *feat_ptr;
feat_iter = list_iterator_create(job_ptr->details->feature_list);
while((feat_ptr = (struct feature_record *)
list_next(feat_iter))) {
if (feat_ptr->count == 0)
continue;
tmp_node_set_size = 0;
/* _pick_best_nodes() is destructive of the node_set
* data structure, so we need to copy then purge */
for (i=0; i<node_set_size; i++) {
if (!_match_feature(feat_ptr->name,
node_set_ptr[i].features))
continue;
tmp_node_set_ptr[tmp_node_set_size].cpus_per_node =
node_set_ptr[i].cpus_per_node;
tmp_node_set_ptr[tmp_node_set_size].real_memory =
node_set_ptr[i].real_memory;
tmp_node_set_ptr[tmp_node_set_size].nodes =
node_set_ptr[i].nodes;
tmp_node_set_ptr[tmp_node_set_size].weight =
node_set_ptr[i].weight;
tmp_node_set_ptr[tmp_node_set_size].features =
xstrdup(node_set_ptr[i].features);
tmp_node_set_ptr[tmp_node_set_size].feature_bits =
bit_copy(node_set_ptr[i].feature_bits);
tmp_node_set_ptr[tmp_node_set_size].my_bitmap =
bit_copy(node_set_ptr[i].my_bitmap);
tmp_node_set_size++;
}
feature_bitmap = NULL;
min_nodes = feat_ptr->count;
req_nodes = feat_ptr->count;
job_ptr->details->min_nodes = feat_ptr->count;
job_ptr->num_procs = feat_ptr->count;
error_code = _pick_best_nodes(tmp_node_set_ptr,
tmp_node_set_size, &feature_bitmap,
job_ptr, part_ptr, min_nodes,
max_nodes, req_nodes);
#if 0
{
char *tmp_str = bitmap2node_name(feature_bitmap);
info("job %u needs %u nodes with feature %s, using %s",
job_ptr->job_id, feat_ptr->count,
feat_ptr->name, tmp_str);
xfree(tmp_str);
}
#endif
for (i=0; i<tmp_node_set_size; i++) {
xfree(tmp_node_set_ptr[i].features);
FREE_NULL_BITMAP(tmp_node_set_ptr[i].feature_bits);
FREE_NULL_BITMAP(tmp_node_set_ptr[i].my_bitmap);
}
if (error_code != SLURM_SUCCESS)
break;
if (feature_bitmap) {
if (accumulate_bitmap) {
bit_or(accumulate_bitmap, feature_bitmap);
bit_free(feature_bitmap);
} else
accumulate_bitmap = feature_bitmap;
}
}
list_iterator_destroy(feat_iter);
}
/* restore most of job state and accumulate remaining resources */
min_nodes = saved_min_nodes;
req_nodes = saved_req_nodes;
job_ptr->details->min_nodes = saved_job_min_nodes;
job_ptr->num_procs = saved_num_procs;
if (saved_req_node_bitmap) {
job_ptr->details->req_node_bitmap =
bit_copy(saved_req_node_bitmap);
}
if (accumulate_bitmap) {
if (job_ptr->details->req_node_bitmap) {
bit_or(job_ptr->details->req_node_bitmap,
accumulate_bitmap);
FREE_NULL_BITMAP(accumulate_bitmap);
} else
job_ptr->details->req_node_bitmap = accumulate_bitmap;
}
xfree(tmp_node_set_ptr);
if (error_code == SLURM_SUCCESS) {
error_code = _pick_best_nodes(node_set_ptr, node_set_size,
select_bitmap, job_ptr, part_ptr, min_nodes,
max_nodes, req_nodes);
}
/* restore job's initial required node bitmap */
FREE_NULL_BITMAP(job_ptr->details->req_node_bitmap);
job_ptr->details->req_node_bitmap = saved_req_node_bitmap;
return error_code;
}
* _pick_best_nodes - from a weigh order list of all nodes satisfying a
* job's specifications, select the "best" for use
* IN node_set_ptr - pointer to node specification information
* IN node_set_size - number of entries in records pointed to by node_set_ptr
* OUT select_bitmap - returns bitmap of selected nodes, must FREE_NULL_BITMAP
* IN job_ptr - pointer to job being scheduled
* IN part_ptr - pointer to the partition in which the job is being scheduled
* IN min_nodes - minimum count of nodes required by the job
* IN max_nodes - maximum count of nodes required by the job (0==no limit)
* IN req_nodes - requested (or desired) count of nodes
* RET SLURM_SUCCESS on success,
* ESLURM_NODES_BUSY if request can not be satisfied now,
* ESLURM_REQUESTED_NODE_CONFIG_UNAVAILABLE if request can never
* be satisfied , or
* ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE if the job can not be
* initiated until the parition's configuration changes
* NOTE: the caller must FREE_NULL_BITMAP memory pointed to by select_bitmap
* Notes: The algorithm is
* 1) If required node list is specified, determine implicitly required
* processor and node count
* 2) Determine how many disjoint required "features" are represented
* (e.g. "FS1|FS2|FS3")
* 3) For each feature: find matching node table entries, identify nodes
* that are up and available (idle or shared) and add them to a bit
* 4) If nodes _not_ shared then call select_g_job_test() to select the
* "best" of those based upon topology, else call _pick_best_load()
* to pick the "best" nodes in terms of workload
* 5) If request can't be satisfied now, execute select_g_job_test()
* against the list of nodes that exist in any state (perhaps DOWN
* DRAINED or ALLOCATED) to determine if the request can
* ever be satified.
