salloc(1)

salloc(1) 💅✨

okay so bestie, this is the Slurm Commands manual page for salloc. Buckle up, because this one is literally about grabbing nodes and making them do your bidding. 🫶

🎀 NAME 🎀

• salloc — obtains a Slurm job allocation (a set of nodes), executes a command, and then releases the allocation when the command is finished. 💖

✨ SYNOPSIS ✨

salloc [OPTIONS(0)...] [ : [OPTIONS(N)...]] [command(0) [args(0)...]]

Option(s) define multiple jobs in a co-scheduled heterogeneous job. For more deets about heterogeneous jobs, peep this document: https://slurm.schedmd.com/heterogeneous_jobs.html

🌸 DESCRIPTION 🌸

okay so salloc is used to allocate a Slurm job allocation, which is a set of resources (nodes), possibly with some set of constraints (e.g. number of processors per node). When salloc successfully obtains the requested allocation, it then runs the command specified by the user. Finally, when the user-specified command is complete, salloc relinquishes the job allocation. iconic behavior, honestly. 💅

The command may be any program the user wishes. Some typical commands are xterm, a shell script containing srun commands, and srun (see the EXAMPLES section). If no command is specified, then salloc runs the user's default shell. 🐚

The following document describes the influence of various options on the allocation of cpus to jobs and tasks: https://slurm.schedmd.com/cpu_management.html

NOTE: The salloc logic includes support to save and restore the terminal line settings and is designed to be executed in the foreground. If you need to execute salloc in the background, set its standard input to some file, for example: salloc -n16 a.out </dev/null &

💀 RETURN VALUE 💀

If salloc is unable to execute the user command, it will return 1 and print errors to stderr. Else if success or if killed by signals HUP, INT, KILL, or QUIT: it will return 0. that's the whole vibe check, no cap. ✨

🦋 COMMAND PATH RESOLUTION 🦋

If provided, the command is resolved in the following order:

1. If command starts with ., then path is constructed as: current working directory / command
2. If command starts with a /, then path is considered absolute.
3. If command can be resolved through PATH. See path_resolution(7).
4. If command is in current working directory.

Current working directory is the calling process working directory unless the --chdir argument is passed, which will override the current working directory. lowkey important, bestie. 🎀

✨ OPTIONS ✨

• -A, --account=<account> — Charge resources used by this job to specified account. The account is an arbitrary string. The account name may be changed after job submission using the scontrol command. 💸
• --acctg-freq=<datatype>=<interval>[,<datatype>=<interval>...] — Define the job accounting and profiling sampling intervals in seconds. This can be used to override the JobAcctGatherFrequency parameter in the slurm.conf file. <datatype>=<interval> specifies the task sampling interval for the jobacct_gather plugin or a sampling interval for a profiling type by the acct_gather_profile plugin. Multiple comma-separated <datatype>=<interval> pairs may be specified. Supported datatype values are:
  • task — Sampling interval for the jobacct_gather plugins and for task profiling by the acct_gather_profile plugin. NOTE: This frequency is used to monitor memory usage. If memory limits are enforced the highest frequency a user can request is what is configured in the slurm.conf file. It can not be disabled. 🧠
  • energy — Sampling interval for energy profiling using the acct_gather_energy plugin. ⚡
  • network — Sampling interval for infiniband profiling using the acct_gather_interconnect plugin. 🌐
  • filesystem — Sampling interval for filesystem profiling using the acct_gather_filesystem plugin. 💾

The default value for the task sampling interval is 30 seconds. The default value for all other intervals is 0. An interval of 0 disables sampling of the specified type. If the task sampling interval is 0, accounting information is collected only at job termination (reducing Slurm interference with the job). Smaller (non-zero) values have a greater impact upon job performance, but a value of 30 seconds is not likely to be noticeable for applications having less than 10,000 tasks. 🌈

• --bb=<spec> — Burst buffer specification. The form of the specification is system dependent. Note the burst buffer may not be accessible from a login node, but require that salloc spawn a shell on one of its allocated compute nodes. When the --bb option is used, Slurm parses this option and creates a temporary burst buffer script file that is used internally by the burst buffer plugins. See Slurm's burst buffer guide for more information and examples: https://slurm.schedmd.com/burst_buffer.html 🫧
• --bbf=<file_name> — Path of file containing burst buffer specification. The form of the specification is system dependent. Also see --bb. Note the burst buffer may not be accessible from a login node, but require that salloc spawn a shell on one of its allocated compute nodes. See Slurm's burst buffer guide for more information and examples: https://slurm.schedmd.com/burst_buffer.html 🫧
• --begin=<time> — Defer eligibility of this job allocation until the specified time. ⏰

Time may be of the form HH:MM:SS to run a job at a specific time of day (seconds are optional). (If that time is already past, the next day is assumed.) You may also specify midnight, noon, elevenses (11 AM), fika (3 PM) or teatime (4 PM) and you can have a time-of-day suffixed with AM or PM for running in the morning or the evening. You can also say what day the job will be run, by specifying a date of the form MMDDYY or MM/DD/YY YYYY-MM-DD. Combine date and time using the following format YYYY-MM-DD[THH:MM[:SS]]. You can also give times like now + count time-units, where the time-units can be seconds (default), minutes, hours, days, or weeks. The keywords today and tomorrow may also be used. The value may be changed after job submission using the scontrol command. For example:

--begin=16:00
--begin=now+1hour
--begin=now+60           (seconds by default)
--begin=2010-01-20T12:34:00

Notes on date/time specifications:

• Although the 'seconds' field of the HH:MM:SS time specification is allowed by the code, note that the poll time of the Slurm scheduler is not precise enough to guarantee dispatch of the job on the exact second. The job will be eligible to start on the next poll following the specified time. The exact poll interval depends on the Slurm scheduler (e.g., 60 seconds with the default sched/builtin). ⏳
• If no time (HH:MM:SS) is specified, the default is (00:00:00).
• If a date is specified without a year (e.g., MM/DD) then the current year is assumed, unless the combination of MM/DD and HH:MM:SS has already passed for that year, in which case the next year is used. 🗓️

• --bell — Force salloc to ring the terminal bell when the job allocation is granted (and only if stdout is a tty). By default, salloc only rings the bell if the allocation is pending for more than ten seconds (and only if stdout is a tty). Also see the option --no-bell. 🔔
• -D, --chdir=<path> — Change directory to path before beginning execution. The path can be specified as full path or relative path to the directory where the command is executed. 📁
• --cluster-constraint=<list> — Specifies features that a federated cluster must have to have a sibling job submitted to it. Slurm will attempt to submit a sibling job to a cluster if it has at least one of the specified features. 🌐
• -M, --clusters=<string> — Clusters to issue commands to. Multiple cluster names may be comma separated. The job will be submitted to the one cluster providing the earliest expected job initiation time. The default value is the current cluster. A value of 'all' will query to run on all clusters. Note that the slurmdbd must be up for this option to work properly, unless running in a federation with FederationParameters=fed_display configured. 👑
• --comment=<string> — An arbitrary comment. 💬
• --consolidate-segments — Ensure that all segments from the allocation will be consolidated into one higher-level aggregated block. 🧩

NOTE: This option will only work with the topology/block plugin.

