1salloc(1) Slurm Commands salloc(1)
2
6 salloc - Obtain a Slurm job allocation (a set of nodes), execute a com‐
7 mand, and then release the allocation when the command is finished.
8
11 salloc [OPTIONS(0)...] [ : [OPTIONS(N)...]] [command(0) [args(0)...]]
12 Option(s) define multiple jobs in a co-scheduled heterogeneous job.
14 For more details about heterogeneous jobs see the document
15 https://slurm.schedmd.com/heterogeneous_jobs.html
16
19 salloc is used to allocate a Slurm job allocation, which is a set of
20 resources (nodes), possibly with some set of constraints (e.g. number
21 of processors per node). When salloc successfully obtains the
22 requested allocation, it then runs the command specified by the user.
23 Finally, when the user specified command is complete, salloc relin‐
24 quishes the job allocation.
25 The command may be any program the user wishes. Some typical commands
27 are xterm, a shell script containing srun commands, and srun (see the
28 EXAMPLES section). If no command is specified, then salloc runs the
29 user's default shell.
30 The following document describes the influence of various options on
32 the allocation of cpus to jobs and tasks.
33 https://slurm.schedmd.com/cpu_management.html
34 NOTE: The salloc logic includes support to save and restore the termi‐
36 nal line settings and is designed to be executed in the foreground. If
37 you need to execute salloc in the background, set its standard input to
38 some file, for example: "salloc -n16 a.out </dev/null &"
39
42 If salloc is unable to execute the user command, it will return 1 and
43 print errors to stderr. Else if success or if killed by signals HUP,
44 INT, KILL, or QUIT: it will return 0.
45
48 If provided, the command is resolved in the following order:
49 1. If command starts with ".", then path is constructed as: current
51 working directory / command
52 2. If command starts with a "/", then path is considered absolute.
53 3. If command can be resolved through PATH. See path_resolution(7).
54 4. If command is in current working directory.
55 Current working directory is the calling process working directory
57 unless the --chdir argument is passed, which will override the current
58 working directory.
59
62 -A, --account=<account>
63 Charge resources used by this job to specified account. The
64 account is an arbitrary string. The account name may be changed
65 after job submission using the scontrol command.
66 --acctg-freq
69 Define the job accounting and profiling sampling intervals.
70 This can be used to override the JobAcctGatherFrequency parame‐
71 ter in Slurm's configuration file, slurm.conf. The supported
72 format is as follows:
73 --acctg-freq=<datatype>=<interval>
75 where <datatype>=<interval> specifies the task sam‐
76 pling interval for the jobacct_gather plugin or a
77 sampling interval for a profiling type by the
78 acct_gather_profile plugin. Multiple, comma-sepa‐
79 rated <datatype>=<interval> intervals may be speci‐
80 fied. Supported datatypes are as follows:
81 task=<interval>
83 where <interval> is the task sampling inter‐
84 val in seconds for the jobacct_gather plugins
85 and for task profiling by the
86 acct_gather_profile plugin. NOTE: This fre‐
87 quency is used to monitor memory usage. If
88 memory limits are enforced the highest fre‐
89 quency a user can request is what is config‐
90 ured in the slurm.conf file. They can not
91 turn it off (=0) either.
92 energy=<interval>
94 where <interval> is the sampling interval in
95 seconds for energy profiling using the
96 acct_gather_energy plugin
97 network=<interval>
99 where <interval> is the sampling interval in
100 seconds for infiniband profiling using the
101 acct_gather_interconnect plugin.
102 filesystem=<interval>
104 where <interval> is the sampling interval in
105 seconds for filesystem profiling using the
106 acct_gather_filesystem plugin.
107 The default value for the task sampling
109 interval
110 is 30. The default value for all other intervals is 0. An
111 interval of 0 disables sampling of the specified type. If the
112 task sampling interval is 0, accounting information is collected
113 only at job termination (reducing Slurm interference with the
114 job).
115 Smaller (non-zero) values have a greater impact upon job perfor‐
116 mance, but a value of 30 seconds is not likely to be noticeable
117 for applications having less than 10,000 tasks.
118 -B --extra-node-info=<sockets[:cores[:threads]]>
121 Restrict node selection to nodes with at least the specified
122 number of sockets, cores per socket and/or threads per core.
123 NOTE: These options do not specify the resource allocation size.
124 Each value specified is considered a minimum. An asterisk (*)
125 can be used as a placeholder indicating that all available
126 resources of that type are to be utilized. Values can also be
127 specified as min-max. The individual levels can also be speci‐
128 fied in separate options if desired:
129 --sockets-per-node=<sockets>
130 --cores-per-socket=<cores>
131 --threads-per-core=<threads>
132 If task/affinity plugin is enabled, then specifying an alloca‐
133 tion in this manner also results in subsequently launched tasks
134 being bound to threads if the -B option specifies a thread
135 count, otherwise an option of cores if a core count is speci‐
136 fied, otherwise an option of sockets. If SelectType is config‐
137 ured to select/cons_res, it must have a parameter of CR_Core,
138 CR_Core_Memory, CR_Socket, or CR_Socket_Memory for this option
139 to be honored. If not specified, the scontrol show job will
140 display 'ReqS:C:T=*:*:*'. This option applies to job alloca‐
141 tions. NOTE: This option is mutually exclusive with --hint,
142 --threads-per-core and --ntasks-per-core.
143 --bb=<spec>
146 Burst buffer specification. The form of the specification is
147 system dependent. Note the burst buffer may not be accessible
148 from a login node, but require that salloc spawn a shell on one
149 of its allocated compute nodes.
150 --bbf=<file_name>
153 Path of file containing burst buffer specification. The form of
154 the specification is system dependent. Also see --bb. Note the
155 burst buffer may not be accessible from a login node, but
156 require that salloc spawn a shell on one of its allocated com‐
157 pute nodes.
158 --begin=<time>
161 Defer eligibility of this job allocation until the specified
162 time.
163 Time may be of the form HH:MM:SS to run a job at a specific time
165 of day (seconds are optional). (If that time is already past,
166 the next day is assumed.) You may also specify midnight, noon,
167 fika (3 PM) or teatime (4 PM) and you can have a time-of-day
168 suffixed with AM or PM for running in the morning or the
169 evening. You can also say what day the job will be run, by
170 specifying a date of the form MMDDYY or MM/DD/YY YYYY-MM-DD.
171 Combine date and time using the following format
172 YYYY-MM-DD[THH:MM[:SS]]. You can also give times like now +
173 count time-units, where the time-units can be seconds (default),
174 minutes, hours, days, or weeks and you can tell Slurm to run the
175 job today with the keyword today and to run the job tomorrow
176 with the keyword tomorrow. The value may be changed after job
177 submission using the scontrol command. For example:
178 --begin=16:00
179 --begin=now+1hour
180 --begin=now+60 (seconds by default)
181 --begin=2010-01-20T12:34:00
182 Notes on date/time specifications:
185 - Although the 'seconds' field of the HH:MM:SS time specifica‐
186 tion is allowed by the code, note that the poll time of the
187 Slurm scheduler is not precise enough to guarantee dispatch of
188 the job on the exact second. The job will be eligible to start
189 on the next poll following the specified time. The exact poll
190 interval depends on the Slurm scheduler (e.g., 60 seconds with
191 the default sched/builtin).
192 - If no time (HH:MM:SS) is specified, the default is
193 (00:00:00).
194 - If a date is specified without a year (e.g., MM/DD) then the
195 current year is assumed, unless the combination of MM/DD and
196 HH:MM:SS has already passed for that year, in which case the
197 next year is used.
198 --bell Force salloc to ring the terminal bell when the job allocation
201 is granted (and only if stdout is a tty). By default, salloc
202 only rings the bell if the allocation is pending for more than
203 ten seconds (and only if stdout is a tty). Also see the option
204 --no-bell.
205 --cluster-constraint=<list>
208 Specifies features that a federated cluster must have to have a
209 sibling job submitted to it. Slurm will attempt to submit a sib‐
210 ling job to a cluster if it has at least one of the specified
211 features.
212 --comment=<string>
215 An arbitrary comment.
216 -C, --constraint=<list>
219 Nodes can have features assigned to them by the Slurm adminis‐
220 trator. Users can specify which of these features are required
221 by their job using the constraint option. Only nodes having
222 features matching the job constraints will be used to satisfy
223 the request. Multiple constraints may be specified with AND,
224 OR, matching OR, resource counts, etc. (some operators are not
225 supported on all system types). Supported constraint options
226 include:
227 Single Name
229 Only nodes which have the specified feature will be used.
230 For example, --constraint="intel"
231 Node Count
233 A request can specify the number of nodes needed with
234 some feature by appending an asterisk and count after the
235 feature name. For example, --nodes=16 --con‐
236 straint="graphics*4 ..." indicates that the job requires
237 16 nodes and that at least four of those nodes must have
238 the feature "graphics."
239 AND If only nodes with all of specified features will be
241 used. The ampersand is used for an AND operator. For
242 example, --constraint="intel&gpu"
243 OR If only nodes with at least one of specified features
245 will be used. The vertical bar is used for an OR opera‐
246 tor. For example, --constraint="intel|amd"
247 Matching OR
249 If only one of a set of possible options should be used
250 for all allocated nodes, then use the OR operator and
251 enclose the options within square brackets. For example,
252 --constraint="[rack1|rack2|rack3|rack4]" might be used to
253 specify that all nodes must be allocated on a single rack
254 of the cluster, but any of those four racks can be used.
255 Multiple Counts
257 Specific counts of multiple resources may be specified by
258 using the AND operator and enclosing the options within
259 square brackets. For example, --con‐
260 straint="[rack1*2&rack2*4]" might be used to specify that
261 two nodes must be allocated from nodes with the feature
262 of "rack1" and four nodes must be allocated from nodes
263 with the feature "rack2".