static int
_pick_best_nodes(struct node_set *node_set_ptr, int node_set_size,
bitstr_t ** select_bitmap, struct job_record *job_ptr,
struct part_record *part_ptr,
uint32_t min_nodes, uint32_t max_nodes, uint32_t req_nodes)
int error_code = SLURM_SUCCESS, i, j, pick_code;
int avail_nodes = 0, avail_cpus = 0;
int avail_mem = 0; /* avail_: resources available for use now */
bitstr_t *avail_bitmap = NULL, *total_bitmap = NULL;
bitstr_t *backup_bitmap = NULL;
bitstr_t *partially_idle_node_bitmap = NULL, *possible_bitmap = NULL;
int max_feature, min_feature;
bool runable_ever = false; /* Job can ever run */
bool runable_avail = false; /* Job can run with available nodes */
uint32_t cr_enabled = 0;
int shared = 0;
select_type_plugin_info_t cr_type = SELECT_TYPE_INFO_NONE;
if (node_set_size == 0) {
info("_pick_best_nodes: empty node set for selection");
return ESLURM_REQUESTED_NODE_CONFIG_UNAVAILABLE;
/* Is Consumable Resources enabled? */
error_code = select_g_get_info_from_plugin (SELECT_CR_PLUGIN,
&cr_enabled);
if (error_code != SLURM_SUCCESS)
return error_code;
shared = _resolve_shared_status(job_ptr->details->shared,
part_ptr->max_share, cr_enabled);
job_ptr->details->shared = shared;
job_ptr->cr_enabled = cr_enabled; /* CR enabled for this job */
cr_type = (select_type_plugin_info_t) slurmctld_conf.
select_type_param;
if ((cr_type == CR_CORE) ||
(cr_type == CR_CPU) || (cr_type == CR_SOCKET)) {
job_ptr->details->job_max_memory = 0;
}
debug3("Job %u shared %d cr_enabled %d CR type %d num_procs %d",
job_ptr->job_id, shared, cr_enabled, cr_type,
if (shared == 0) {
partially_idle_node_bitmap = bit_copy(idle_node_bitmap);
} else {
/* Update partially_idle_node_bitmap to reflect the
* idle and partially idle nodes */
error_code = select_g_get_info_from_plugin (
&partially_idle_node_bitmap);
}
if (error_code != SLURM_SUCCESS) {
FREE_NULL_BITMAP(partially_idle_node_bitmap);
return error_code;
}
}
if (job_ptr->details->req_node_bitmap) { /* specific nodes required */
/* we have already confirmed that all of these nodes have a
* usable configuration and are in the proper partition */
if (min_nodes != 0) {
total_nodes = bit_set_count(
job_ptr->details->req_node_bitmap);
}
if (total_nodes > max_nodes) {
/* exceeds node limit */
if (cr_enabled)
FREE_NULL_BITMAP(partially_idle_node_bitmap);
return ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE;
/* check the availability of these nodes */
/* Should we check memory availability on these nodes? */
if (!bit_super_set(job_ptr->details->req_node_bitmap,
avail_node_bitmap)) {
if (cr_enabled)
FREE_NULL_BITMAP(partially_idle_node_bitmap);
return ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE;
}
if (cr_enabled) {
if (!bit_super_set(job_ptr->details->req_node_bitmap,
partially_idle_node_bitmap)) {
FREE_NULL_BITMAP(partially_idle_node_bitmap);
return ESLURM_NODES_BUSY;
}
}
if (!bit_super_set(job_ptr->details->req_node_bitmap,
FREE_NULL_BITMAP(
partially_idle_node_bitmap);
} else {
if (!bit_super_set(job_ptr->details->req_node_bitmap,
idle_node_bitmap)) {
if (cr_enabled)
FREE_NULL_BITMAP(
partially_idle_node_bitmap);
/* still must go through select_g_job_test() to
* determine validity of request and/or perform
* set-up before job launch */
total_nodes = 0; /* reinitialize */