• -C, --constraint=<list> — Nodes can have features assigned to them by the Slurm administrator. Users can specify which of these features are required by their job using the constraint option. If you are looking for 'soft' constraints please see --prefer for more information. Only nodes having features matching the job constraints will be used to satisfy the request. Multiple constraints may be specified with AND, OR, matching OR, resource counts, etc. (some operators are not supported on all system types). 🎯

NOTE: Changeable features are features defined by a NodeFeatures plugin.

Supported --constraint options include:

• Single Name — Only nodes which have the specified feature will be used. For example, --constraint="intel" 💻
• Node Count — A request can specify the number of nodes needed with some feature by appending an asterisk and count after the feature name. For example, --nodes=16 --constraint="graphics*4" indicates that the job requires 16 nodes and that at least four of those nodes must have the feature "graphics." If requesting more than one feature and using node counts, the request must have square brackets surrounding it. 🖥️

NOTE: This option is not supported by the helpers NodeFeatures plugin. Heterogeneous jobs can be used instead.

• AND — Only nodes with all of specified features will be used. The ampersand is used for an AND operator. For example, --constraint="intel&gpu" 🔥

OR

Only nodes rockin' at least one of the features you name will get used, periodt. The vertical bar is the OR operator, iykyk. If you're not asking for changeable features, the nodes in your allocation can lowkey have totally different features from each other. Like, salloc -N2 --constraint="intel|amd" can totally result in a job allocation where one node has the intel feature and the other node has the amd feature. 💅

But bestie, the second your expression contains a changeable feature, all OR operators are automatically treated as Matching OR, so that all nodes in the job allocation have the same set of features. Example time:

salloc -N2 --constraint="foo|bar&baz"

That gets you two nodes where both nodes have foo, or bar and baz (one or both nodes could have foo, bar, and baz, she's flexible like that). The helpers NodeFeatures plugin will find the first set of node features that matches all nodes in the job allocation; those features get set as active features on the node and passed to RebootProgram (see slurm.conf(5)) and the helper script (see helpers.conf(5)). In this case, the helpers plugin uses the first of "foo" or "bar,baz" that match the two nodes in the job allocation. Slay. ✨

Matching OR

Okay so if only one of a set of possible options should be used for all allocated nodes, use the OR operator and enclose the options within square brackets. Like --constraint="[rack1|rack2|rack3|rack4]" might be used to specify that all nodes must be allocated on a single rack of the cluster, but any of those four racks can be used. Iconic behavior, no cap. 🎀

Multiple Counts

Specific counts of multiple resources? You can specify those by using the AND operator and enclosing the options within square brackets. For example, --constraint="[rack1*2&rack2*4]" might be used to specify that two nodes must be allocated from nodes with the feature of "rack1" and four nodes must be allocated from nodes with the feature "rack2".

NOTE: This option is not supported by the helpers NodeFeatures plugin. 🫶

NOTE: Multiple Counts can cause jobs to be allocated with a non-optimal network layout. Don't say we didn't warn you. 😭

Brackets

Brackets can be used to indicate that you're looking for a set of nodes with the different requirements contained within the brackets. For example, --constraint="[(rack1|rack2)*1&(rack3)*2]" will get you one node with either the "rack1" or "rack2" features and two nodes with the "rack3" feature. If you're requesting more than one feature and using node counts, the request must have square brackets surrounding it. Fr fr, no exceptions.

NOTE: Brackets are only reserved for Multiple Counts and Matching OR syntax. AND operators require a count for each feature inside square brackets (i.e. "[quad*2&hemi*1]"). Slurm will only allow a single set of bracketed constraints per job. 👑

NOTE: Square brackets are not supported by the helpers NodeFeatures plugin. Matching OR can be requested without square brackets by using the vertical bar character with at least one changeable feature.

Parentheses

Parentheses can be used to group like node features together. For example, --constraint="[(knl&snc4&flat)*4&haswell*1]" might be used to specify that four nodes with the features "knl", "snc4" and "flat" plus one node with the feature "haswell" are required. 🦋

Parentheses can also group operations. Without parentheses, node features are parsed strictly from left to right — literally, no vibes-based reinterpretation. So:

• --constraint="foo&bar|baz" requests nodes with foo and bar, or baz.
• --constraint="foo|bar&baz" requests nodes with foo and baz, or bar and baz (note how baz was AND'd with everything, she's the main character of that expression 😭).
• --constraint="foo&(bar|baz)" requests nodes with foo and at least one of bar or baz.

NOTE: OR within parentheses should not be used with a KNL NodeFeatures plugin but is supported by the helpers NodeFeatures plugin.

---

• --container=<path_to_container> — absolute path to the OCI container bundle. That's it, that's the whole vibe. 🧚
• --container-id=<container_id> — a unique name for the OCI container, so each one gets her own little name tag.
• --container-type=<container_type> — the job container type to use for the job.
• --contiguous — if set, the allocated nodes must form a contiguous set. Cute, tidy, all lined up. ✨

NOTE: Whether this option is honored depends on the topology plugin! It's enforced by topology/flat and topology/tree. Under topology/tree it also disables the topology-aware node selection optimizations, so the resulting allocation may span nodes with greater network distance between them. It is silently ignored by all other topology plugins. 🙈

• -S, --core-spec=<num> — count of Specialized Cores per node reserved by the job for system operations and not used by the application. If AllowSpecResourcesUsage is enabled, a job can override the CoreSpecCount of all its allocated nodes with this option. The overridden Specialized Cores will still be reserved for system processes. The job will get an implicit --exclusive allocation for the rest of the Cores on the nodes, resulting in the job's processes being able to use (and being charged for) all the Cores on the nodes except for the overridden Specialized Cores. This option can not be used with the --thread-spec option. 💀

NOTE: Explicitly setting a job's specialized core value implicitly sets the --exclusive option. The way that just happens... iconic but also read the fine print.