264 NOTE: This construct does not support multiple Intel KNL
266 NUMA or MCDRAM modes. For example, while --con‐
267 straint="[(knl&quad)*2&(knl&hemi)*4]" is not supported,
268 --constraint="[haswell*2&(knl&hemi)*4]" is supported.
269 Specification of multiple KNL modes requires the use of a
270 heterogeneous job.
271 Brackets
273 Brackets can be used to indicate that you are looking for
274 a set of nodes with the different requirements contained
275 within the brackets. For example, --con‐
276 straint="[(rack1|rack2)*1&(rack3)*2]" will get you one
277 node with either the "rack1" or "rack2" features and two
278 nodes with the "rack3" feature. The same request without
279 the brackets will try to find a single node that meets
280 those requirements.
281 Parenthesis
283 Parenthesis can be used to group like node features
284 together. For example, --con‐
285 straint="[(knl&snc4&flat)*4&haswell*1]" might be used to
286 specify that four nodes with the features "knl", "snc4"
287 and "flat" plus one node with the feature "haswell" are
288 required. All options within parenthesis should be
289 grouped with AND (e.g. "&") operands.
290 --contiguous
293 If set, then the allocated nodes must form a contiguous set.
294 NOTE: If SelectPlugin=cons_res this option won't be honored with
296 the topology/tree or topology/3d_torus plugins, both of which
297 can modify the node ordering.
298 --cores-per-socket=<cores>
301 Restrict node selection to nodes with at least the specified
302 number of cores per socket. See additional information under -B
303 option above when task/affinity plugin is enabled.
304 --cpu-freq =<p1[-p2[:p3]]>
307 Request that job steps initiated by srun commands inside this
309 allocation be run at some requested frequency if possible, on
310 the CPUs selected for the step on the compute node(s).
311 p1 can be [#### | low | medium | high | highm1] which will set
313 the frequency scaling_speed to the corresponding value, and set
314 the frequency scaling_governor to UserSpace. See below for defi‐
315 nition of the values.
316 p1 can be [Conservative | OnDemand | Performance | PowerSave]
318 which will set the scaling_governor to the corresponding value.
319 The governor has to be in the list set by the slurm.conf option
320 CpuFreqGovernors.
321 When p2 is present, p1 will be the minimum scaling frequency and
323 p2 will be the maximum scaling frequency.
324 p2 can be [#### | medium | high | highm1] p2 must be greater
326 than p1.
327 p3 can be [Conservative | OnDemand | Performance | PowerSave |
329 UserSpace] which will set the governor to the corresponding
330 value.
331 If p3 is UserSpace, the frequency scaling_speed will be set by a
333 power or energy aware scheduling strategy to a value between p1
334 and p2 that lets the job run within the site's power goal. The
335 job may be delayed if p1 is higher than a frequency that allows
336 the job to run within the goal.
337 If the current frequency is < min, it will be set to min. Like‐
339 wise, if the current frequency is > max, it will be set to max.
340 Acceptable values at present include:
342 #### frequency in kilohertz
344 Low the lowest available frequency
346 High the highest available frequency
348 HighM1 (high minus one) will select the next highest
350 available frequency
351 Medium attempts to set a frequency in the middle of the
353 available range
354 Conservative attempts to use the Conservative CPU governor
356 OnDemand attempts to use the OnDemand CPU governor (the
358 default value)
359 Performance attempts to use the Performance CPU governor
361 PowerSave attempts to use the PowerSave CPU governor
363 UserSpace attempts to use the UserSpace CPU governor
365 The following informational environment variable is set
368 in the job
369 step when --cpu-freq option is requested.
370 SLURM_CPU_FREQ_REQ
371 This environment variable can also be used to supply the value
373 for the CPU frequency request if it is set when the 'srun' com‐
374 mand is issued. The --cpu-freq on the command line will over‐
375 ride the environment variable value. The form on the environ‐
376 ment variable is the same as the command line. See the ENVIRON‐
377 MENT VARIABLES section for a description of the
378 SLURM_CPU_FREQ_REQ variable.
379 NOTE: This parameter is treated as a request, not a requirement.
381 If the job step's node does not support setting the CPU fre‐
382 quency, or the requested value is outside the bounds of the
383 legal frequencies, an error is logged, but the job step is
384 allowed to continue.
385 NOTE: Setting the frequency for just the CPUs of the job step
387 implies that the tasks are confined to those CPUs. If task con‐
388 finement (i.e., TaskPlugin=task/affinity or TaskPlu‐
389 gin=task/cgroup with the "ConstrainCores" option) is not config‐
390 ured, this parameter is ignored.
391 NOTE: When the step completes, the frequency and governor of
393 each selected CPU is reset to the previous values.
394 NOTE: When submitting jobs with the --cpu-freq option with lin‐
396 uxproc as the ProctrackType can cause jobs to run too quickly
397 before Accounting is able to poll for job information. As a
398 result not all of accounting information will be present.
399 --cpus-per-gpu=<ncpus>
402 Advise Slurm that ensuing job steps will require ncpus proces‐
403 sors per allocated GPU. Not compatible with the --cpus-per-task
404 option.
405 -c, --cpus-per-task=<ncpus>
408 Advise Slurm that ensuing job steps will require ncpus proces‐
409 sors per task. By default Slurm will allocate one processor per
410 task.
411 For instance, consider an application that has 4 tasks, each
413 requiring 3 processors. If our cluster is comprised of
414 quad-processors nodes and we simply ask for 12 processors, the
415 controller might give us only 3 nodes. However, by using the
416 --cpus-per-task=3 options, the controller knows that each task
417 requires 3 processors on the same node, and the controller will
418 grant an allocation of 4 nodes, one for each of the 4 tasks.
419 --deadline=<OPT>
422 remove the job if no ending is possible before this deadline
423 (start > (deadline - time[-min])). Default is no deadline.
424 Valid time formats are:
425 HH:MM[:SS] [AM|PM]
426 MMDD[YY] or MM/DD[/YY] or MM.DD[.YY]
427 MM/DD[/YY]-HH:MM[:SS]
428 YYYY-MM-DD[THH:MM[:SS]]]
429 now[+count[seconds(default)|minutes|hours|days|weeks]]
430 --delay-boot=<minutes>
433 Do not reboot nodes in order to satisfied this job's feature
434 specification if the job has been eligible to run for less than
435 this time period. If the job has waited for less than the spec‐
436 ified period, it will use only nodes which already have the
437 specified features. The argument is in units of minutes. A
438 default value may be set by a system administrator using the
439 delay_boot option of the SchedulerParameters configuration
440 parameter in the slurm.conf file, otherwise the default value is
441 zero (no delay).
442 -d, --dependency=<dependency_list>
445 Defer the start of this job until the specified dependencies
446 have been satisfied completed. <dependency_list> is of the form
447 <type:job_id[:job_id][,type:job_id[:job_id]]> or
448 <type:job_id[:job_id][?type:job_id[:job_id]]>. All dependencies
449 must be satisfied if the "," separator is used. Any dependency
450 may be satisfied if the "?" separator is used. Only one separa‐
451 tor may be used. Many jobs can share the same dependency and
452 these jobs may even belong to different users. The value may
453 be changed after job submission using the scontrol command.
454 Dependencies on remote jobs are allowed in a federation. Once a
455 job dependency fails due to the termination state of a preceding
456 job, the dependent job will never be run, even if the preceding
457 job is requeued and has a different termination state in a sub‐
458 sequent execution.
459 after:job_id[[+time][:jobid[+time]...]]
461 After the specified jobs start or are cancelled and
462 'time' in minutes from job start or cancellation happens,
463 this job can begin execution. If no 'time' is given then
464 there is no delay after start or cancellation.
465 afterany:job_id[:jobid...]
467 This job can begin execution after the specified jobs
468 have terminated.
469 afterburstbuffer:job_id[:jobid...]
471 This job can begin execution after the specified jobs
472 have terminated and any associated burst buffer stage out
473 operations have completed.
474 aftercorr:job_id[:jobid...]
476 A task of this job array can begin execution after the
477 corresponding task ID in the specified job has completed
478 successfully (ran to completion with an exit code of
479 zero).
480 afternotok:job_id[:jobid...]
482 This job can begin execution after the specified jobs
483 have terminated in some failed state (non-zero exit code,
484 node failure, timed out, etc).
485 afterok:job_id[:jobid...]
487 This job can begin execution after the specified jobs
488 have successfully executed (ran to completion with an
489 exit code of zero).
490 expand:job_id
492 Resources allocated to this job should be used to expand
493 the specified job. The job to expand must share the same
494 QOS (Quality of Service) and partition. Gang scheduling
495 of resources in the partition is also not supported.
496 "expand" is not allowed for jobs that didn't originate on
497 the same cluster as the submitted job.
498 singleton
500 This job can begin execution after any previously
501 launched jobs sharing the same job name and user have
502 terminated. In other words, only one job by that name
503 and owned by that user can be running or suspended at any
504 point in time. In a federation, a singleton dependency
505 must be fulfilled on all clusters unless DependencyParam‐
506 eters=disable_remote_singleton is used in slurm.conf.
507 -D, --chdir=<path>
510 Change directory to path before beginning execution. The path
511 can be specified as full path or relative path to the directory
512 where the command is executed.
513 --exclusive[=user|mcs]
516 The job allocation can not share nodes with other running jobs
517 (or just other users with the "=user" option or with the "=mcs"
518 option). The default shared/exclusive behavior depends on sys‐
519 tem configuration and the partition's OverSubscribe option takes
520 precedence over the job's option.
521 -F, --nodefile=<node file>
524 Much like --nodelist, but the list is contained in a file of
525 name node file. The node names of the list may also span multi‐
526 ple lines in the file. Duplicate node names in the file will
527 be ignored. The order of the node names in the list is not
528 important; the node names will be sorted by Slurm.
