1
1
This file describes changes in recent versions of SLURM. It primarily
2
2
documents those changes that are of interest to users and admins.
4
* Changes in SLURM 2.0.5
5
========================
6
-- BLUEGENE - Added support for emulating systems with a X-dimension of 4.
7
-- BLUEGENE - When a nodecard goes down on a non-Dynamic system SLURM will
8
now only drain blocks under 1 midplane, if no such block exists then SLURM
9
will drain the entire midplane and not mark any block in error state.
10
Previously SLURM would drain every overlapping block of the nodecard
11
making it possible for a large block to make other blocks not work since
12
they overlap some other part of the block that really isn't bad.
13
-- BLUEGENE - Handle L3 errors on boot better.
14
-- Don't revoke a pending batch launch request from the slurmctld if the
15
job is immediately suspended (a normal event with gang scheduling).
16
-- BLUEGENE - Fixed issue with restart of slurmctld would allow error block
17
nodes to be considered for building new blocks when testing if a job would
18
run. This is a visual bug only, jobs would never run on new block, but
19
the block would appear in slurm tools.
20
-- Better responsiveness when starting new allocations when running with the
22
-- Fixed race condition when reconfiguring the slurmctld and using the
23
consumable resources plugin which would cause the controller to core.
24
-- Fixed race condition that sometimes caused jobs to stay in completing
25
state longer than necessary after being terminated.
26
-- Fixed issue where if a parent account has a qos added and then a child
27
account has the qos removed the users still get the qos.
28
-- BLUEGENE - New blocks in dynamic mode will only be made in the system
29
when the block is actually needed for a job, not when testing.
30
-- BLUEGENE - Don't remove larger block used for small block until job starts.
31
-- Add new squeue output format and sort option of "%L" to print a job's time
32
left (time limit minus time used).
33
-- BLUEGENE - Fixed draining state count for sinfo/sview.
34
-- Fix for sview to not core when viewing nodes allocated to a partition
35
and the all jobs finish.
36
-- Fix cons_res to not core dump when finishing a job running on a
38
-- Don't require a node to have --ntasks-per-node CPUs for use when the
39
--overcommit option is also used.
40
-- Increase the maximum number of tasks which can be launched by a job step
41
per node from 64 to 128.
42
-- sview - make right click on popup window title show sorted list.
43
-- scontrol now displays correct units for job min memory and min tmp disk.
44
-- better support for salloc/sbatch arbitrary layout for setting correct
46
-- Env var SLURM_CPUS_ON_NODE is now set correctly depending on the
47
FastSchedule configuration parameter.
48
-- Correction to topology/3d_torus plugin calculation when coordinate value
49
exceeds "9" (i.e. a hex value).
50
-- In sched/wiki2 - Strip single and double quotes out of a node's reason
51
string to avoid confusing Moab's parser.
52
-- Modified scancel to cancel any pending jobs before cancelling any other
53
-- Updated sview config info
54
-- Fix a couple of bugs with respect to scheduling with overlapping
55
reservations (one with a flag of "Maintenance").
56
-- Fix bug when updating a pending job's nice value after explicitly setting
58
-- We no longer add blank QOS'
59
-- Fix task affinity for systems running fastschedule!=0 and they have less
60
resources configured than in existence.
61
-- Slurm.pm loads without warning now on AIX systems
62
-- modified pmi code to do strncpy's on the correct len
63
-- Fix for filling in a qos structure to return SLURM_SUCCESS on success.
64
-- BLUEGENE - Added SLURM_BG_NUM_NODES with cnode count of allocation,
65
SLURM_JOB_NUM_NODES represents midplane counts until 2.1.
66
-- BLUEGENE - Added fix for if a block is in error state and the midplane
67
containning the block is also set to drain/down. This previously
68
prevented dynamic creation of new blocks when this state was present.
69
-- Fixed bug where a users association limits were not enforced, only
70
parent limits were being enforced.
71
-- For OpenMPI use of SLURM reserved ports, reserve a count of ports equal to
72
the maximum task count on any node plus one (the plus one is a correction).
73
-- Do not reset SLURM_TASKS_PER_NODE when srun --preserve-env option is used
75
-- Fix possible assert failure in task/affinity if a node is configured with
76
more resources than physically exist.
77
-- Sview can now resize columns.
78
-- Avoid clearing a drained node's reason field when state is changed from
79
down (i.e. returned to service). Note the drain state flag stays set.