• --cores-per-socket=<cores> — restrict node selection to nodes with at least the specified number of cores per socket. See additional information under the -B option above when the task/affinity plugin is enabled.

NOTE: This option may implicitly set the number of tasks (if -n was not specified) as one task per requested thread. Sneaky but logical. 🍓

• --cpu-freq=<p1>[-p2][:p3] — request that job steps initiated by srun commands inside this allocation be run at some requested frequency if possible, on the CPUs selected for the step on the compute node(s). Bestie, this is the whole frequency saga, buckle up:

p1 can be [#### | low | medium | high | highm1] which will set the frequency scaling_speed to the corresponding value, and set the frequency scaling_governor to UserSpace. See below for definition of the values.

p1 can be [Conservative | OnDemand | Performance | PowerSave] which will set the scaling_governor to the corresponding value. The governor has to be in the list set by the slurm.conf option CpuFreqGovernors.

When p2 is present, p1 will be the minimum scaling frequency and p2 will be the maximum scaling frequency. In that case the governor p3 or CpuFreqDef cannot be UserSpace since it doesn't support a range.

p2 can be [#### | medium | high | highm1]. p2 must be greater than p1 and is incompatible with UserSpace governor.

p3 can be [Conservative | OnDemand | Performance | PowerSave | SchedUtil | UserSpace] which will set the governor to the corresponding value.

If p3 is UserSpace, the frequency scaling_speed, scaling_max_freq and scaling_min_freq will be statically set to the value defined by p1.

Any requested frequency below the minimum available frequency will be rounded up to the minimum available frequency. In the same way, any requested frequency above the maximum available frequency will be rounded to the maximum available frequency. Very "close enough" of her. 🌸

The CpuFreqDef parameter in slurm.conf will be used to set the governor in absence of p3. If there's no CpuFreqDef, the default governor will be to use the system current governor set in each cpu. Specifying a range without CpuFreqDef or a specific governor is therefore not allowed. 🚫

Acceptable values at present include:

• #### — frequency in kilohertz
• Low — the lowest available frequency
• High — the highest available frequency
• HighM1 — (high minus one) will select the next highest available frequency
• Medium — attempts to set a frequency in the middle of the available range
• Conservative — attempts to use the Conservative CPU governor
• OnDemand — attempts to use the OnDemand CPU governor (the default value)
• Performance — attempts to use the Performance CPU governor
• PowerSave — attempts to use the PowerSave CPU governor
• UserSpace — attempts to use the UserSpace CPU governor

The following informational environment variable is set in the job step when --cpu-freq option is requested:

SLURM_CPU_FREQ_REQ

This environment variable can also be used to supply the value for the CPU frequency request if it is set when the srun command is issued. The --cpu-freq on the command line will override the environment variable value. The form on the environment variable is the same as the command line. See the ENVIRONMENT VARIABLES section for a description of the SLURM_CPU_FREQ_REQ variable. ✨

NOTE: This parameter is treated as a request, not a requirement. If the job step's node does not support setting the CPU frequency, or the requested value is outside the bounds of the legal frequencies, an error is logged, but the job step is allowed to continue. She's doing her best. 😭

NOTE: Setting the frequency for just the CPUs of the job step implies that the tasks are confined to those CPUs. If task confinement (i.e. the task/affinity TaskPlugin is enabled, or the task/cgroup TaskPlugin is enabled with "ConstrainCores=yes" set in cgroup.conf) is not configured, this parameter is ignored. 🫶

NOTE: When the step completes, the frequency and governor of each selected CPU is reset to the previous values. Cleanup queen. 💜

⚠️ NOTE: if you're submitting jobs with --cpu-freq and linuxproc is your ProctrackType, your jobs can legitimately run so fast that Accounting doesn't get a chance to poll for job info 😭. Result? Not all of your accounting info will be there. Speed demon problems fr.

• --cpus-per-gpu=<ncpus> — bestie, this asks for ncpus processors to be allocated for every GPU you get. Any steps that inherit this value will imply --exact. And heads up: this is not compatible with the --cpus-per-task option, so don't try to run both at once 💅.
• -c, --cpus-per-task=<ncpus> — a little heads-up to Slurm that the job steps coming up are gonna need ncpus processors per task. Default behavior? Slurm allocates one processor per task, no extra ✨.

Okay so picture this: you've got an app with 4 tasks, and each one needs 3 processors. Your cluster is made of quad-processor nodes, and if you just ask for 12 processors flat out, the controller might hand you only 3 nodes. But the moment you use --cpus-per-task=3, the controller knows every task needs 3 processors on the same node — so it'll grant you an allocation of 4 nodes, one for each of the 4 tasks. King behavior honestly 👑.

• --deadline=<OPT> — cancels/removes the job if there's no possible ending before this deadline (start > (deadline - time[-min])). Default? No deadline at all. But lowkey a catch: if neither DefaultTime nor MaxTime are configured on the partition your job is in, then the job has to specify some form of time limit (--time[-min]) for a deadline to even work 🙈.

Valid time formats, get into it:

HH:MM[:SS] [AM|PM]
MMDD[YY] or MM/DD[/YY] or MM.DD[.YY]
MM/DD[/YY]-HH:MM[:SS]
YYYY-MM-DD[THH:MM[:SS]]
now[+count[seconds(default)|minutes|hours|days|weeks]]
midnight, elevenses (11 AM), noon, fika (3 PM), teatime (4 PM),
or tomorrow

You can stack one or more time strings (like 'tomorrow18:00') 🫶. And if they conflict? The last one silently wins. No drama, no warning, it just takes precedence.

• --delay-boot=<minutes> — okay so this says: don't reboot nodes to satisfy this job's feature specification if the job's only been eligible to run for less than this time period. If the job has waited less than the specified period, it'll only use nodes that already have the features you asked for 🦋. The argument is in units of minutes. A default can be set by your sysadmin using the delay_boot option of the SchedulerParameters configuration parameter in the slurm.conf file — otherwise the default is zero (no delay, bestie, zero patience).
• -d, --dependency=<dependency_list> — this defers the start of your job until the specified dependencies have been satisfied. Once a dependency is satisfied, it's removed from the job ✨. The <dependency_list> looks like <type:job_id[:job_id][,type:job_id[:job_id]]> or <type:job_id[:job_id][?type:job_id[:job_id]]>. All dependencies must be satisfied if you use the "," separator. Any dependency may be satisfied if you use the "?" separator. Only one separator may be used, periodt. For example:

-d afterok:20:21,afterany:23

means the job can only run after a 0 return code from jobs 20 and 21 AND completion of job 23. But then:

-d afterok:20:21?afterany:23

means any of the conditions (afterok:20 OR afterok:21 OR afterany:23) is enough to release the job 😌.