529 --get-user-env[=timeout][mode]
532 This option will load login environment variables for the user
533 specified in the --uid option. The environment variables are
534 retrieved by running something of this sort "su - <username> -c
535 /usr/bin/env" and parsing the output. Be aware that any envi‐
536 ronment variables already set in salloc's environment will take
537 precedence over any environment variables in the user's login
538 environment. The optional timeout value is in seconds. Default
539 value is 3 seconds. The optional mode value control the "su"
540 options. With a mode value of "S", "su" is executed without the
541 "-" option. With a mode value of "L", "su" is executed with the
542 "-" option, replicating the login environment. If mode not
543 specified, the mode established at Slurm build time is used.
544 Example of use include "--get-user-env", "--get-user-env=10"
545 "--get-user-env=10L", and "--get-user-env=S". NOTE: This option
546 only works if the caller has an effective uid of "root".
547 --gid=<group>
550 Submit the job with the specified group's group access permis‐
551 sions. group may be the group name or the numerical group ID.
552 In the default Slurm configuration, this option is only valid
553 when used by the user root.
554 -G, --gpus=[<type>:]<number>
557 Specify the total number of GPUs required for the job. An
558 optional GPU type specification can be supplied. For example
559 "--gpus=volta:3". Multiple options can be requested in a comma
560 separated list, for example: "--gpus=volta:3,kepler:1". See
561 also the --gpus-per-node, --gpus-per-socket and --gpus-per-task
562 options.
563 --gpu-bind=[verbose,]<type>
566 Bind tasks to specific GPUs. By default every spawned task can
567 access every GPU allocated to the job. If "verbose," is speci‐
568 fied before <type>, then print out GPU binding information.
569 Supported type options:
571 closest Bind each task to the GPU(s) which are closest. In a
573 NUMA environment, each task may be bound to more than
574 one GPU (i.e. all GPUs in that NUMA environment).
575 map_gpu:<list>
577 Bind by setting GPU masks on tasks (or ranks) as spec‐
578 ified where <list> is
579 <gpu_id_for_task_0>,<gpu_id_for_task_1>,... GPU IDs
580 are interpreted as decimal values unless they are pre‐
581 ceded with '0x' in which case they interpreted as
582 hexadecimal values. If the number of tasks (or ranks)
583 exceeds the number of elements in this list, elements
584 in the list will be reused as needed starting from the
585 beginning of the list. To simplify support for large
586 task counts, the lists may follow a map with an aster‐
587 isk and repetition count. For example
588 "map_gpu:0*4,1*4". If the task/cgroup plugin is used
589 and ConstrainDevices is set in cgroup.conf, then the
590 GPU IDs are zero-based indexes relative to the GPUs
591 allocated to the job (e.g. the first GPU is 0, even if
592 the global ID is 3). Otherwise, the GPU IDs are global
593 IDs, and all GPUs on each node in the job should be
594 allocated for predictable binding results.
595 mask_gpu:<list>
597 Bind by setting GPU masks on tasks (or ranks) as spec‐
598 ified where <list> is
599 <gpu_mask_for_task_0>,<gpu_mask_for_task_1>,... The
600 mapping is specified for a node and identical mapping
601 is applied to the tasks on every node (i.e. the lowest
602 task ID on each node is mapped to the first mask spec‐
603 ified in the list, etc.). GPU masks are always inter‐
604 preted as hexadecimal values but can be preceded with
605 an optional '0x'. To simplify support for large task
606 counts, the lists may follow a map with an asterisk
607 and repetition count. For example
608 "mask_gpu:0x0f*4,0xf0*4". If the task/cgroup plugin
609 is used and ConstrainDevices is set in cgroup.conf,
610 then the GPU IDs are zero-based indexes relative to
611 the GPUs allocated to the job (e.g. the first GPU is
612 0, even if the global ID is 3). Otherwise, the GPU IDs
613 are global IDs, and all GPUs on each node in the job
614 should be allocated for predictable binding results.
615 single:<tasks_per_gpu>
617 Like --gpu-bind=closest, except that each task can
618 only be bound to a single GPU, even when it can be
619 bound to multiple GPUs that are equally close. The
620 GPU to bind to is determined by <tasks_per_gpu>, where
621 the first <tasks_per_gpu> tasks are bound to the first
622 GPU available, the second <tasks_per_gpu> tasks are
623 bound to the second GPU available, etc. This is basi‐
624 cally a block distribution of tasks onto available
625 GPUs, where the available GPUs are determined by the
626 socket affinity of the task and the socket affinity of
627 the GPUs as specified in gres.conf's Cores parameter.
628 --gpu-freq=[<type]=value>[,<type=value>][,verbose]
631 Request that GPUs allocated to the job are configured with spe‐
632 cific frequency values. This option can be used to indepen‐
633 dently configure the GPU and its memory frequencies. After the
634 job is completed, the frequencies of all affected GPUs will be
635 reset to the highest possible values. In some cases, system
636 power caps may override the requested values. The field type
637 can be "memory". If type is not specified, the GPU frequency is
638 implied. The value field can either be "low", "medium", "high",
639 "highm1" or a numeric value in megahertz (MHz). If the speci‐
640 fied numeric value is not possible, a value as close as possible
641 will be used. See below for definition of the values. The ver‐
642 bose option causes current GPU frequency information to be
643 logged. Examples of use include "--gpu-freq=medium,memory=high"
644 and "--gpu-freq=450".
645 Supported value definitions:
647 low the lowest available frequency.
649 medium attempts to set a frequency in the middle of the
651 available range.
652 high the highest available frequency.
654 highm1 (high minus one) will select the next highest avail‐
656 able frequency.
657 --gpus-per-node=[<type>:]<number>
660 Specify the number of GPUs required for the job on each node
661 included in the job's resource allocation. An optional GPU type
662 specification can be supplied. For example
663 "--gpus-per-node=volta:3". Multiple options can be requested in
664 a comma separated list, for example:
665 "--gpus-per-node=volta:3,kepler:1". See also the --gpus,
666 --gpus-per-socket and --gpus-per-task options.
667 --gpus-per-socket=[<type>:]<number>
670 Specify the number of GPUs required for the job on each socket
671 included in the job's resource allocation. An optional GPU type
672 specification can be supplied. For example
673 "--gpus-per-socket=volta:3". Multiple options can be requested
674 in a comma separated list, for example:
675 "--gpus-per-socket=volta:3,kepler:1". Requires job to specify a
676 sockets per node count ( --sockets-per-node). See also the
677 --gpus, --gpus-per-node and --gpus-per-task options.
678 --gpus-per-task=[<type>:]<number>
681 Specify the number of GPUs required for the job on each task to
682 be spawned in the job's resource allocation. An optional GPU
683 type specification can be supplied. For example
684 "--gpus-per-task=volta:1". Multiple options can be requested in
685 a comma separated list, for example:
686 "--gpus-per-task=volta:3,kepler:1". See also the --gpus,
687 --gpus-per-socket and --gpus-per-node options. This option
688 requires an explicit task count, e.g. -n, --ntasks or "--gpus=X
689 --gpus-per-task=Y" rather than an ambiguous range of nodes with
690 -N, --nodes.
691 NOTE: This option will not have any impact on GPU binding,
692 specifically it won't limit the number of devices set for
693 CUDA_VISIBLE_DEVICES.
694 --gres=<list>
697 Specifies a comma delimited list of generic consumable
698 resources. The format of each entry on the list is
699 "name[[:type]:count]". The name is that of the consumable
700 resource. The count is the number of those resources with a
701 default value of 1. The count can have a suffix of "k" or "K"
702 (multiple of 1024), "m" or "M" (multiple of 1024 x 1024), "g" or
703 "G" (multiple of 1024 x 1024 x 1024), "t" or "T" (multiple of
704 1024 x 1024 x 1024 x 1024), "p" or "P" (multiple of 1024 x 1024
705 x 1024 x 1024 x 1024). The specified resources will be allo‐
706 cated to the job on each node. The available generic consumable
707 resources is configurable by the system administrator. A list
708 of available generic consumable resources will be printed and
709 the command will exit if the option argument is "help". Exam‐
710 ples of use include "--gres=gpu:2,mic:1", "--gres=gpu:kepler:2",
711 and "--gres=help".
712 --gres-flags=<type>
715 Specify generic resource task binding options.
716 disable-binding
718 Disable filtering of CPUs with respect to generic
719 resource locality. This option is currently required to
720 use more CPUs than are bound to a GRES (i.e. if a GPU is
721 bound to the CPUs on one socket, but resources on more
722 than one socket are required to run the job). This
723 option may permit a job to be allocated resources sooner
724 than otherwise possible, but may result in lower job per‐
725 formance.
726 NOTE: This option is specific to SelectType=cons_res.
727 enforce-binding
729 The only CPUs available to the job will be those bound to
730 the selected GRES (i.e. the CPUs identified in the
731 gres.conf file will be strictly enforced). This option
732 may result in delayed initiation of a job. For example a
733 job requiring two GPUs and one CPU will be delayed until
734 both GPUs on a single socket are available rather than
735 using GPUs bound to separate sockets, however, the appli‐
736 cation performance may be improved due to improved commu‐
737 nication speed. Requires the node to be configured with
738 more than one socket and resource filtering will be per‐
739 formed on a per-socket basis.
740 NOTE: This option is specific to SelectType=cons_tres.
741 -H, --hold
744 Specify the job is to be submitted in a held state (priority of
745 zero). A held job can now be released using scontrol to reset
746 its priority (e.g. "scontrol release <job_id>").
747 -h, --help
750 Display help information and exit.
751 --hint=<type>
754 Bind tasks according to application hints.
755 NOTE: This option cannot be used in conjunction with
756 --ntasks-per-core, --threads-per-core or -B. If --hint is speci‐
757 fied as a command line argument, it will take precedence over
758 the environment.
759 compute_bound
761 Select settings for compute bound applications: use all
762 cores in each socket, one thread per core.