81
* Changes in SLURM 2.0.4
82
========================
83
-- Permit node suspend/resume logic to be enabled through "scontrol reconfig"
84
given appropriate changes to slurm configuration file.
85
-- Check for return codes on functions with warn_unused_result set.
86
-- Fix memory leak in getting step information (as used by squeue -s).
87
-- Better logging for when job's request bad output file locations.
88
-- Fix issue where if user specified non-existant file to write to slurmstepd
89
will regain privileges before sending batch script ended to the controller.
90
-- Fix bug when using the priority_multifactor plugin with no associations
92
-- BLUEGENE - we no longer check for the images to sync state. This was
93
needed long ago when rebooting blocks wasn't a possibility and should
94
had been removed when that functionality was available.
95
-- Added message about no connection with the database for sacctmgr.
96
-- On BlueGene, let srun or salloc exit on SIGINT if slurmctld dies while
98
-- In select/cons_res fix bug that could result in invalid memory pointer
99
if node configurations in slurm.conf contains 8 or more distinct
100
socket/core/thread counts.
101
-- Modify select/cons_res to recognize updated memory size upon node startup
103
-- Fixed bug if not enforcing associations, but running with them and the
104
priority/multifactor, the slurmctld will not core dump on processing usage.
105
-- QOS will not be reset to the default when added back a previously deleted
107
-- Do not set a job's virtual memory limit based upon the job's specified
108
memory limit (which should be a real memory limit, not virtual).
109
-- BLUEGENE - fix for sinfo/sview for displaying proper node count for nodes
111
-- Fix for sview when viewing a certain part of a group (like 1 job) so it
112
doesn't core when the part is gone.
113
-- BLUEGENE - Changed order of SYNC's to be on the front of the list to
114
avoid having a job terminated with a TERM before the SYNC of the
116
-- Validate configured PluginDir value is a valid directory before trying to
118
-- Fix to resolve agent_queue_request symbol from some checkpoint plugins.
119
-- Fix possible execve error for sbatch script read from stdin.
120
-- Modify handling of user ID/name and group ID/name in the slurm.conf file
121
to properly handle user names that contain all digits. Return error code
122
from uid_from_string() and gid_from_string() functions rather than a uid of
123
-1, which might be a valid uid or gid on some systems.
124
-- Fix in re-calcuation of job priorities due to DOWN or DRAINED nodes.
126
* Changes in SLURM 2.0.3
127
========================
128
-- Add reservation creation/update flag of Ignore_Jobs to enable the creation
129
of a reservation that overlaps jobs expected to still be running when
130
the reservation starts. This would be especially useful to reserve all
131
nodes for system maintenence without adjusting time limits of running
132
jobs before creating the reservation. Without this flag, nodes allocated
133
jobs expected to running when the reservation begins can not be placed
135
-- In task/affinity plugin, add layer of abstraction to logic translating
136
block masks to physical machine masks. Patch from Matthieu Hautreux, CEA.
137
-- Fix for setting the node_bitmap in a job to NULL if the job does not
138
start correctly when expected to start.
139
-- Fixed bug in srun --pty logic. Output from the task was split up
140
arbitrarily into stdout and stderr streams, and sometimes was printed
142
-- If job requests minimum and maximum node count range with select/cons_res,
143
try to satisfy the higher value (formerly only allocated the minimum).
144
-- Fix for checking for a non-existant job when querying steps.
145
-- For job steps with the --exclusive option, base initial wait time in
146
partly upon the process ID for better performance with many job steps
147
started at the same time. Maintain exponential back-off as needed.
148
-- Fix for correct step ordering in sview.
149
-- Support optional argument to srun and salloc --immediate option. Specify
150
timeout value in seconds for job or step to be allocated resources.
152
* Changes in SLURM 2.0.2
153
========================
154
-- Fix, don't remove job details when a job is cancelled while pending.
155
-- Do correct type for mktime so garbage isn't returned on 64bit systems
156
for accounting archival.
157
-- Better checking in sacctmgr to avoid infinite loops.
158
-- Fix minor memory leak in fake_slurm_step_layout_create()
159
-- Fix node weight (scheduling priority) calculation for powered down
160
nodes. Patch from Hongjia Cao, NUDT.
161
-- Fix node suspend/resume rate calculations. Patch from Hongjia Cao, NUDT.