Many jobs can share the same dependency, and those jobs can even belong to different users. The value can be changed after job submission using the scontrol command. Dependencies on remote jobs are allowed in a federation. And big warning: once a job dependency fails because of the termination state of a preceding job, the dependent job will never be run — even if the preceding job is requeued and ends up with a different termination state in a later execution 💀.

Here are the dependency types, iykyk:

• after:job_id[[+time][:jobid[+time]...]] — after the specified jobs start or are cancelled, plus time in minutes from that start or cancellation, this job can begin execution. If no time is given, there's no delay after start or cancellation 🕐.
• afterany:job_id[:jobid...] — this job can begin execution after the specified jobs have terminated. This is the default dependency type, the vanilla of dependencies 🍦.
• afterburstbuffer:job_id[:jobid...] — this job can begin execution after the specified jobs have terminated and any associated burst buffer stage out operations have completed. Patient queen behavior.
• aftercorr:job_id[:jobid...] — a task of this job array can begin execution after the corresponding task ID in the specified job has completed successfully (ran to completion with an exit code of zero). If the specified job isn't an array, this is treated the same as afterok 🎀.
• afternotok:job_id[:jobid...] — this job can begin execution after the specified jobs have terminated in some failed state (non-zero exit code, node failure, timed out, etc.). This job must be submitted while the specified job is still active, or within MinJobAge seconds after the specified job has ended. If the dependent job id isn't found and is on the same cluster as the job submission, the job is rejected. If the dependent job id isn't found and is on a different cluster from the job submission, the dependency is marked as failed 😭.
• afterok:job_id[:jobid...] — this job can begin execution after the specified jobs have successfully executed (ran to completion with an exit code of zero). Same rules: this job must be submitted while the specified job is still active, or within MinJobAge seconds after the specified job has ended. If the dependent job id isn't found and is on the same cluster as the job submission, the job is rejected. If the dependent job id isn't found and is on a different cluster from the job submission, the dependency is marked as failed 🌸.
• singleton — this job can begin execution after any previously launched jobs sharing the same job name and user have terminated. In other words: only one job by that name, owned by that user, can be running or suspended at any point in time. Main character energy, only one of you on stage 👑. In a federation, a singleton dependency must be fulfilled on all clusters unless DependencyParameters=disable_remote_singleton is used in slurm.conf.

• -m, --distribution={*|block|cyclic|arbitrary|plane=<size>}[:{*|block|cyclic|fcyclic}[:{*|block|cyclic|fcyclic}]][,{Pack|NoPack}] — this one's for specifying alternate distribution methods for remote processes 🌀. For job allocation, this sets environment variables that'll be used by subsequent srun requests and also affects which cores get picked for job allocation.

So this option controls the distribution of tasks to the nodes where resources have been allocated, and the distribution of those resources to tasks for binding (task affinity). Let's break it down: the first distribution method (before the first ":") controls the distribution of tasks to nodes. The second distribution method (after the first ":") controls the distribution of allocated CPUs across sockets for binding to tasks. The third distribution method (after the second ":") controls the distribution of allocated CPUs across cores for binding to tasks. The second and third distributions only apply if task affinity is enabled. And the third distribution is only supported if the task/cgroup plugin is configured. The default value for each distribution type is specified by * ⭐.

Lowkey important note: with select/cons_tres, the number of CPUs allocated to each socket and node may be different. Check out https://slurm.schedmd.com/mc_support.html for more on resource allocation, distribution of tasks to nodes, and binding of tasks to CPUs 🫶.

First distribution method (distribution of tasks across nodes):

• * — use the default method for distributing tasks to nodes (block). If a topology plugin is configured and the user doesn't specify a distribution method, tasks are distributed following topology order rather than node index order. See CR_NO_DIST_TOPO_BLOCK in SelectTypeParameters to disable this behavior 🌈.
• block — the block distribution method distributes tasks to a node such that consecutive tasks share a node. Example time: consider an allocation of three nodes, each with two cpus. A four-task block distribution request will distribute those tasks to the nodes with tasks one and two on the first node, task three on the second node, and task four on the third node. Block distribution is the default behavior if the number of tasks exceeds the number of allocated nodes 💖.
• cyclic — the cyclic distribution method distributes tasks to a node such that consecutive tasks are distributed over consecutive nodes (round-robin style, we love a rotation 🔄). Example: consider an allocation of three nodes, each with two cpus. A four-task cyclic distribution request will distribute those tasks to the nodes with tasks one and four on the first node, task two on the second node, and task three on the third node. Note that when SelectType is select/cons_tres, the same number of CPUs may not be allocated on each node. Task distribution will be round-robin among all the nodes with CPUs yet to be assigned to tasks. Cyclic distribution is the default behavior if the number of tasks is no larger than the number of allocated nodes ✨.

• plane — tasks get handed out in blocks of size <size>, and you gotta give the size or set SLURM_DIST_PLANESIZE or it's not happening. The number of tasks spread onto each node is the same as with cyclic distribution, but the taskids handed to each node depend on the plane size. Heads up: you can't stack extra distribution specs on top of this one. Want the full glow-up with examples and diagrams? Go peep https://slurm.schedmd.com/mc_support.html and https://slurm.schedmd.com/dist_plane.html 💅
• arbitrary — this method allocates processes in-order, exactly as they're listed in the file named by the SLURM_HOSTFILE environment variable. If that variable is set, it overrides any other method you specified. If it's not set, the method defaults to block. The hostfile needs to contain at minimum the number of hosts requested, and they have to be one per line or comma separated. If you specify a task count (-n, --ntasks=<number>), your tasks will be laid out on the nodes in the order of the file.

NOTE: arbitrary distribution is only supported with the generic allocation method — it cannot be used with features like topology, CR_LLN, CR_Pack_Nodes, or pack_serial_at_end. It's meant primarily to control a job step's task layout inside an existing job allocation for the srun command.