763 memory_bound
765 Select settings for memory bound applications: use only
766 one core in each socket, one thread per core.
767 [no]multithread
769 [don't] use extra threads with in-core multi-threading
770 which can benefit communication intensive applications.
771 Only supported with the task/affinity plugin.
772 help show this help message
774 -I, --immediate[=<seconds>]
777 exit if resources are not available within the time period spec‐
778 ified. If no argument is given (seconds defaults to 1),
779 resources must be available immediately for the request to suc‐
780 ceed. If defer is configured in SchedulerParameters and sec‐
781 onds=1 the allocation request will fail immediately; defer con‐
782 flicts and takes precedence over this option. By default,
783 --immediate is off, and the command will block until resources
784 become available. Since this option's argument is optional, for
785 proper parsing the single letter option must be followed immedi‐
786 ately with the value and not include a space between them. For
787 example "-I60" and not "-I 60".
788 -J, --job-name=<jobname>
791 Specify a name for the job allocation. The specified name will
792 appear along with the job id number when querying running jobs
793 on the system. The default job name is the name of the "com‐
794 mand" specified on the command line.
795 -K, --kill-command[=signal]
798 salloc always runs a user-specified command once the allocation
799 is granted. salloc will wait indefinitely for that command to
800 exit. If you specify the --kill-command option salloc will send
801 a signal to your command any time that the Slurm controller
802 tells salloc that its job allocation has been revoked. The job
803 allocation can be revoked for a couple of reasons: someone used
804 scancel to revoke the allocation, or the allocation reached its
805 time limit. If you do not specify a signal name or number and
806 Slurm is configured to signal the spawned command at job termi‐
807 nation, the default signal is SIGHUP for interactive and SIGTERM
808 for non-interactive sessions. Since this option's argument is
809 optional, for proper parsing the single letter option must be
810 followed immediately with the value and not include a space
811 between them. For example "-K1" and not "-K 1".
812 -k, --no-kill [=off]
815 Do not automatically terminate a job if one of the nodes it has
816 been allocated fails. The user will assume the responsibilities
817 for fault-tolerance should a node fail. When there is a node
818 failure, any active job steps (usually MPI jobs) on that node
819 will almost certainly suffer a fatal error, but with --no-kill,
820 the job allocation will not be revoked so the user may launch
821 new job steps on the remaining nodes in their allocation.
822 Specify an optional argument of "off" disable the effect of the
824 SALLOC_NO_KILL environment variable.
825 By default Slurm terminates the entire job allocation if any
827 node fails in its range of allocated nodes.
828 -L, --licenses=<license>
831 Specification of licenses (or other resources available on all
832 nodes of the cluster) which must be allocated to this job.
833 License names can be followed by a colon and count (the default
834 count is one). Multiple license names should be comma separated
835 (e.g. "--licenses=foo:4,bar").
836 -M, --clusters=<string>
839 Clusters to issue commands to. Multiple cluster names may be
840 comma separated. The job will be submitted to the one cluster
841 providing the earliest expected job initiation time. The default
842 value is the current cluster. A value of 'all' will query to run
843 on all clusters. Note that the SlurmDBD must be up for this
844 option to work properly.
845 -m, --distribution=
848 arbitrary|<block|cyclic|plane=<options>[:block|cyclic|fcyclic]>
849 Specify alternate distribution methods for remote processes. In
851 salloc, this only sets environment variables that will be used
852 by subsequent srun requests. This option controls the assign‐
853 ment of tasks to the nodes on which resources have been allo‐
854 cated, and the distribution of those resources to tasks for
855 binding (task affinity). The first distribution method (before
856 the ":") controls the distribution of resources across nodes.
857 The optional second distribution method (after the ":") controls
858 the distribution of resources across sockets within a node.
859 Note that with select/cons_res, the number of cpus allocated on
860 each socket and node may be different. Refer to
861 https://slurm.schedmd.com/mc_support.html for more information
862 on resource allocation, assignment of tasks to nodes, and bind‐
863 ing of tasks to CPUs.
864 First distribution method:
866 block The block distribution method will distribute tasks to a
868 node such that consecutive tasks share a node. For exam‐
869 ple, consider an allocation of three nodes each with two
870 cpus. A four-task block distribution request will dis‐
871 tribute those tasks to the nodes with tasks one and two
872 on the first node, task three on the second node, and
873 task four on the third node. Block distribution is the
874 default behavior if the number of tasks exceeds the num‐
875 ber of allocated nodes.
876 cyclic The cyclic distribution method will distribute tasks to a
878 node such that consecutive tasks are distributed over
879 consecutive nodes (in a round-robin fashion). For exam‐
880 ple, consider an allocation of three nodes each with two
881 cpus. A four-task cyclic distribution request will dis‐
882 tribute those tasks to the nodes with tasks one and four
883 on the first node, task two on the second node, and task
884 three on the third node. Note that when SelectType is
885 select/cons_res, the same number of CPUs may not be allo‐
886 cated on each node. Task distribution will be round-robin
887 among all the nodes with CPUs yet to be assigned to
888 tasks. Cyclic distribution is the default behavior if
889 the number of tasks is no larger than the number of allo‐
890 cated nodes.
891 plane The tasks are distributed in blocks of a specified size.
893 The number of tasks distributed to each node is the same
894 as for cyclic distribution, but the taskids assigned to
895 each node depend on the plane size. Additional distribu‐
896 tion specifications cannot be combined with this option.
897 For more details (including examples and diagrams),
898 please see
899 https://slurm.schedmd.com/mc_support.html
900 and
901 https://slurm.schedmd.com/dist_plane.html
902 arbitrary
904 The arbitrary method of distribution will allocate pro‐
905 cesses in-order as listed in file designated by the envi‐
906 ronment variable SLURM_HOSTFILE. If this variable is
907 listed it will over ride any other method specified. If
908 not set the method will default to block. Inside the
909 hostfile must contain at minimum the number of hosts
910 requested and be one per line or comma separated. If
911 specifying a task count (-n, --ntasks=<number>), your
912 tasks will be laid out on the nodes in the order of the
913 file.
914 NOTE: The arbitrary distribution option on a job alloca‐
915 tion only controls the nodes to be allocated to the job
916 and not the allocation of CPUs on those nodes. This
917 option is meant primarily to control a job step's task
918 layout in an existing job allocation for the srun com‐
919 mand.
920 Second distribution method:
923 block The block distribution method will distribute tasks to
925 sockets such that consecutive tasks share a socket.
926 cyclic The cyclic distribution method will distribute tasks to
928 sockets such that consecutive tasks are distributed over
929 consecutive sockets (in a round-robin fashion). Tasks
930 requiring more than one CPU will have all of those CPUs
931 allocated on a single socket if possible.
932 fcyclic
934 The fcyclic distribution method will distribute tasks to
935 sockets such that consecutive tasks are distributed over
936 consecutive sockets (in a round-robin fashion). Tasks
937 requiring more than one CPU will have each CPUs allocated
938 in a cyclic fashion across sockets.
939 --mail-type=<type>
942 Notify user by email when certain event types occur. Valid type
943 values are NONE, BEGIN, END, FAIL, REQUEUE, ALL (equivalent to
944 BEGIN, END, FAIL, INVALID_DEPEND, REQUEUE, and STAGE_OUT),
945 INVALID_DEPEND (dependency never satisfied), STAGE_OUT (burst
946 buffer stage out and teardown completed), TIME_LIMIT,
947 TIME_LIMIT_90 (reached 90 percent of time limit), TIME_LIMIT_80
948 (reached 80 percent of time limit), and TIME_LIMIT_50 (reached
949 50 percent of time limit). Multiple type values may be speci‐
950 fied in a comma separated list. The user to be notified is
951 indicated with --mail-user.
952 --mail-user=<user>
955 User to receive email notification of state changes as defined
956 by --mail-type. The default value is the submitting user.
957 --mcs-label=<mcs>
960 Used only when the mcs/group plugin is enabled. This parameter
961 is a group among the groups of the user. Default value is cal‐
962 culated by the Plugin mcs if it's enabled.
963 --mem=<size[units]>
966 Specify the real memory required per node. Default units are
967 megabytes. Different units can be specified using the suffix
968 [K|M|G|T]. Default value is DefMemPerNode and the maximum value
969 is MaxMemPerNode. If configured, both of parameters can be seen
970 using the scontrol show config command. This parameter would
971 generally be used if whole nodes are allocated to jobs (Select‐
972 Type=select/linear). Also see --mem-per-cpu and --mem-per-gpu.
973 The --mem, --mem-per-cpu and --mem-per-gpu options are mutually
974 exclusive. If --mem, --mem-per-cpu or --mem-per-gpu are speci‐
975 fied as command line arguments, then they will take precedence
976 over the environment.
977 NOTE: A memory size specification of zero is treated as a spe‐
979 cial case and grants the job access to all of the memory on each
980 node. If the job is allocated multiple nodes in a heterogeneous
981 cluster, the memory limit on each node will be that of the node
982 in the allocation with the smallest memory size (same limit will
983 apply to every node in the job's allocation).
984 NOTE: Enforcement of memory limits currently relies upon the
986 task/cgroup plugin or enabling of accounting, which samples mem‐
987 ory use on a periodic basis (data need not be stored, just col‐
988 lected). In both cases memory use is based upon the job's Resi‐
989 dent Set Size (RSS). A task may exceed the memory limit until
990 the next periodic accounting sample.
991 --mem-per-cpu=<size[units]>
994 Minimum memory required per allocated CPU. Default units are
995 megabytes. Different units can be specified using the suffix
996 [K|M|G|T]. The default value is DefMemPerCPU and the maximum
997 value is MaxMemPerCPU (see exception below). If configured, both
998 parameters can be seen using the scontrol show config command.