162
-- Change calculations using ResumeRate and SuspendRate to provide higher
164
-- Log the IP address for incoming messages having an invalid protocol
166
-- Fix for sacct to show jobs that start the same second as the sacct
168
-- BLUEGENE - Fix for -n option to work on correct cpu counts for each
169
midplane instead of treating -n as a c-node count.
170
-- salloc now sets SLURM_NTASKS_PER_NODE if --ntasks-per-node option is set.
171
-- Fix select/linear to properly set a job's count of allocated processors
172
(all processors on the allocated nodes).
173
-- Fix select/cons_res to allocate proper CPU count when --ntasks-per-node
174
option is used without a task count in the job request.
175
-- Insure that no node is allocated to a job for which the CPU count is less
176
than --ntasks-per-node * --cpus-per-task.
177
-- Correct AllocProcs reported by "scontrol show node" when ThreadsPerCore
178
is greater than 1 and select/cons_res is used.
179
-- Fix scontrol show config for accounting information when values are
180
not set in the slurm.conf.
181
-- Added a set of SBATCH_CPU_BIND* and SBATCH_MEM_BIND* env variables to keep
182
jobsteps launched from within a batch script from inheriting the CPU and
183
memory affinity that was applied to the batch script. Patch from Matthieu
185
-- Ignore the extra processors on a node above configured size if either
186
sched/gang or select/cons_res is configured.
187
-- Fix bug in tracking memory allocated on a node for select/cons_res plugin.
188
-- Fixed a race condition when writing labelled output with a file per task
189
or per node, which potentially closed a file before all data was written.
190
-- BLUEGENE - Fix, for if a job comes in spanning both less than and
191
over 1 midplane in size we check the connection type appropriately.
192
-- Make sched/backfill properly schedule jobs with constraints having node
193
counts. NOTE: Backfill of jobs with constraings having exclusive OR
194
operators are not fully supported.
195
-- If srun is cancelled by SIGINT, set the job state to cancelled, not
197
-- BLUEGENE - Fix, for if you are setting an subbp into an error mode
198
where the subbp stated isn't the first ionode in a nodecard.
199
-- Fix for backfill to not core when checking shared nodes.
200
-- Fix for scontrol to not core when hitting just return in interactive mode.
201
-- Improve sched/backfill logic with respect to shared nodes (multiple jobs
203
-- In sched/wiki (Maui interface) add job info fields QOS, RCLASS, DMEM and
204
TASKSPERNODE. Patch from Bjorn-Helge Mevik, University of Oslo.
206
* Changes in SLURM 2.0.1
207
========================
208
-- Fix, truncate time of start and end for job steps in sacct.
209
-- Initialize all messages to slurmdbd. Previously uninitialized string could
210
cause slurmctld to fail with invalid memory reference.
211
-- BLUEGENE - Fix, for when trying to finish a torus on a block already
212
visited. Even though this may be possible electrically this isn't valid
213
in the under lying infrastructure.
214
-- Fix, in mysql plugins change mediumints to int to support full 32bit
216
-- Add sinfo node state filtering support for NO_RESPOND, POWER_SAVE, FAIL,
217
MAINT, DRAINED and DRAINING states. The state filter of DRAIN still maps
218
to any node in either DRAINED or DRAINING state.
219
-- Fix reservation logic when job requests specific nodes that are already
220
in some reservation the job can not use.
221
-- Fix recomputation of a job's end time when allocated nodes which are
222
being powered up. The end time would be set in the past if the job's
223
time limit was INFINITE, resulting in it being prematurely terminated.
224
-- Permit regular user to change the time limit of his pending jobs up to
225
the partition's limit.
226
-- Fix "-Q" (quiet) option for salloc and sbatch which was previously
228
-- BLUEGENE - fix for finding odd shaped blocks in dynamic mode.
229
-- Fix logic supporting SuspendRate and ResumeRate configuration parameters.
230
Previous logic was changing state of one too many nodes per minute.
231
-- Save new reservation state file on shutdown (even if no changes).
232
-- Fix, when partitions are deleted the sched and select plugins are notified.
233
-- Fix for slurmdbd to create wckeyid's when they don't exist
234
-- Fix linking problem that prevented checkpoint/aix from working.
236
* Changes in SLURM 2.0.0
237
========================
238
-- Fix for bluegene systems to be able to create 32 node blocks with only
239
16 psets defined in dynamic layout mode.