NOTE: if you give the number of tasks and also give a list of requested nodes, the number of nodes used from that list gets reduced to match the number of tasks — but only if the list has more nodes than you have tasks. math is mathing ✨

🧵 Second distribution method (distribution of CPUs across sockets for binding)

• * — just use the default method for distributing CPUs across sockets (cyclic). main character behavior, honestly.
• block — this one distributes allocated CPUs consecutively from the same socket for binding to tasks, and only then moves on to the next consecutive socket.
• cyclic — distributes allocated CPUs for binding to a given task consecutively from the same socket, then from the next consecutive socket for the next task, round-robin style across all the sockets. Tasks that need more than one CPU get all of those CPUs on a single socket if it's possible.

NOTE: on nodes with hyper-threading enabled, a task that isn't requesting full cores may get distributed across sockets. You can dodge that by specifying --ntasks-per-core=1, which forces tasks to allocate full cores. no crumbs left behind 🫶

• fcyclic — distributes allocated CPUs for binding to tasks from consecutive sockets in a round-robin fashion across the sockets. Tasks requiring more than one CPU get each of those CPUs allocated in a cyclic fashion across sockets.

🌈 Third distribution method (distribution of CPUs across cores for binding)

• * — use the default method for distributing CPUs across cores, which is inherited from the second distribution method. it's giving hereditary 💜
• block — distributes allocated CPUs consecutively from the same core for binding to tasks, before using the next consecutive core.
• cyclic — distributes allocated CPUs for binding to a given task consecutively from the same core, and from the next consecutive core for the next task, in a round-robin fashion across cores.
• fcyclic — distributes allocated CPUs for binding to tasks from consecutive cores in a round-robin fashion across the cores.

🎀 Optional control for task distribution over nodes

• Pack — instead of spreading a job step's tasks evenly across its allocated nodes, pack them as tightly as possible onto the nodes. This only applies when the block task distribution method is used. bestie is CLINGY.
• NoPack — instead of packing a job step's tasks as tightly as possible, distribute them evenly. This user option supersedes the SelectTypeParameters CR_Pack_Nodes configuration parameter. personal space respected, periodt.

---

• -x, --exclude=<node_name_list> — explicitly yeet certain nodes out of the resources granted to the job. not on the list, not in the allocation 💅
• --exclusive[={user|mcs|topo}] — the job allocation cannot share nodes with other running jobs (or, with =topo, cannot share a topology segment), or with just other users if you use the =user option, or with the =mcs option. If user/mcsi/topo aren't specified (i.e. the job allocation can't share nodes with other running jobs), the job gets allocated all CPUs and GRES on all nodes in the allocation — but it's only allocated as much memory as it requested. That's on purpose, to support gang scheduling, because suspended jobs still sit in memory. Want all the memory on a node? Use --mem=0. The default shared/exclusive behavior depends on system configuration, and the partition's OverSubscribe option takes precedence over the job's option.

NOTE: since shared GRES (MPS) can't be allocated at the same time as a sharing GRES (GPU), this option only allocates all sharing GRES and no underlying shared GRES.

NOTE: this option is mutually exclusive with --oversubscribe. pick a lane, babe 😤

• --extra=<string> — an arbitrary string, wrapped in single or double quotes if you're using spaces or some special characters. If SchedulerParameters=extra_constraints is enabled, this string is used for node filtering based on the Extra field in each node. cute little custom tag, iykyk ✨
• -B, --extra-node-info=<sockets>[:cores[:threads]] — restrict node selection to nodes with at least the specified number of sockets, cores per socket and/or threads per core.

NOTE: these options do not specify the resource allocation size. every value you give is treated as a minimum. You can use an asterisk (*) as a placeholder so node selection isn't restricted for that type. The individual levels can also be specified in separate options if that's more your vibe:

--sockets-per-node=<sockets>
--cores-per-socket=<cores>
--threads-per-core=<threads>

If the task/affinity plugin is enabled, then specifying an allocation this way also results in subsequently launched tasks being bound to threads if the -B option specifies a thread count — otherwise to cores if a core count is specified, otherwise to sockets. If SelectType is configured to select/cons_tres, it must have a parameter of CR_Core, CR_Core_Memory, CR_Socket, or CR_Socket_Memory for this option to be honored. If it's not specified, scontrol show job will display 'ReqS:C:T=*:*:*'.

NOTE: this option is mutually exclusive with --hint, --threads-per-core and --ntasks-per-core.

NOTE: this option may implicitly set the number of tasks (if -n wasn't specified) as one task per requested thread. sneaky little side effect, we love a multitasker 🦋

• --gpu-bind=[verbose,]<type> — equivalent to --tres-bind=gres/gpu:[verbose,]<type>. See --tres-bind for all the options and the full documentation. same energy, different name 💫
• --gpu-freq=[<type]=value>[,<type=value>][,verbose] — request that the GPUs allocated to the job get configured with specific frequency values. You can use this to independently configure the GPU and its memory frequencies. Once the job is done, the frequencies of all affected GPUs get reset to the highest possible values. In some cases, system power caps may override the values you requested. The field type can be "memory". If type isn't specified, the GPU frequency is implied. The value field can be "low", "medium", "high", "highm1" or a numeric value in megahertz (MHz). If the numeric value you specified isn't possible, a value as close as possible will be used. See below for what those values actually mean. The verbose option causes current GPU frequency information to be logged. Examples of use include "--gpu-freq=medium,memory=high" and "--gpu-freq=450". tuning your GPU like it's a playlist, obsessed 🎧 Supported value definitions:
  • low — the lowest available frequency.
  • medium — attempts to set a frequency in the middle of the available range.
  • high — the highest available frequency.
  • highm1 — (high minus one) selects the next highest available frequency. so close to the ceiling bestie 🎀
• -G, --gpus=[type:]<number> — specify the total number of GPUs required for the job. You can optionally tack on a GPU type specification. For example "--gpus=volta:3". See also the --gpus-per-node, --gpus-per-socket and --gpus-per-task options.

NOTE: the allocation has to contain at least one GPU per node, or one of each GPU type per node if types are used. If different nodes need different GPU types, use heterogeneous jobs. no mixing and matching, that's the rule 🌸

• --gpus-per-node=[type:]<number> — specify the number of GPUs required for the job on each node included in the job's resource allocation. You can optionally supply a GPU type spec. For example "--gpus-per-node=volta:3". Multiple options can be requested in a comma separated list, for example: "--gpus-per-node=volta:3,kepler:1". See also the --gpus, --gpus-per-socket and --gpus-per-task options.