/* identify the min and max feature values for exclusive OR */
max_feature = -1;
min_feature = MAX_FEATURES;
for (i = 0; i < node_set_size; i++) {
j = bit_ffs(node_set_ptr[i].feature_bits);
if ((j >= 0) && (j < min_feature))
min_feature = j;
j = bit_fls(node_set_ptr[i].feature_bits);
if ((j >= 0) && (j > max_feature))
max_feature = j;
for (j = min_feature; j <= max_feature; j++) {
for (i = 0; i < node_set_size; i++) {
if (!bit_test(node_set_ptr[i].feature_bits, j))
continue;
if (!runable_ever) {
error_code = _add_node_set_info(
&node_set_ptr[i],
&total_bitmap,
job_ptr);
if (error_code != SLURM_SUCCESS) {
if (cr_enabled) {
FREE_NULL_BITMAP(
partially_idle_node_bitmap);
}
FREE_NULL_BITMAP(total_bitmap);
FREE_NULL_BITMAP(possible_bitmap);
return error_code;
bit_and(node_set_ptr[i].my_bitmap, avail_node_bitmap);
if (cr_enabled) {
bit_and(node_set_ptr[i].my_bitmap,
partially_idle_node_bitmap);
}
bit_and(node_set_ptr[i].my_bitmap,
share_node_bitmap);
} else {
bit_and(node_set_ptr[i].my_bitmap,
idle_node_bitmap);
}
node_set_ptr[i].nodes =
bit_set_count(node_set_ptr[i].my_bitmap);
avail_mem = job_ptr->details->job_max_memory;
error_code = _add_node_set_info(&node_set_ptr[i],
&avail_bitmap,
&avail_nodes,
&avail_cpus,
cr_enabled,
job_ptr);
if (error_code != SLURM_SUCCESS) {
if (cr_enabled) {
FREE_NULL_BITMAP(
partially_idle_node_bitmap);
FREE_NULL_BITMAP(total_bitmap);
FREE_NULL_BITMAP(avail_bitmap);
FREE_NULL_BITMAP(possible_bitmap);
return error_code;
#ifndef HAVE_BG
if (shared)
continue; /* Keep accumulating */
#endif
if (avail_nodes == 0)
continue; /* Keep accumulating */
if ((job_ptr->details->req_node_bitmap) &&
(!bit_super_set(job_ptr->details->req_node_bitmap,
avail_bitmap)))
continue;
if ((avail_nodes < min_nodes) ||
((req_nodes > min_nodes) &&
(avail_nodes < req_nodes)))
continue; /* Keep accumulating nodes */
if (avail_cpus < job_ptr->num_procs)
continue; /* Keep accumulating CPUs */
/* NOTE: select_g_job_test() is destructive of
* avail_bitmap, so save a backup copy */
backup_bitmap = bit_copy(avail_bitmap);
pick_code = select_g_job_test(job_ptr,
avail_bitmap,
min_nodes,
max_nodes,
req_nodes,
false);
if (pick_code == SLURM_SUCCESS) {
FREE_NULL_BITMAP(backup_bitmap);
if (bit_set_count(avail_bitmap) > max_nodes) {
/* end of tests for this feature */
avail_nodes = 0;
break;
}
FREE_NULL_BITMAP(total_bitmap);
FREE_NULL_BITMAP(possible_bitmap);
if (cr_enabled) {
FREE_NULL_BITMAP(
partially_idle_node_bitmap);
}
*select_bitmap = avail_bitmap;
} else {
FREE_NULL_BITMAP(avail_bitmap);
avail_bitmap = backup_bitmap;
} /* for (i = 0; i < node_set_size; i++) */
#ifndef HAVE_BG
pick_code = 1;
if (job_ptr->details->req_node_bitmap &&
!bit_super_set(job_ptr->details->req_node_bitmap,
avail_bitmap))
pick_code = 0;
if ((avail_nodes < min_nodes) ||
((req_nodes > min_nodes) && (avail_nodes < req_nodes)))
pick_code = 0;
if (avail_cpus < job_ptr->num_procs)
pick_code = 0;
if (pick_code && cr_enabled) {
/* now that we have all possible resources,
* let's call the select plugin */
backup_bitmap = bit_copy(avail_bitmap);
pick_code = select_g_job_test(job_ptr,
avail_bitmap,
min_nodes,
max_nodes,
req_nodes,
false);
if (pick_code == SLURM_SUCCESS) {
FREE_NULL_BITMAP(backup_bitmap);
FREE_NULL_BITMAP(total_bitmap);
FREE_NULL_BITMAP(possible_bitmap);