240
-- Improve srun_cr handling of child srun forking. Patch from Hongjia Cao,
242
-- Configuration parameter ResumeDelay replaced by SuspendTimeout and
244
-- BLUEGENE - sview/sinfo now displays correct cnode numbers for drained nodes
245
or blocks in error state.
246
-- Fix some batch job launch bugs when powering up suspended nodes.
247
-- Added option '-T' for sacct to truncate time of start and end and set
248
default of --starttime to Midnight of current day.
250
* Changes in SLURM 2.0.0-rc2
251
============================
252
-- Change fanout logic to start on calling node instead of first node in
254
-- Fix bug so that smap builds properly on Sun Constellation system.
255
-- Filter white-space out from node feature specification.
256
-- Fixed issue with duration not being honored when updating start time in
258
-- Fix bug in sched/wiki and sched/wiki2 plugins for reporting job resource
259
allocation properly when node names are configured out of sort order
260
with more than one numeric suffix (e.g. "tux10-1" is configured after
262
-- Avoid re-use of job_id (if specified at submit time) when the existing
263
job is in completing state (possible race condition with Moab).
264
-- Added SLURM_DISTRIBUTION to env for salloc.
265
-- Add support for "scontrol takeover" command for backup controller to
266
assume control immediately. Patch from Matthieu Hautreux, CEA.
267
-- If srun is unable to communicate with the slurmd tasks are now marked as
268
failed with the controller.
269
-- Fixed issues with requeued jobs not being accounted for correctly in
271
-- Clear node's POWER_SAVE flag if configuration changes to one lacking a
273
-- Extend a job's time limit as appropriate due to delays powering up nodes.
274
-- If sbatch is used to launch a job step within an existing allocation (as
275
used by LSF) and the required node is powered down, print the message
276
"Job step creation temporarily disabled, retrying", sleep, and retry.
277
-- Configuration parameter ResumeDelay added to control how much time must
278
after a node has been suspended before resume it (e.g. powering it back
280
-- Fix CPU binding for batch program. Patch from Matthieu Hautreux, CEA.
281
-- Fix for front end systems non-responding nodes now show up correctly in
284
* Changes in SLURM 2.0.0-rc1
285
============================
286
-- Fix bug in preservation of advanced reservations when slurmctld restarts.
287
-- Updated perlapi to match correctly with slurm.h structures
288
-- Do not install the srun command on BlueGene systems (mpirun must be used to
290
-- Corrections to scheduling logic for topology/tree in configurations where
291
nodes are configured in multiple leaf switches.
292
-- Patch from Matthieu Hautreux for backup mysql deamon support.
293
-- Changed DbdBackup to DbdBackupHost for slurmdbd.conf file
294
-- Add support for spank_strerror() function and improve error handling in
295
general for SPANK plugins.
296
-- Added configuration parameter SrunIOTimeout to optionally ping srun's tasks
297
for better fault tolerance (e.g. killed and restarteed SLURM daemons on
299
-- Add slurmctld and slurmd binding to appropriate communications address
300
based upon NodeAddr, ControllerAddr and BackupAddr configuration
301
parameters. Based upon patch from Matthieu Hautreux, CEA.
302
NOTE: Fails when SlurmDBD is configured with some configurations.
303
NOTE: You must define BIND_SPECIFIC_ADDR to enable this option.
304
-- Avoid using powered down nodes when scheduling work if possible.
305
Fix possible invalid memory reference in power save logic.
307
* Changes in SLURM 1.4.0-pre13
308
==============================
309
-- Added new partition option AllocNodes which controls the hosts from
310
which jobs can be submitted to this partition. From Matthieu Hautreux, CEA.
311
-- Better support the --contiguous option for job allocations.
312
-- Add new scontrol option: show topology (reports contents of topology.conf
313
file via RPC if topology/tree plugin is configured).
314
-- Add advanced reservation display to smap command.
315
-- Replaced remaining references to SLURM_JOBID with SLURM_JOB_ID - except
316
when needed for backwards compatibility.
317
-- Fix logic to properly excise a DOWN node from the allocation of a job
318
with the --no-kill option.
319
-- The MySQL and PgSQL plugins for accounting storage and job completion are
320
now only built if the underlying database libraries exists (previously
321
the plugins were built to produce a fatal error when used).
322
-- BLUEGENE - scontrol show config will now display bluegene.conf information.
324
* Changes in SLURM 1.4.0-pre12
325
==============================
326
-- Added support for hard time limit by associations with added configuration
327
option PriorityUsageResetPeriod. This specifies the interval at which to
328
clear the record of time used. This is currently only available with the
329
priority/multifactor plugin.