NOTE: this option is mutually exclusive with --gres=gpu. can't have both, bestie, that's not how this works 😭

• --gpus-per-socket=[type:]<number> — how many GPUs you want on each socket in the job's resource allocation, bestie 🎮. You can tack on an optional GPU type if you're feeling picky, like "--gpus-per-socket=volta:3". Wanna ask for a whole mix? Comma-separated list, no cap: "--gpus-per-socket=volta:3,kepler:1". This one requires the job to specify a sockets per node count (--sockets-per-node) — it's giving "you can't sit with us" energy otherwise 💅. Also peep the --gpus, --gpus-per-node and --gpus-per-task options.
• --gpus-per-task=[type:]<number> — same idea but per task spawned in the job's resource allocation ✨. Optional GPU type can be supplied, e.g. "--gpus-per-task=volta:1". Multiple options in a comma separated list, e.g. "--gpus-per-task=volta:3,kepler:1". See also --gpus, --gpus-per-socket and --gpus-per-node. This one requires an explicit task count, like -n, --ntasks or "--gpus=X --gpus-per-task=Y", rather than an ambiguous range of nodes with -N, --nodes — no vague vibes allowed 🙅. It'll implicitly set --tres-bind=gres/gpu:per_task:<gpus_per_task>, or if multiple gpu types are specified, --tres-bind=gres/gpu:per_task:<gpus_per_task_type_sum>. Lowkey you can override that with an explicit --tres-bind=gres/gpu specification 🫶.
• --gres=<list> — a comma-delimited list of generic consumable resources requested per node 🌈. Format for each entry is "name[[:type]:count]". The name is the type of consumable resource (e.g. gpu). The type is an optional classification for the resource (e.g. a100). The count is how many of those resources, defaulting to 1. And the count can flex a suffix: "k" or "K" (multiple of 1024), "m" or "M" (multiple of 1024 x 1024), "g" or "G" (multiple of 1024 x 1024 x 1024), "t" or "T" (multiple of 1024 x 1024 x 1024 x 1024), "p" or "P" (multiple of 1024 x 1024 x 1024 x 1024 x 1024) — unit glow-up fr fr 🧚. The specified resources get allocated to the job on each node. What's actually available is configurable by the system administrator. If you pass "help" as the option argument, a list of available generic consumable resources gets printed and the command exits — free info, slay 💖. Examples of use include "--gres=gpu:2", "--gres=gpu:kepler:2", and "--gres=help".

NOTE: This option is mutually exclusive with --gpus-per-node. Pick a lane bestie 💀

• --gres-flags=<type> — specify generic resource task binding options 🔧. The types are:
  • multiple-tasks-per-sharing — negate one-task-per-sharing. Useful if it's set by default in SelectTypeParameters.
  • disable-binding — negate enforce-binding. Useful if that's set by default in SelectTypeParameters.
  • enforce-binding — the only CPUs available to the job will be those bound to the selected GRES (i.e. the CPUs identified in the gres.conf file will be strictly enforced). Heads up: this option may result in delayed initiation of a job 😭. For example a job requiring two GPUs and one CPU will be delayed until both GPUs on a single socket are available rather than using GPUs bound to separate sockets, however, the application performance may be improved due to improved communication speed. Requires the node to be configured with more than one socket and resource filtering will be performed on a per-socket basis.

NOTE: This option can be set by default in SelectTypeParameters. NOTE: This option is specific to SelectType=cons_tres. NOTE: This option can give undefined results if attempting to enforce binding on multiple gres on multiple sockets — be careful babes 🚨

• one-task-per-sharing — do not allow different tasks in to be allocated shared gres from the same sharing gres.

NOTE: This flag is only enforced if shared gres are requested with --tres-per-task. NOTE: This option can be set by default with SelectTypeParameters=ONE_TASK_PER_SHARING_GRES. NOTE: This option is specific to SelectTypeParameters=MULTIPLE_SHARING_GRES_PJ.

• -h, --help — display help information and exit. The "just tell me what to do" button 🎀.
• --hint=<type> — bind tasks according to application hints 🧠.

NOTE: This option implies specific values for certain related options, which prevents its use with any user-specified values for --ntasks-per-core, --cores-per-socket, --sockets-per-node, --threads-per-core or -B. These conflicting options will override --hint when specified as command line arguments. If a conflicting option is specified as an environment variable, --hint as a command line argument will take precedence. Yes it's a whole hierarchy, iykyk 👑

• compute_bound — select settings for compute bound applications: use all cores in each socket, one thread per core.
• memory_bound — select settings for memory bound applications: use only one core in each socket, one thread per core.
• multithread — use extra threads with in-core multi-threading which can benefit communication intensive applications. Only supported with the task/affinity plugin.
• nomultithread — don't use extra threads with in-core multi-threading; restricts tasks to one thread per core. Only supported with the task/affinity plugin.
• help — show this help message.

• -H, --hold — specify the job is to be submitted in a held state (priority of zero) ✋. Good news: a held job can now be released using scontrol to reset its priority (e.g. "scontrol release <job_id>"). Un-pause that era whenever you're ready 💜.
• -I, --immediate[=<seconds>] — exit if resources are not available within the time period specified ⏰. If no argument is given (seconds defaults to 1), resources must be available immediately for the request to succeed. If defer is configured in SchedulerParameters and seconds=1 the allocation request will fail immediately; defer conflicts and takes precedence over this option. By default, --immediate is off, and the command will block until resources become available. Since this option's argument is optional, for proper parsing the single letter option must be followed immediately with the value and not include a space between them — so "-I60" and not "-I 60". The space is a trap bestie, don't fall for it 🙈.
• -J, --job-name=<jobname> — specify a name for the job allocation 🏷️. The specified name will appear along with the job id number when querying running jobs on the system. The default job name is the name of the "command" specified on the command line. Main character name energy, optional but iconic ⭐.
• -K, --kill-command[=signal] — salloc always runs a user-specified command once the allocation is granted. By default salloc will wait indefinitely for that command to exit 😳😳. If you specify the --kill-command option, salloc will send a signal to your command any time that the Slurm controller tells salloc that its job allocation has been revoked. The job allocation can be revoked for a couple of reasons: someone used scancel to revoke the allocation, or the allocation reached its time limit. If you do not specify a signal name or number and Slurm is configured to signal the spawned command at job termination, the default signal is SIGHUP for interactive and SIGTERM for non-interactive sessions. And again, since this option's argument is optional, for proper parsing the single letter option must be followed immediately with the value and not include a space between them — like "-K1" and not "-K 1" 🫠.
• -L, --licenses=<license>[@db][:count][,license[@db][:count]...] — specification of licenses (or other resources available on all nodes of the cluster) which must be allocated to this job 🎟️. License names can be followed by a colon and count (the default count is one). Multiple licenses can be requested. If they're separated by a comma (, meaning AND), then all requested licenses are required for the job. For example, "--licenses=foo:4,bar". If they're separated by a pipe ('|' meaning OR), then only one of the license requests is required for the job. For example, "--licenses=foo:4|bar". AND and OR cannot both be used — pick your fighter ⚔️. To submit jobs using remote licenses (those served by slurmdbd), specify the name of the server providing the licenses. For example, "--license=nastran@slurmdb:12". If LicenseParameters=RemoteFuzzyMatch is set in slurm.conf, it is possible to omit the server name.