999 Note that if the job's --mem-per-cpu value exceeds the config‐
1000 ured MaxMemPerCPU, then the user's limit will be treated as a
1001 memory limit per task; --mem-per-cpu will be reduced to a value
1002 no larger than MaxMemPerCPU; --cpus-per-task will be set and the
1003 value of --cpus-per-task multiplied by the new --mem-per-cpu
1004 value will equal the original --mem-per-cpu value specified by
1005 the user. This parameter would generally be used if individual
1006 processors are allocated to jobs (SelectType=select/cons_res).
1007 If resources are allocated by core, socket, or whole nodes, then
1008 the number of CPUs allocated to a job may be higher than the
1009 task count and the value of --mem-per-cpu should be adjusted
1010 accordingly. Also see --mem and --mem-per-gpu. The --mem,
1011 --mem-per-cpu and --mem-per-gpu options are mutually exclusive.
1012 NOTE: If the final amount of memory requested by a job can't be
1014 satisfied by any of the nodes configured in the partition, the
1015 job will be rejected. This could happen if --mem-per-cpu is
1016 used with the --exclusive option for a job allocation and
1017 --mem-per-cpu times the number of CPUs on a node is greater than
1018 the total memory of that node.
1019 --mem-per-gpu=<size[units]>
1022 Minimum memory required per allocated GPU. Default units are
1023 megabytes. Different units can be specified using the suffix
1024 [K|M|G|T]. Default value is DefMemPerGPU and is available on
1025 both a global and per partition basis. If configured, the
1026 parameters can be seen using the scontrol show config and scon‐
1027 trol show partition commands. Also see --mem. The --mem,
1028 --mem-per-cpu and --mem-per-gpu options are mutually exclusive.
1029 --mem-bind=[{quiet,verbose},]type
1032 Bind tasks to memory. Used only when the task/affinity plugin is
1033 enabled and the NUMA memory functions are available. Note that
1034 the resolution of CPU and memory binding may differ on some
1035 architectures. For example, CPU binding may be performed at the
1036 level of the cores within a processor while memory binding will
1037 be performed at the level of nodes, where the definition of
1038 "nodes" may differ from system to system. By default no memory
1039 binding is performed; any task using any CPU can use any memory.
1040 This option is typically used to ensure that each task is bound
1041 to the memory closest to its assigned CPU. The use of any type
1042 other than "none" or "local" is not recommended.
1043 NOTE: To have Slurm always report on the selected memory binding
1045 for all commands executed in a shell, you can enable verbose
1046 mode by setting the SLURM_MEM_BIND environment variable value to
1047 "verbose".
1048 The following informational environment variables are set when
1050 --mem-bind is in use:
1051 SLURM_MEM_BIND_LIST
1053 SLURM_MEM_BIND_PREFER
1054 SLURM_MEM_BIND_SORT
1055 SLURM_MEM_BIND_TYPE
1056 SLURM_MEM_BIND_VERBOSE
1057 See the ENVIRONMENT VARIABLES section for a more detailed
1059 description of the individual SLURM_MEM_BIND* variables.
1060 Supported options include:
1062 help show this help message
1064 local Use memory local to the processor in use
1066 map_mem:<list>
1068 Bind by setting memory masks on tasks (or ranks) as spec‐
1069 ified where <list> is
1070 <numa_id_for_task_0>,<numa_id_for_task_1>,... The map‐
1071 ping is specified for a node and identical mapping is
1072 applied to the tasks on every node (i.e. the lowest task
1073 ID on each node is mapped to the first ID specified in
1074 the list, etc.). NUMA IDs are interpreted as decimal
1075 values unless they are preceded with '0x' in which case
1076 they interpreted as hexadecimal values. If the number of
1077 tasks (or ranks) exceeds the number of elements in this
1078 list, elements in the list will be reused as needed
1079 starting from the beginning of the list. To simplify
1080 support for large task counts, the lists may follow a map
1081 with an asterisk and repetition count. For example
1082 "map_mem:0x0f*4,0xf0*4". For predictable binding
1083 results, all CPUs for each node in the job should be
1084 allocated to the job.
1085 mask_mem:<list>
1087 Bind by setting memory masks on tasks (or ranks) as spec‐
1088 ified where <list> is
1089 <numa_mask_for_task_0>,<numa_mask_for_task_1>,... The
1090 mapping is specified for a node and identical mapping is
1091 applied to the tasks on every node (i.e. the lowest task
1092 ID on each node is mapped to the first mask specified in
1093 the list, etc.). NUMA masks are always interpreted as
1094 hexadecimal values. Note that masks must be preceded
1095 with a '0x' if they don't begin with [0-9] so they are
1096 seen as numerical values. If the number of tasks (or
1097 ranks) exceeds the number of elements in this list, ele‐
1098 ments in the list will be reused as needed starting from
1099 the beginning of the list. To simplify support for large
1100 task counts, the lists may follow a mask with an asterisk
1101 and repetition count. For example "mask_mem:0*4,1*4".
1102 For predictable binding results, all CPUs for each node
1103 in the job should be allocated to the job.
1104 no[ne] don't bind tasks to memory (default)
1106 p[refer]
1108 Prefer use of first specified NUMA node, but permit
1109 use of other available NUMA nodes.
1110 q[uiet]
1112 quietly bind before task runs (default)
1113 rank bind by task rank (not recommended)
1115 sort sort free cache pages (run zonesort on Intel KNL nodes)
1117 v[erbose]
1119 verbosely report binding before task runs
1120 --mincpus=<n>
1123 Specify a minimum number of logical cpus/processors per node.
1124 -N, --nodes=<minnodes[-maxnodes]>
1127 Request that a minimum of minnodes nodes be allocated to this
1128 job. A maximum node count may also be specified with maxnodes.
1129 If only one number is specified, this is used as both the mini‐
1130 mum and maximum node count. The partition's node limits super‐
1131 sede those of the job. If a job's node limits are outside of
1132 the range permitted for its associated partition, the job will
1133 be left in a PENDING state. This permits possible execution at
1134 a later time, when the partition limit is changed. If a job
1135 node limit exceeds the number of nodes configured in the parti‐
1136 tion, the job will be rejected. Note that the environment vari‐
1137 able SLURM_JOB_NODES will be set to the count of nodes actually
1138 allocated to the job. See the ENVIRONMENT VARIABLES section for
1139 more information. If -N is not specified, the default behavior
1140 is to allocate enough nodes to satisfy the requirements of the
1141 -n and -c options. The job will be allocated as many nodes as
1142 possible within the range specified and without delaying the
1143 initiation of the job. The node count specification may include
1144 a numeric value followed by a suffix of "k" (multiplies numeric
1145 value by 1,024) or "m" (multiplies numeric value by 1,048,576).
1146 -n, --ntasks=<number>
1149 salloc does not launch tasks, it requests an allocation of
1150 resources and executed some command. This option advises the
1151 Slurm controller that job steps run within this allocation will
1152 launch a maximum of number tasks and sufficient resources are
1153 allocated to accomplish this. The default is one task per node,
1154 but note that the --cpus-per-task option will change this
1155 default.
1156 --network=<type>
1159 Specify information pertaining to the switch or network. The
1160 interpretation of type is system dependent. This option is sup‐
1161 ported when running Slurm on a Cray natively. It is used to
1162 request using Network Performance Counters. Only one value per
1163 request is valid. All options are case in-sensitive. In this
1164 configuration supported values include:
1165 system
1167 Use the system-wide network performance counters. Only
1168 nodes requested will be marked in use for the job alloca‐
1169 tion. If the job does not fill up the entire system the
1170 rest of the nodes are not able to be used by other jobs
1171 using NPC, if idle their state will appear as PerfCnts.
1172 These nodes are still available for other jobs not using
1173 NPC.
1174 blade Use the blade network performance counters. Only nodes
1176 requested will be marked in use for the job allocation.
1177 If the job does not fill up the entire blade(s) allocated
1178 to the job those blade(s) are not able to be used by other
1179 jobs using NPC, if idle their state will appear as PerfC‐
1180 nts. These nodes are still available for other jobs not
1181 using NPC.
1182 In all cases the job allocation request must specify the
1185 --exclusive option. Otherwise the request will be denied.
1186 Also with any of these options steps are not allowed to share
1188 blades, so resources would remain idle inside an allocation if
1189 the step running on a blade does not take up all the nodes on
1190 the blade.
1191 The network option is also supported on systems with IBM's Par‐
1193 allel Environment (PE). See IBM's LoadLeveler job command key‐
1194 word documentation about the keyword "network" for more informa‐
1195 tion. Multiple values may be specified in a comma separated
1196 list. All options are case in-sensitive. Supported values
1197 include:
1198 BULK_XFER[=<resources>]
1200 Enable bulk transfer of data using Remote Direct-
1201 Memory Access (RDMA). The optional resources speci‐
1202 fication is a numeric value which can have a suffix
1203 of "k", "K", "m", "M", "g" or "G" for kilobytes,
1204 megabytes or gigabytes. NOTE: The resources speci‐
1205 fication is not supported by the underlying IBM in‐
1206 frastructure as of Parallel Environment version 2.2
1207 and no value should be specified at this time.
1208 CAU=<count> Number of Collectve Acceleration Units (CAU)
1210 required. Applies only to IBM Power7-IH processors.
1211 Default value is zero. Independent CAU will be
1212 allocated for each programming interface (MPI, LAPI,
1213 etc.)
1214 DEVNAME=<name>
1216 Specify the device name to use for communications
1217 (e.g. "eth0" or "mlx4_0").
1218 DEVTYPE=<type>
1220 Specify the device type to use for communications.
1221 The supported values of type are: "IB" (InfiniBand),
1222 "HFI" (P7 Host Fabric Interface), "IPONLY" (IP-Only
1223 interfaces), "HPCE" (HPC Ethernet), and "KMUX" (Ker‐
1224 nel Emulation of HPCE). The devices allocated to a
1225 job must all be of the same type. The default value
1226 depends upon depends upon what hardware is available
1227 and in order of preferences is IPONLY (which is not
1228 considered in User Space mode), HFI, IB, HPCE, and
1229 KMUX.
1230 IMMED =<count>
1232 Number of immediate send slots per window required.
1233 Applies only to IBM Power7-IH processors. Default
1234 value is zero.