330
-- Added SLURM_SUBMIT_DIR to sbatch's output environment variables.
331
-- Backup slurmdbd support implemented.
332
-- Update to checkpoint/xlch logic from Hongjia Cao, NUDT.
333
-- Added configuration parameter AccountingStorageBackupHost.
335
* Changes in SLURM 1.4.0-pre11
336
==============================
337
-- Fix slurm.spec file for RPM build.
339
* Changes in SLURM 1.4.0-pre10
340
==============================
341
-- Critical bug fix in task/affinity when the CoresPerSocket is greater
342
than the ThreadsPerCore (invalid memory reference).
343
-- Add DebugFlag parameter of "Wiki" to log sched/wiki and wiki2
344
communications in greater detail.
345
-- Add "-d <slurmstepd_path>" as an option to the slurmd daemon to
346
specifying a non-stardard slurmstepd file, used for testing purposes.
347
-- Minor cleanup to crypto/munge plugin.
348
- Restrict uid allowed to decode job credentials in crypto/munge
349
- Get slurm user id early in crypto/munge
350
- Remove buggy error code handling in crypto/munge
351
-- Added sprio command - works only with the priority/multifactor plugin
352
-- Add real topology plugin infrastructure (it was initially added
353
directly into slurmctld code). To specify topology information,
354
set TopologyType=topology/tree and add configuration information
355
to a new file called topology.conf. See "man topology.conf" or
356
topology.html web page for details.
357
-- Set "/proc/self/oom_adj" for slurmd and slurmstepd daemons based upon
358
the values of SLURMD_OOM_ADJ and SLURMSTEPD_OOM_ADJ environment
359
variables. This can be used to prevent daemons being killed when
360
a node's memory is exhausted. Based upon patch by Hongjia Cao, NUDT.
361
-- Fix several bugs in task/affinity: cpuset logic was broken and
362
--cpus-per-task option not properly handled.
363
-- Ensure slurmctld adopts SlurmUser GID as well as UID on startup.
365
* Changes in SLURM 1.4.0-pre9
366
=============================
367
-- OpenMPI users only: Add srun logic to automatically recreate and
368
re-launch a job step if the step fails with a reserved port conflict.
369
-- Added TopologyPlugin configuration parameter.
370
-- Added switch topology data structure to slurmctld (for use by select
371
plugin) add load it based upon new slurm.conf parameters: SwitchName,
372
Nodes, Switches and LinkSpeed.
373
-- Modify select/linear and select/cons_res plugins to optimize resource
374
allocation with respect to network topology.
375
-- Added support for new configuration parameter EpilogSlurmctld (executed
376
by slurmctld daemon).
377
-- Added checkpoint/blcr plugin, SLURM now support job checkpoint/restart
378
using BLCR. Patch from Hongjia Cao, NUDT, China.
379
-- Made a variety of new environment variables available to PrologSlurmctld
380
and EpilogSlurmctld. See the "Prolog and Epilog Scripts" section of the
381
slurm.conf man page for details.
382
-- NOTE: Cold-start (without preserving state) required for upgrade from
385
* Changes in SLURM 1.4.0-pre8
386
=============================
387
-- In order to create a new partition using the scontrol command, use
388
the "create" option rather than "update" (which will only operate
389
upon partitions that already exist).
390
-- Added environment variable SLURM_RESTART_COUNT to batch jobs to
391
indicated the count of job restarts made.
392
-- Added sacctmgr command "show config".
393
-- Added the scancel option --nodelist to cancel any jobs running on a
395
-- Add partition-specific DefaultTime (default time limit for jobs,
396
if not specified use MaxTime for the partition. Patch from Par
397
Andersson, National Supercomputer Centre, Sweden.
398
-- Add support for the scontrol command to be able change the Weight
399
associated with nodes. Patch from Krishnakumar Ravi[KK] (HP).
400
-- Add DebugFlag configuration option of "CPU_Bind" for detailed CPU
401
binding information to be logged.
402
-- Fix some significant bugs in task binding logic (possible infinite loops
403
and memory corruption).
404
-- Add new node state flag of NODE_STATE_MAINT indicating the node is in
405
a reservation of type MAINT.
406
-- Modified task/affinity plugin to automatically bind tasks to sockets,
407
cores, or threads as appropriated based upon resource allocation and
408
task count. User can override with srun's --cpu_bind option.