FREE_NULL_BITMAP(partially_idle_node_bitmap);
*select_bitmap = avail_bitmap;
return SLURM_SUCCESS;
} else {
FREE_NULL_BITMAP(avail_bitmap);
avail_bitmap = backup_bitmap;
}
} else if (pick_code && shared) {
/* try picking the lightest load from all
available nodes with this feature set */
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
backup_bitmap = bit_copy(avail_bitmap);
pick_code = _pick_best_load(job_ptr,
avail_bitmap,
min_nodes,
max_nodes,
req_nodes,
false);
if (pick_code == SLURM_SUCCESS) {
FREE_NULL_BITMAP(backup_bitmap);
if (bit_set_count(avail_bitmap) > max_nodes) {
avail_nodes = 0;
} else {
FREE_NULL_BITMAP(total_bitmap);
FREE_NULL_BITMAP(possible_bitmap);
if (cr_enabled) {
FREE_NULL_BITMAP(
partially_idle_node_bitmap);
}
*select_bitmap = avail_bitmap;
return SLURM_SUCCESS;
}
} else {
FREE_NULL_BITMAP(avail_bitmap);
avail_bitmap = backup_bitmap;
}
}
#endif
/* try to get req_nodes now for this feature */
if (avail_bitmap
&& (req_nodes > min_nodes)
&& (avail_nodes >= min_nodes)
&& (avail_nodes < req_nodes)
&& ((job_ptr->details->req_node_bitmap == NULL) ||
bit_super_set(job_ptr->details->req_node_bitmap,
avail_bitmap))) {
pick_code = select_g_job_test(job_ptr, avail_bitmap,
if ((pick_code == SLURM_SUCCESS) &&
(bit_set_count(avail_bitmap) <= max_nodes)) {
FREE_NULL_BITMAP(total_bitmap);
FREE_NULL_BITMAP(possible_bitmap);
FREE_NULL_BITMAP(
partially_idle_node_bitmap);
*select_bitmap = avail_bitmap;
return SLURM_SUCCESS;
}
}
/* determine if job could possibly run (if all configured
* nodes available) */
if (total_bitmap
&& (!runable_ever || !runable_avail)
&& (total_nodes >= min_nodes)
&& ((slurmctld_conf.fast_schedule == 0) ||
(total_cpus >= job_ptr->num_procs))
&& ((job_ptr->details->req_node_bitmap == NULL) ||
(bit_super_set(job_ptr->details->req_node_bitmap,
if (!runable_avail) {
FREE_NULL_BITMAP(avail_bitmap);
avail_bitmap = bit_copy(total_bitmap);
if (avail_bitmap == NULL)
fatal("bit_copy malloc failure");
bit_and(avail_bitmap, avail_node_bitmap);
pick_code = select_g_job_test(job_ptr,
avail_bitmap,
min_nodes,
true);
if (cr_enabled)
job_ptr->cr_enabled = 1;
if (pick_code == SLURM_SUCCESS) {
runable_ever = true;
if (bit_set_count(avail_bitmap) <=
max_nodes)
runable_avail = true;
FREE_NULL_BITMAP(possible_bitmap);
possible_bitmap = avail_bitmap;
avail_bitmap = NULL;
if (!runable_ever) {
pick_code = select_g_job_test(job_ptr,
total_bitmap,
min_nodes,
true);
if (cr_enabled)
job_ptr->cr_enabled = 1;
if (pick_code == SLURM_SUCCESS) {
FREE_NULL_BITMAP(possible_bitmap);
possible_bitmap = total_bitmap;
total_bitmap = NULL;
runable_ever = true;
FREE_NULL_BITMAP(avail_bitmap);
FREE_NULL_BITMAP(total_bitmap);
if (error_code != SLURM_SUCCESS)
if (cr_enabled)
FREE_NULL_BITMAP(partially_idle_node_bitmap);
/* The job is not able to start right now, return a
* value indicating when the job can start */
if (!runable_avail)
error_code = ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE;
if (!runable_ever) {
error_code = ESLURM_REQUESTED_NODE_CONFIG_UNAVAILABLE;
info("_pick_best_nodes %u : job never runnable", job_ptr->job_id);
if (error_code == SLURM_SUCCESS) {
error_code = ESLURM_NODES_BUSY;
*select_bitmap = possible_bitmap;
} else {
FREE_NULL_BITMAP(possible_bitmap);
return error_code;
}