1235 INSTANCES =<count>
1237 Specify number of network connections for each task
1238 on each network connection. The default instance
1239 count is 1.
1240 IPV4 Use Internet Protocol (IP) version 4 communications
1242 (default).
1243 IPV6 Use Internet Protocol (IP) version 6 communications.
1245 LAPI Use the LAPI programming interface.
1247 MPI Use the MPI programming interface. MPI is the
1249 default interface.
1250 PAMI Use the PAMI programming interface.
1252 SHMEM Use the OpenSHMEM programming interface.
1254 SN_ALL Use all available switch networks (default).
1256 SN_SINGLE Use one available switch network.
1258 UPC Use the UPC programming interface.
1260 US Use User Space communications.
1262 Some examples of network specifications:
1265 Instances=2,US,MPI,SN_ALL
1267 Create two user space connections for MPI communica‐
1268 tions on every switch network for each task.
1269 US,MPI,Instances=3,Devtype=IB
1271 Create three user space connections for MPI communi‐
1272 cations on every InfiniBand network for each task.
1273 IPV4,LAPI,SN_Single
1275 Create a IP version 4 connection for LAPI communica‐
1276 tions on one switch network for each task.
1277 Instances=2,US,LAPI,MPI
1279 Create two user space connections each for LAPI and
1280 MPI communications on every switch network for each
1281 task. Note that SN_ALL is the default option so
1282 every switch network is used. Also note that
1283 Instances=2 specifies that two connections are
1284 established for each protocol (LAPI and MPI) and
1285 each task. If there are two networks and four tasks
1286 on the node then a total of 32 connections are
1287 established (2 instances x 2 protocols x 2 networks
1288 x 4 tasks).
1289 --nice[=adjustment]
1292 Run the job with an adjusted scheduling priority within Slurm.
1293 With no adjustment value the scheduling priority is decreased by
1294 100. A negative nice value increases the priority, otherwise
1295 decreases it. The adjustment range is +/- 2147483645. Only priv‐
1296 ileged users can specify a negative adjustment.
1297 --ntasks-per-core=<ntasks>
1300 Request the maximum ntasks be invoked on each core. Meant to be
1301 used with the --ntasks option. Related to --ntasks-per-node
1302 except at the core level instead of the node level. NOTE: This
1303 option is not supported unless SelectType=cons_res is configured
1304 (either directly or indirectly on Cray systems) along with the
1305 node's core count.
1306 --ntasks-per-gpu=<ntasks>
1309 Request that there are ntasks tasks invoked for every GPU. This
1310 option can work in two ways: 1) either specify --ntasks in addi‐
1311 tion, in which case a type-less GPU specification will be auto‐
1312 matically determined to satisfy --ntasks-per-gpu, or 2) specify
1313 the GPUs wanted (e.g. via --gpus or --gres) without specifying
1314 --ntasks, and the total task count will be automatically deter‐
1315 mined. The number of CPUs needed will be automatically
1316 increased if necessary to allow for any calculated task count.
1317 This option will implicitly set --gpu-bind=single:<ntasks>, but
1318 that can be overridden with an explicit --gpu-bind specifica‐
1319 tion. This option is not compatible with a node range (i.e.
1320 -N<minnodes-maxnodes>). This option is not compatible with
1321 --gpus-per-task, --gpus-per-socket, or --ntasks-per-node. This
1322 option is not supported unless SelectType=cons_tres is config‐
1323 ured (either directly or indirectly on Cray systems).
1324 --ntasks-per-node=<ntasks>
1327 Request that ntasks be invoked on each node. If used with the
1328 --ntasks option, the --ntasks option will take precedence and
1329 the --ntasks-per-node will be treated as a maximum count of
1330 tasks per node. Meant to be used with the --nodes option. This
1331 is related to --cpus-per-task=ncpus, but does not require knowl‐
1332 edge of the actual number of cpus on each node. In some cases,
1333 it is more convenient to be able to request that no more than a
1334 specific number of tasks be invoked on each node. Examples of
1335 this include submitting a hybrid MPI/OpenMP app where only one
1336 MPI "task/rank" should be assigned to each node while allowing
1337 the OpenMP portion to utilize all of the parallelism present in
1338 the node, or submitting a single setup/cleanup/monitoring job to
1339 each node of a pre-existing allocation as one step in a larger
1340 job script.
1341 --ntasks-per-socket=<ntasks>
1344 Request the maximum ntasks be invoked on each socket. Meant to
1345 be used with the --ntasks option. Related to --ntasks-per-node
1346 except at the socket level instead of the node level. NOTE:
1347 This option is not supported unless SelectType=cons_res is con‐
1348 figured (either directly or indirectly on Cray systems) along
1349 with the node's socket count.
1350 --no-bell
1353 Silence salloc's use of the terminal bell. Also see the option
1354 --bell.
1355 --no-shell
1358 immediately exit after allocating resources, without running a
1359 command. However, the Slurm job will still be created and will
1360 remain active and will own the allocated resources as long as it
1361 is active. You will have a Slurm job id with no associated pro‐
1362 cesses or tasks. You can submit srun commands against this
1363 resource allocation, if you specify the --jobid= option with the
1364 job id of this Slurm job. Or, this can be used to temporarily
1365 reserve a set of resources so that other jobs cannot use them
1366 for some period of time. (Note that the Slurm job is subject to
1367 the normal constraints on jobs, including time limits, so that
1368 eventually the job will terminate and the resources will be
1369 freed, or you can terminate the job manually using the scancel
1370 command.)
1371 -O, --overcommit
1374 Overcommit resources. When applied to job allocation, only one
1375 CPU is allocated to the job per node and options used to specify
1376 the number of tasks per node, socket, core, etc. are ignored.
1377 When applied to job step allocations (the srun command when exe‐
1378 cuted within an existing job allocation), this option can be
1379 used to launch more than one task per CPU. Normally, srun will
1380 not allocate more than one process per CPU. By specifying
1381 --overcommit you are explicitly allowing more than one process
1382 per CPU. However no more than MAX_TASKS_PER_NODE tasks are per‐
1383 mitted to execute per node. NOTE: MAX_TASKS_PER_NODE is defined
1384 in the file slurm.h and is not a variable, it is set at Slurm
1385 build time.
1386 -p, --partition=<partition_names>
1389 Request a specific partition for the resource allocation. If
1390 not specified, the default behavior is to allow the slurm con‐
1391 troller to select the default partition as designated by the
1392 system administrator. If the job can use more than one parti‐
1393 tion, specify their names in a comma separate list and the one
1394 offering earliest initiation will be used with no regard given
1395 to the partition name ordering (although higher priority parti‐
1396 tions will be considered first). When the job is initiated, the
1397 name of the partition used will be placed first in the job
1398 record partition string.
1399 --power=<flags>
1402 Comma separated list of power management plugin options. Cur‐
1403 rently available flags include: level (all nodes allocated to
1404 the job should have identical power caps, may be disabled by the
1405 Slurm configuration option PowerParameters=job_no_level).
1406 --priority=<value>
1409 Request a specific job priority. May be subject to configura‐
1410 tion specific constraints. value should either be a numeric
1411 value or "TOP" (for highest possible value). Only Slurm opera‐
1412 tors and administrators can set the priority of a job.
1413 --profile=<all|none|[energy[,|task[,|lustre[,|network]]]]>
1416 enables detailed data collection by the acct_gather_profile
1417 plugin. Detailed data are typically time-series that are stored
1418 in an HDF5 file for the job or an InfluxDB database depending on
1419 the configured plugin.
1420 All All data types are collected. (Cannot be combined with
1423 other values.)
1424 None No data types are collected. This is the default.
1427 (Cannot be combined with other values.)
1428 Energy Energy data is collected.
1431 Task Task (I/O, Memory, ...) data is collected.
1434 Lustre Lustre data is collected.
1437 Network Network (InfiniBand) data is collected.
1440 -q, --qos=<qos>
1443 Request a quality of service for the job. QOS values can be
1444 defined for each user/cluster/account association in the Slurm
1445 database. Users will be limited to their association's defined
1446 set of qos's when the Slurm configuration parameter, Account‐
1447 ingStorageEnforce, includes "qos" in its definition.
1448 -Q, --quiet
1451 Suppress informational messages from salloc. Errors will still
1452 be displayed.
1453 --reboot
1456 Force the allocated nodes to reboot before starting the job.
1457 This is only supported with some system configurations and will
1458 otherwise be silently ignored. Only root, SlurmUser or admins
1459 can reboot nodes.
1460 --reservation=<reservation_names>
1463 Allocate resources for the job from the named reservation. If
1464 the job can use more than one reservation, specify their names
1465 in a comma separate list and the one offering earliest initia‐
1466 tion. Each reservation will be considered in the order it was
1467 requested. All reservations will be listed in scontrol/squeue
1468 through the life of the job. In accounting the first reserva‐
1469 tion will be seen and after the job starts the reservation used
1470 will replace it.
1471 -s, --oversubscribe
1474 The job allocation can over-subscribe resources with other run‐
1475 ning jobs. The resources to be over-subscribed can be nodes,
1476 sockets, cores, and/or hyperthreads depending upon configura‐
1477 tion. The default over-subscribe behavior depends on system
1478 configuration and the partition's OverSubscribe option takes
1479 precedence over the job's option. This option may result in the
1480 allocation being granted sooner than if the --oversubscribe
1481 option was not set and allow higher system utilization, but
1482 application performance will likely suffer due to competition
1483 for resources. Also see the --exclusive option.