409
-- Fix bug in backfill logic for select/cons_res plugin, resulted in
410
error "cons_res:_rm_job_from_res: node_state mis-count".
411
-- Add logic go bind a batch job to the resources allocated to that job.
412
-- Add configuration parameter MpiParams for (future) OpenMPI port
413
management. Add resv_port_cnt and resv_ports fields to the job step
414
data structures. Add environment variable SLURM_STEP_RESV_PORTS to
415
show what ports are reserved for a job step.
416
-- Add support for SchedulerParameters=interval=<sec> to control the time
417
interval between executions of the backfill scheduler logic.
418
-- Preserve record of last job ID in use even when doing a cold-start unless
419
there is no job state file or there is a change in its format (which only
420
happens when there is a change in SLURM's major or minor version number:
422
-- Added new configuration parameter KillOnBadExit to kill a job step as soon
423
as any task of a job step exits with a non-zero exit code. Patch based
424
on work from Eric Lin, Bull.
425
-- Add spank plugin calls for use by salloc and sbatch command, see
426
"man spank" for details.
427
-- NOTE: Cold-start (without preserving state) required for upgrade from
430
* Changes in SLURM 1.4.0-pre7
431
=============================
432
-- Bug fix for preemption with select/cons_res when there are no idle nodes.
433
-- Bug fix for use of srun options --exclusive and --cpus-per-task together
434
for job step resource allocation (tracking of cpus in use was bad).
435
-- Added the srun option --preserve-env to pass the current values of
436
environment variables SLURM_NNODES and SLURM_NPROCS through to the
437
executable, rather than computing them from commandline parameters.
438
-- For select/cons_res or sched/gang only: Validate a job's resource
439
allocation socket and core count on each allocated node. If the node's
440
configuration has been changed, then abort the job.
441
-- For select/cons_res or sched/gang only: Disable updating a node's
442
processor count if FastSchedule=0. Administrators must set a valid
443
processor count although the memory and disk space configuration can
444
be loaded from the compute node when it starts.
445
-- Add configure option "--disable-iso8601" to disable SLURM use of ISO 8601
446
time format at the time of SLURM build. Default output for all commands
447
is now ISO 8601 (yyyy-mm-ddThh:mm:ss).
448
-- Add support for scontrol to explicity power a node up or down using the
449
configured SuspendProg and ResumeProg programs.
450
-- Fix book select/cons_res logic for tracking the number of allocated
451
CPUs on a node when a partition's Shared value is YES or FORCE.
452
-- Added configure options "--enable-cray-xt" and "--with-apbasil=PATH" for
453
eventual support of Cray-XT systems.
455
* Changes in SLURM 1.4.0-pre6
456
=============================
457
-- Fix job preemption when sched/gang and select/linear are configured with
458
non-sharing partitions.
459
-- In select/cons_res insure that required nodes have available resources.
461
* Changes in SLURM 1.4.0-pre5
462
=============================
463
-- Correction in setting of SLURM_CPU_BIND environment variable.
464
-- Rebuild slurmctld's job select_jobinfo->node_bitmap on restart/reconfigure
465
of the daemon rather than restoring the bitmap since the nodes in a system
466
can change (be added or removed).
467
-- Add configuration option "--with-cpusetdir=PATH" for non-standard
469
-- Get new multi-core data structures working on BlueGene systems.
470
-- Modify PMI_Get_clique_ranks() to return an array of integers rather
471
than a char * to satisfy PMI standard. Correct logic in
472
PMI_Get_clique_size() for when srun --overcommit option is used.
473
-- Fix bug in select/cons_res, allocated a job all of the processors on a
474
node when the --exclusive option is specified as a job submit option.
475
-- Add NUMA cpu_bind support to the task affinity plugin. Binds tasks to
476
a set of CPUs that belong NUMA locality domain with the appropriate
477
--cpu-bind option (ldoms, rank_ldom, map_ldom, and mask_ldom), see
478
"man srun" for more information.
480
* Changes in SLURM 1.4.0-pre4
481
=============================
482
-- For task/affinity, force jobs to use a particular task binding by setting
483
the TaskPluginParam configuration parameter rather than slurmd's
484
SLURM_ENFORCED_CPU_BIND environment variable.
485
-- Enable full preemption of jobs by partition with select/cons_res
486
(cons_res_preempt.patch from Chris Holmes, HP).
487
-- Add configuration parameter DebugFlags to provide detailed logging for
488
specific subsystems (steps and triggers so far).
489
-- srun's --no-kill option is passed to slurmctld so that a job step is
490
killed even if the node where srun executes goes down (unless the
491
--no-kill option is used, previous termination logic would fail if
492
srun was not responding).
493
-- Transfer a job step's core bitmap from the slurmctld to the slurmd
494
within the job step credential.
495
-- Add cpu_bind, cpu_bind_type, mem_bind and mem_bind_type to job allocation
496
request and job_details structure in slurmctld. Add support to --cpu_bind
497
and --mem_bind options from salloc and sbatch commands.
499
* Changes in SLURM 1.4.0-pre3
500
=============================
501
-- Internal changes: CPUs per node changed from 32-bit to 16-bit size.
502
Node count fields changed from 16-bit to 32-bit size in some structures.
503
-- Remove select plugin functions select_p_get_extra_jobinfo(),
504
select_p_step_begin() and select_p_step_fini().
505
-- Remove the following slurmctld job structure fields: num_cpu_groups,
506
cpus_per_node, cpu_count_reps, alloc_lps_cnt, alloc_lps, and used_lps.
507
Use equivalent fields in new "select_job" structure, which is filled
508
in by the select plugins.
509
-- Modify mem_per_task in job step request from 16-bit to 32-bit size.
510
Use new "select_job" structure for the job step's memory management.
511
-- Add core_bitmap_job to slurmctld's job step structure to identify
512
which specific cores are allocated to the step.
513
-- Add new configuration option OverTimeLimit to permit jobs to exceed
514
their (soft) time limit by a configurable amount. Backfill scheduling
515
will be based upon the soft time limit.
516
-- Remove select_g_get_job_cores(). That data is now within the slurmctld's
519
* Changes in SLURM 1.4.0-pre2
520
=============================
521
-- Remove srun's --ctrl-comm-ifhn-addr option (for PMI/MPICH2). It is no
523
-- Modify power save mode so that nodes can be powered off when idle. See
524
https://computing.llnl.gov/linux/slurm/power_save.html or
525
"man slurm.conf" (SuspendProgram and related parameters) for more
527
-- Added configuration parameter PrologSlurmctld, which can be used to boot
528
nodes into a particular state for each job. See "man slurm.conf" for
530
-- Add configuration parameter CompleteTime to control how long to wait for
531
a job's completion before allocating already released resources to pending
532
jobs. This can be used to reduce fragmentation of resources. See
533
"man slurm.conf" for details.
534
-- Make default CryptoType=crypto/munge. OpenSSL is now completely optional.
535
-- Make default AuthType=auth/munge rather than auth/none.
536
-- Change output format of "sinfo -R" from "%35R %N" to "%50R %N".
538
* Changes in SLURM 1.4.0-pre1
539
=============================
540
-- Save/restore a job's task_distribution option on slurmctld retart.
541
NOTE: SLURM must be cold-started on converstion from version 1.3.x.
542
-- Remove task_mem from job step credential (only job_mem is used now).
543
-- Remove --task-mem and --job-mem options from salloc, sbatch and srun
544
(use --mem-per-cpu or --mem instead).
545
-- Remove DefMemPerTask from slurm.conf (use DefMemPerCPU or DefMemPerNode
547
-- Modify slurm_step_launch API call. Move launch host from function argument
548
to element in the data structure slurm_step_launch_params_t, which is
549
used as a function argument.
550
-- Add state_reason_string to job state with optional details about why
552
-- Make "scontrol show node" output match scontrol input for some fields
553
("Cores" changed to "CoresPerSocket", etc.).
554
-- Add support for a new node state "FUTURE" in slurm.conf. These node records
555
are created in SLURM tables for future use without a reboot of the SLURM
556
daemons, but are not reported by any SLURM commands or APIs.
558
* Changes in SLURM 1.3.17
559
=========================
560
-- Fix bug in configure script that can clear user specified LIBS.
4
562
* Changes in SLURM 1.3.16
5
563
=========================
564
-- Fix memory leak in forward logic of tree message passing.
565
-- Fix job exit code recorded for srun job allocation.
566
-- Bluegene - Bug fix for too many parameters being passed to a debug statement
567
-- Bluegene - Bug fix for systems running more than 8 in the X dim running
7
570
* Changes in SLURM 1.3.15
8
571
=========================