1484 -S, --core-spec=<num>
1487 Count of specialized cores per node reserved by the job for sys‐
1488 tem operations and not used by the application. The application
1489 will not use these cores, but will be charged for their alloca‐
1490 tion. Default value is dependent upon the node's configured
1491 CoreSpecCount value. If a value of zero is designated and the
1492 Slurm configuration option AllowSpecResourcesUsage is enabled,
1493 the job will be allowed to override CoreSpecCount and use the
1494 specialized resources on nodes it is allocated. This option can
1495 not be used with the --thread-spec option.
1496 --signal=[R:]<sig_num>[@<sig_time>]
1499 When a job is within sig_time seconds of its end time, send it
1500 the signal sig_num. Due to the resolution of event handling by
1501 Slurm, the signal may be sent up to 60 seconds earlier than
1502 specified. sig_num may either be a signal number or name (e.g.
1503 "10" or "USR1"). sig_time must have an integer value between 0
1504 and 65535. By default, no signal is sent before the job's end
1505 time. If a sig_num is specified without any sig_time, the
1506 default time will be 60 seconds. Use the "R:" option to allow
1507 this job to overlap with a reservation with MaxStartDelay set.
1508 To have the signal sent at preemption time see the pre‐
1509 empt_send_user_signal SlurmctldParameter.
1510 --sockets-per-node=<sockets>
1513 Restrict node selection to nodes with at least the specified
1514 number of sockets. See additional information under -B option
1515 above when task/affinity plugin is enabled.
1516 --spread-job
1519 Spread the job allocation over as many nodes as possible and
1520 attempt to evenly distribute tasks across the allocated nodes.
1521 This option disables the topology/tree plugin.
1522 --switches=<count>[@<max-time>]
1525 When a tree topology is used, this defines the maximum count of
1526 switches desired for the job allocation and optionally the maxi‐
1527 mum time to wait for that number of switches. If Slurm finds an
1528 allocation containing more switches than the count specified,
1529 the job remains pending until it either finds an allocation with
1530 desired switch count or the time limit expires. It there is no
1531 switch count limit, there is no delay in starting the job.
1532 Acceptable time formats include "minutes", "minutes:seconds",
1533 "hours:minutes:seconds", "days-hours", "days-hours:minutes" and
1534 "days-hours:minutes:seconds". The job's maximum time delay may
1535 be limited by the system administrator using the SchedulerParam‐
1536 eters configuration parameter with the max_switch_wait parameter
1537 option. On a dragonfly network the only switch count supported
1538 is 1 since communication performance will be highest when a job
1539 is allocate resources on one leaf switch or more than 2 leaf
1540 switches. The default max-time is the max_switch_wait Sched‐
1541 ulerParameters.
1542 -t, --time=<time>
1545 Set a limit on the total run time of the job allocation. If the
1546 requested time limit exceeds the partition's time limit, the job
1547 will be left in a PENDING state (possibly indefinitely). The
1548 default time limit is the partition's default time limit. When
1549 the time limit is reached, each task in each job step is sent
1550 SIGTERM followed by SIGKILL. The interval between signals is
1551 specified by the Slurm configuration parameter KillWait. The
1552 OverTimeLimit configuration parameter may permit the job to run
1553 longer than scheduled. Time resolution is one minute and second
1554 values are rounded up to the next minute.
1555 A time limit of zero requests that no time limit be imposed.
1557 Acceptable time formats include "minutes", "minutes:seconds",
1558 "hours:minutes:seconds", "days-hours", "days-hours:minutes" and
1559 "days-hours:minutes:seconds".
1560 --thread-spec=<num>
1563 Count of specialized threads per node reserved by the job for
1564 system operations and not used by the application. The applica‐
1565 tion will not use these threads, but will be charged for their
1566 allocation. This option can not be used with the --core-spec
1567 option.
1568 --threads-per-core=<threads>
1571 Restrict node selection to nodes with at least the specified
1572 number of threads per core. In task layout, use the specified
1573 maximum number of threads per core. NOTE: "Threads" refers to
1574 the number of processing units on each core rather than the num‐
1575 ber of application tasks to be launched per core. See addi‐
1576 tional information under -B option above when task/affinity
1577 plugin is enabled.
1578 --time-min=<time>
1581 Set a minimum time limit on the job allocation. If specified,
1582 the job may have its --time limit lowered to a value no lower
1583 than --time-min if doing so permits the job to begin execution
1584 earlier than otherwise possible. The job's time limit will not
1585 be changed after the job is allocated resources. This is per‐
1586 formed by a backfill scheduling algorithm to allocate resources
1587 otherwise reserved for higher priority jobs. Acceptable time
1588 formats include "minutes", "minutes:seconds", "hours:min‐
1589 utes:seconds", "days-hours", "days-hours:minutes" and
1590 "days-hours:minutes:seconds".
1591 --tmp=<size[units]>
1594 Specify a minimum amount of temporary disk space per node.
1595 Default units are megabytes. Different units can be specified
1596 using the suffix [K|M|G|T].
1597 --usage
1600 Display brief help message and exit.
1601 --uid=<user>
1604 Attempt to submit and/or run a job as user instead of the invok‐
1605 ing user id. The invoking user's credentials will be used to
1606 check access permissions for the target partition. This option
1607 is only valid for user root. This option may be used by user
1608 root may use this option to run jobs as a normal user in a
1609 RootOnly partition for example. If run as root, salloc will drop
1610 its permissions to the uid specified after node allocation is
1611 successful. user may be the user name or numerical user ID.
1612 --use-min-nodes
1615 If a range of node counts is given, prefer the smaller count.
1616 -V, --version
1619 Display version information and exit.
1620 -v, --verbose
1623 Increase the verbosity of salloc's informational messages. Mul‐
1624 tiple -v's will further increase salloc's verbosity. By default
1625 only errors will be displayed.
1626 -w, --nodelist=<node name list>
1629 Request a specific list of hosts. The job will contain all of
1630 these hosts and possibly additional hosts as needed to satisfy
1631 resource requirements. The list may be specified as a
1632 comma-separated list of hosts, a range of hosts (host[1-5,7,...]
1633 for example), or a filename. The host list will be assumed to
1634 be a filename if it contains a "/" character. If you specify a
1635 minimum node or processor count larger than can be satisfied by
1636 the supplied host list, additional resources will be allocated
1637 on other nodes as needed. Duplicate node names in the list will
1638 be ignored. The order of the node names in the list is not
1639 important; the node names will be sorted by Slurm.
1640 --wait-all-nodes=<value>
1643 Controls when the execution of the command begins with respect
1644 to when nodes are ready for use (i.e. booted). By default, the
1645 salloc command will return as soon as the allocation is made.
1646 This default can be altered using the salloc_wait_nodes option
1647 to the SchedulerParameters parameter in the slurm.conf file.
1648 0 Begin execution as soon as allocation can be made. Do not
1650 wait for all nodes to be ready for use (i.e. booted).
1651 1 Do not begin execution until all nodes are ready for use.
1653 --wckey=<wckey>
1656 Specify wckey to be used with job. If TrackWCKey=no (default)
1657 in the slurm.conf this value is ignored.
1658 -x, --exclude=<node name list>
1661 Explicitly exclude certain nodes from the resources granted to
1662 the job.
1663 --x11[=<all|first|last>]
1666 Sets up X11 forwarding on all, first or last node(s) of the
1667 allocation. This option is only enabled if Slurm was compiled
1668 with X11 support and PrologFlags=x11 is defined in the
1669 slurm.conf. Default is all.
1670
1673 Executing salloc sends a remote procedure call to slurmctld. If enough
1674 calls from salloc or other Slurm client commands that send remote pro‐
1675 cedure calls to the slurmctld daemon come in at once, it can result in
1676 a degradation of performance of the slurmctld daemon, possibly result‐
1677 ing in a denial of service.
1678 Do not run salloc or other Slurm client commands that send remote pro‐
1680 cedure calls to slurmctld from loops in shell scripts or other pro‐
1681 grams. Ensure that programs limit calls to salloc to the minimum neces‐
1682 sary for the information you are trying to gather.
1683
1686 Upon startup, salloc will read and handle the options set in the fol‐
1687 lowing environment variables. Note: Command line options always over‐
1688 ride environment variables settings.
1689 SALLOC_ACCOUNT Same as -A, --account
1692 SALLOC_ACCTG_FREQ Same as --acctg-freq
1694 SALLOC_BELL Same as --bell
1696 SALLOC_BURST_BUFFER Same as --bb
1698 SALLOC_CLUSTERS or SLURM_CLUSTERS
1700 Same as --clusters
1701 SALLOC_CONSTRAINT Same as -C, --constraint
1703 SALLOC_CORE_SPEC Same as --core-spec
1705 SALLOC_CPUS_PER_GPU Same as --cpus-per-gpu
1707 SALLOC_DEBUG Same as -v, --verbose
1709 SALLOC_DELAY_BOOT Same as --delay-boot
1711 SALLOC_EXCLUSIVE Same as --exclusive
1713 SALLOC_GPUS Same as -G, --gpus
1715 SALLOC_GPU_BIND Same as --gpu-bind
1717 SALLOC_GPU_FREQ Same as --gpu-freq
1719 SALLOC_GPUS_PER_NODE Same as --gpus-per-node
1721 SALLOC_GPUS_PER_TASK Same as --gpus-per-task
1723 SALLOC_GRES Same as --gres
1725 SALLOC_GRES_FLAGS Same as --gres-flags
1727 SALLOC_HINT or SLURM_HINT
1729 Same as --hint
1730 SALLOC_IMMEDIATE Same as -I, --immediate
1732 SALLOC_KILL_CMD Same as -K, --kill-command
1734 SALLOC_MEM_BIND Same as --mem-bind
1736 SALLOC_MEM_PER_CPU Same as --mem-per-cpu
1738 SALLOC_MEM_PER_GPU Same as --mem-per-gpu
1740 SALLOC_MEM_PER_NODE Same as --mem
1742 SALLOC_NETWORK Same as --network
1744 SALLOC_NO_BELL Same as --no-bell
1746 SALLOC_NO_KILL Same as -k, --no-kill
1748 SALLOC_OVERCOMMIT Same as -O, --overcommit
1750 SALLOC_PARTITION Same as -p, --partition
1752 SALLOC_POWER Same as --power
1754 SALLOC_PROFILE Same as --profile
1756 SALLOC_QOS Same as --qos
1758 SALLOC_REQ_SWITCH When a tree topology is used, this defines the
1760 maximum count of switches desired for the job
1761 allocation and optionally the maximum time to
1762 wait for that number of switches. See --switches.
1763 SALLOC_RESERVATION Same as --reservation
1765 SALLOC_SIGNAL Same as --signal
1767 SALLOC_SPREAD_JOB Same as --spread-job
1769 SALLOC_THREAD_SPEC Same as --thread-spec
1771 SALLOC_TIMELIMIT Same as -t, --time
1773 SALLOC_USE_MIN_NODES Same as --use-min-nodes
1775 SALLOC_WAIT_ALL_NODES Same as --wait-all-nodes
1777 SALLOC_WCKEY Same as --wckey
1779 SALLOC_WAIT4SWITCH Max time waiting for requested switches. See
1781 --switches
1782 SLURM_CONF The location of the Slurm configuration file.
1784 SLURM_EXIT_ERROR Specifies the exit code generated when a Slurm
1786 error occurs (e.g. invalid options). This can be
1787 used by a script to distinguish application exit
1788 codes from various Slurm error conditions. Also
1789 see SLURM_EXIT_IMMEDIATE.
1790 SLURM_EXIT_IMMEDIATE Specifies the exit code generated when the
1792 --immediate option is used and resources are not
1793 currently available. This can be used by a
1794 script to distinguish application exit codes from
1795 various Slurm error conditions. Also see
1796 SLURM_EXIT_ERROR.
1797
1800 salloc will set the following environment variables in the environment
1801 of the executed program:
1802 SLURM_*_HET_GROUP_#
1804 For a heterogeneous job allocation, the environment variables
1805 are set separately for each component.
1806 SLURM_CLUSTER_NAME
1808 Name of the cluster on which the job is executing.
1809 SLURM_CPUS_PER_GPU
1811 Number of CPUs requested per allocated GPU. Only set if the
1812 --cpus-per-gpu option is specified.
1813 SLURM_CPUS_PER_TASK
1815 Number of CPUs requested per task. Only set if the
1816 --cpus-per-task option is specified.
1817 SLURM_DIST_PLANESIZE
1819 Plane distribution size. Only set for plane distributions. See
1820 -m, --distribution.
1821 SLURM_DISTRIBUTION
1823 Only set if the -m, --distribution option is specified.
1824 SLURM_GPUS
1826 Number of GPUs requested. Only set if the -G, --gpus option is
1827 specified.
1828 SLURM_GPU_BIND
1830 Requested binding of tasks to GPU. Only set if the --gpu-bind
1831 option is specified.
1832 SLURM_GPU_FREQ
1834 Requested GPU frequency. Only set if the --gpu-freq option is
1835 specified.
1836 SLURM_GPUS_PER_NODE
1838 Requested GPU count per allocated node. Only set if the
1839 --gpus-per-node option is specified.
1840 SLURM_GPUS_PER_SOCKET
1842 Requested GPU count per allocated socket. Only set if the
1843 --gpus-per-socket option is specified.
1844 SLURM_GPUS_PER_TASK
1846 Requested GPU count per allocated task. Only set if the
1847 --gpus-per-task option is specified.
1848 SLURM_JOB_ACCOUNT
1850 Account name associated of the job allocation.
1851 SLURM_JOB_ID (and SLURM_JOBID for backwards compatibility)
1853 The ID of the job allocation.
1854 SLURM_JOB_CPUS_PER_NODE
1856 Count of processors available to the job on this node. Note the
1857 select/linear plugin allocates entire nodes to jobs, so the
1858 value indicates the total count of CPUs on each node. The
1859 select/cons_res plugin allocates individual processors to jobs,
1860 so this number indicates the number of processors on each node
1861 allocated to the job allocation.
1862 SLURM_JOB_NODELIST (and SLURM_NODELIST for backwards compatibility)
1864 List of nodes allocated to the job.
1865 SLURM_JOB_NUM_NODES (and SLURM_NNODES for backwards compatibility)
1867 Total number of nodes in the job allocation.
1868 SLURM_JOB_PARTITION
1870 Name of the partition in which the job is running.
1871 SLURM_JOB_QOS
1873 Quality Of Service (QOS) of the job allocation.
1874 SLURM_JOB_RESERVATION
1876 Advanced reservation containing the job allocation, if any.
1877 SLURM_MEM_BIND
1879 Set to value of the --mem-bind option.
1880 SLURM_MEM_BIND_LIST
1882 Set to bit mask used for memory binding.
1883 SLURM_MEM_BIND_PREFER
1885 Set to "prefer" if the --mem-bind option includes the prefer
1886 option.
1887 SLURM_MEM_BIND_SORT
1889 Sort free cache pages (run zonesort on Intel KNL nodes)
1890 SLURM_MEM_BIND_TYPE
1892 Set to the memory binding type specified with the --mem-bind
1893 option. Possible values are "none", "rank", "map_map",
1894 "mask_mem" and "local".
1895 SLURM_MEM_BIND_VERBOSE
1897 Set to "verbose" if the --mem-bind option includes the verbose
1898 option. Set to "quiet" otherwise.
1899 SLURM_MEM_PER_CPU
1901 Same as --mem-per-cpu
1902 SLURM_MEM_PER_GPU
1904 Requested memory per allocated GPU. Only set if the
1905 --mem-per-gpu option is specified.
1906 SLURM_MEM_PER_NODE
1908 Same as --mem
1909 SLURM_HET_SIZE
1911 Set to count of components in heterogeneous job.
1912 SLURM_SUBMIT_DIR
1914 The directory from which salloc was invoked or, if applicable,
1915 the directory specified by the -D, --chdir option.
1916 SLURM_SUBMIT_HOST
1918 The hostname of the computer from which salloc was invoked.
1919 SLURM_NODE_ALIASES
1921 Sets of node name, communication address and hostname for nodes
1922 allocated to the job from the cloud. Each element in the set if
1923 colon separated and each set is comma separated. For example:
1924 SLURM_NODE_ALIASES=ec0:1.2.3.4:foo,ec1:1.2.3.5:bar
1925 SLURM_NTASKS
1927 Same as -n, --ntasks
1928 SLURM_NTASKS_PER_CORE
1930 Set to value of the --ntasks-per-core option, if specified.
1931 SLURM_NTASKS_PER_GPU
1933 Set to value of the --ntasks-per-gpu option, if specified.
1934 SLURM_NTASKS_PER_NODE
1936 Set to value of the --ntasks-per-node option, if specified.
1937 SLURM_NTASKS_PER_SOCKET
1939 Set to value of the --ntasks-per-socket option, if specified.
1940 SLURM_PROFILE
1942 Same as --profile
1943 SLURM_TASKS_PER_NODE
1945 Number of tasks to be initiated on each node. Values are comma
1946 separated and in the same order as SLURM_JOB_NODELIST. If two
1947 or more consecutive nodes are to have the same task count, that
1948 count is followed by "(x#)" where "#" is the repetition count.
1949 For example, "SLURM_TASKS_PER_NODE=2(x3),1" indicates that the
1950 first three nodes will each execute two tasks and the fourth
1951 node will execute one task.
1952
1955 While salloc is waiting for a PENDING job allocation, most signals will
1956 cause salloc to revoke the allocation request and exit.
1957 However if the allocation has been granted and salloc has already
1959 started the specified command, then salloc will ignore most signals.
1960 salloc will not exit or release the allocation until the command exits.
1961 One notable exception is SIGHUP. A SIGHUP signal will cause salloc to
1962 release the allocation and exit without waiting for the command to fin‐
1963 ish. Another exception is SIGTERM, which will be forwarded to the
1964 spawned process.
1965
1968 To get an allocation, and open a new xterm in which srun commands may
1969 be typed interactively:
1970 $ salloc -N16 xterm
1972 salloc: Granted job allocation 65537
1973 (at this point the xterm appears, and salloc waits for xterm to
1974 exit)
1975 salloc: Relinquishing job allocation 65537
1976 To grab an allocation of nodes and launch a parallel application on one
1978 command line (See the salloc man page for more examples):
1979 salloc -N5 srun -n10 myprogram
1981 To create a heterogeneous job with 3 components, each allocating a
1983 unique set of nodes:
1984 salloc -w node[2-3] : -w node4 : -w node[5-7] bash
1986 salloc: job 32294 queued and waiting for resources
1987 salloc: job 32294 has been allocated resources
1988 salloc: Granted job allocation 32294
1989
1992 Copyright (C) 2006-2007 The Regents of the University of California.
1993 Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
1994 Copyright (C) 2008-2010 Lawrence Livermore National Security.
1995 Copyright (C) 2010-2018 SchedMD LLC.
1996 This file is part of Slurm, a resource management program. For
1998 details, see <https://slurm.schedmd.com/>.
1999 Slurm is free software; you can redistribute it and/or modify it under
2001 the terms of the GNU General Public License as published by the Free
2002 Software Foundation; either version 2 of the License, or (at your
2003 option) any later version.
2004 Slurm is distributed in the hope that it will be useful, but WITHOUT
2006 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
2007 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
2008 for more details.
2009
2012 sinfo(1), sattach(1), sbatch(1), squeue(1), scancel(1), scontrol(1),
2013 slurm.conf(5), sched_setaffinity (2), numa (3)
2014November 2020 Slurm Commands salloc(1)