NOTE: When submitting heterogeneous jobs, license requests may only be made on the first component job. For example "salloc -L ansys:2 :". First one carries the group chat, as always 💅

NOTE: If licenses are tracked in AccountingStorageTres and OR is used, ReqTRES will display all requested tres separated by commas. AllocTRES will display only the license that was allocated to the job.

NOTE: When a job requests OR'd licenses, Slurm will attempt to allocate the licenses in the order in which they are requested. This specified order will take precedence even if the rest of requested licenses could be satisfied on a requested reservation. This also applies to backfill planning when SchedulerParameters=bf_licenses is configured. Order matters, periodt 🔥

• --mail-type=<type> — notify user by email when certain event types occur 📬. Valid type values are NONE, BEGIN, END, FAIL, REQUEUE, ALL (equivalent to BEGIN, END, FAIL, INVALID_DEPEND, REQUEUE, and STAGE_OUT), INVALID_DEPEND (dependency never satisfied), STAGE_OUT (burst buffer stage out and teardown completed), TIME_LIMIT, TIME_LIMIT_90 (reached 90 percent of time limit), TIME_LIMIT_80 (reached 80 percent of time limit), and TIME_LIMIT_50 (reached 50 percent of time limit). Multiple type values may be specified in a comma separated list. NONE will suppress all event notifications, ignoring any other values specified — it's the "leave me on read" setting 🙃. By default no email notifications are sent. The user to be notified is indicated with --mail-user.
• --mail-user=<user> — who gets the email notifications for state changes, as defined by --mail-type. It can be a full email address or just a username. If you give a username, Slurm tacks on the value from MailDomain in slurm.conf to build the address for you 🫶. Default? The submitting user. So you're already signed up, bestie.
• --mcs-label=<mcs> — only does anything when a compatible MCSPlugin is enabled, so vibe check your config first. This parameter is either a group that the user belongs to (mcs/group) or an arbitrary label string (mcs/label). In both cases, no label will be assigned by default — it's opt-in only, no cap. More deets live in the MCS documentation: https://slurm.schedmd.com/mcs.html ✨
• --mem=<size>[units] — the real memory you need per node. Default units are mebibytes, and you can switch it up with the suffix [K|M|G|T]. Default value is DefMemPerNode and the max is MaxMemPerNode — if configured, you can peek at both with the scontrol show config command 👀. This one's generally the move when whole nodes are getting allocated to jobs (SelectType=select/linear). Also go see --mem-per-cpu and --mem-per-gpu. Oh and heads up: --mem, --mem-per-cpu and --mem-per-gpu are mutually exclusive — pick one, babe. If --mem, --mem-per-cpu or --mem-per-gpu are given as command line arguments, they take precedence over the environment. Periodt.

NOTE: a memory size specification of zero is treated as a special case and grants the job access to all of the memory on each node. Zero = eat everything, basically 💀

NOTE: the memory used by each slurmstepd process is included in the job's total memory usage. It typically consumes between 20MiB and 200MiB, though this can vary depending on system configuration and any loaded plugins. Sneaky little overhead, ngl.

NOTE: memory requests will not be strictly enforced unless Slurm is configured to use an enforcement mechanism. See ConstrainRAMSpace in the cgroup.conf(5) man page and OverMemoryKill in the slurm.conf(5) man page for more details. Don't say I didn't warn you 🙈

• --mem-bind=[{quiet|verbose},]<type> — binds tasks to memory. Only works when the task/affinity plugin is enabled and the NUMA memory functions are available, so like… conditional slay. Note that the resolution of CPU and memory binding may differ on some architectures — for example, CPU binding may be performed at the level of the cores within a processor while memory binding will be performed at the level of nodes, where the definition of "nodes" may differ from system to system 🌀. By default no memory binding is performed; any task using any CPU can use any memory. This option is typically used to ensure that each task is bound to the memory closest to its assigned CPU. The use of any type other than "none" or "local" is not recommended — the other options are giving risky, lowkey not worth it.

NOTE: to have Slurm always report on the selected memory binding for all commands executed in a shell, you can enable verbose mode by setting the SLURM_MEM_BIND environment variable value to "verbose".

The following informational environment variables are set when --mem-bind is in use:

SLURM_MEM_BIND_LIST
SLURM_MEM_BIND_PREFER
SLURM_MEM_BIND_TYPE
SLURM_MEM_BIND_VERBOSE

See the ENVIRONMENT VARIABLES section for a more detailed description of the individual SLURM_MEM_BIND* variables 💜

Supported options include:

• help — show this help message
• local — use memory local to the processor in use
• map_mem:<list> — bind by setting memory masks on tasks (or ranks) as specified, where <list> is <numa_id_for_task_0>,<numa_id_for_task_1>,... The mapping is specified for a node and identical mapping is applied to the tasks on every node (i.e. the lowest task ID on each node is mapped to the first ID specified in the list, etc.). NUMA IDs are interpreted as decimal values unless they are preceded with '0x', in which case they're interpreted as hexadecimal values. If the number of tasks (or ranks) exceeds the number of elements in this list, elements in the list will be reused as needed starting from the beginning of the list. To simplify support for large task counts, the lists may follow a map with an asterisk and repetition count — for example "map_mem:0x0f*4,0xf0*4" ✨. For predictable binding results, all CPUs for each node in the job should be allocated to the job.
• mask_mem:<list> — bind by setting memory masks on tasks (or ranks) as specified, where <list> is <numa_mask_for_task_0>,<numa_mask_for_task_1>,... The mapping is specified for a node and identical mapping is applied to the tasks on every node (i.e. the lowest task ID on each node is mapped to the first mask specified in the list, etc.). NUMA masks are always interpreted as hexadecimal values. Note that masks must be preceded with a '0x' if they don't begin with [0-9] so they're seen as numerical values. If the number of tasks (or ranks) exceeds the number of elements in this list, elements in the list will be reused as needed starting from the beginning of the list. To simplify support for large task counts, the lists may follow a mask with an asterisk and repetition count — for example "mask_mem:0*4,1*4". For predictable binding results, all CPUs for each node in the job should be allocated to the job.
• no[ne] — don't bind tasks to memory (default)
• p[refer] — prefer use of first specified NUMA node, but permit use of other available NUMA nodes
• q[uiet] — quietly bind before task runs (default)
• rank — bind by task rank (not recommended)
• v[erbose] — verbosely report binding before task runs

• --mem-per-cpu=<size>[units] — minimum memory required per usable allocated CPU. Default units are mebibytes, and yes you can suffix it with [K|M|G|T]. The default value is DefMemPerCPU and the maximum value is MaxMemPerCPU (see the exception below, bestie, it's coming). If configured, both parameters can be seen using the scontrol show config command 🔍. Note that if the job's --mem-per-cpu value exceeds the configured MaxMemPerCPU, then the user's limit will be treated as a memory limit per task; --mem-per-cpu will be reduced to a value no larger than MaxMemPerCPU; --cpus-per-task will be set and the value of --cpus-per-task multiplied by the new --mem-per-cpu value will equal the original --mem-per-cpu value specified by the user. Math! She does math 💅. This parameter would generally be used if individual processors are allocated to jobs (SelectType=select/cons_tres). If resources are allocated by core, socket, or whole nodes, then the number of CPUs allocated to a job may be higher than the task count and the value of --mem-per-cpu should be adjusted accordingly. Also see --mem and --mem-per-gpu. And remember: --mem, --mem-per-cpu and --mem-per-gpu are mutually exclusive.

NOTE: if the final amount of memory requested by a job can't be satisfied by any of the nodes configured in the partition, the job will be rejected. This could happen if --mem-per-cpu is used with the --exclusive option for a job allocation and --mem-per-cpu times the number of CPUs on a node is greater than the total memory of that node 😭

NOTE: this applies to usable allocated CPUs in a job allocation. This is important when more than one thread per core is configured. If a job requests --threads-per-core with fewer threads on a core than exist on the core (or --hint=nomultithread which implies --threads-per-core=1), the job will be unable to use those extra threads on the core and those threads will not be included in the memory per CPU calculation. But if the job has access to all threads on the core, those threads will be included in the memory per CPU calculation even if the job did not explicitly request those threads. It's giving "you get what you get" energy.

In the following examples, each core has two threads.

In this first example, two tasks can run on separate hyperthreads in the same core because --threads-per-core is not used. The third task uses both threads of the second core. The allocated memory per cpu includes all threads:

$ salloc -n3 --mem-per-cpu=100
salloc: Granted job allocation 17199
$ sacct -j $SLURM_JOB_ID -X -o jobid%7,reqtres%35,alloctres%35
  JobID                             ReqTRES                           AllocTRES
------- ----------------------------------- -----------------------------------
  17199     billing=3,cpu=3,mem=300M,node=1     billing=4,cpu=4,mem=400M,node=1

In this second example, because of --threads-per-core=1, each task is allocated an entire core but is only able to use one thread per core. Allocated CPUs includes all threads on each core. However, allocated memory per cpu includes only the usable thread in each core.

$ salloc -n3 --mem-per-cpu=100 --threads-per-core=1
salloc: Granted job allocation 17200
$ sacct -j $SLURM_JOB_ID -X -o jobid%7,reqtres%35,alloctres%35
  JobID                             ReqTRES                           AllocTRES
------- ----------------------------------- -----------------------------------
  17200     billing=3,cpu=3,mem=300M,node=1     billing=6,cpu=6,mem=300M,node=1

See how the numbers shifted? That's the whole point, icon behavior 🎀

• --mem-per-gpu=<size>[units] — minimum memory required per allocated GPU. Default units are mebibytes, suffixes [K|M|G|T] apply like everywhere else. Default value is DefMemPerGPU and it's available on both a global and per partition basis 🌈. If configured, the parameters can be seen using the scontrol show config and scontrol show partition commands. Also see --mem. And the classic reminder: --mem, --mem-per-cpu and --mem-per-gpu are mutually exclusive — one of them, not all three, bestie 💀
• --mem-update=<margin>@<delay> — bestie, this one's the memory glow-down 💅 After your job has been running for <delay>, Slurm automatically shrinks the job's memory limit. The new limit = the job's current memory usage + <margin> percent. So like, <margin> is a percentage (20 means 20% above current usage, iykyk). And <delay> is literally how long after job start to go sample the memory usage and apply the reduction.

Accepted time formats for <delay> — memorize these, they're the vibe check:

minutes
minutes:seconds
hours:minutes:seconds
days-hours
days-hours:minutes
days-hours:minutes:seconds

⚠️ Important: the memory limit can only be reduced, never increased. If the newly calculated limit isn't actually lower than the current one, nothing happens, no change, periodt. And there's a floor of 10% of the original memory request enforced so nobody reduces it to some unreasonably tiny value (we don't do that here 🙅‍♀️).

Also — memory usage is measured as RSS (resident physical memory) only. Swap usage is not included. When ConstrainSwapSpace is configured, the swap limit gets reduced proportionally along with the memory limit, slay.

This works with both --mem and --mem-per-cpu jobs ✨ For --mem-per-cpu jobs, the new limit is applied as a per-node value, compared against the highest per-node allocation across all nodes.

This is a one-shot operation — the reduction is applied at most once per job run. On job requeue, the auto-reduction may trigger again. And heads up: this option requires SlurmctldParameters=enable_stepmgr and JobAcctGatherType=jobacct_gather/cgroup to be configured. No config, no mem-update, no cap 🧢

• --mincpus=<n> — specify a minimum number of logical cpus/processors per node. That's it, that's the whole thing 🎀
• --network=<type> — give info about the switch or network. The interpretation of type is system dependent, so like, it depends on your setup, bestie.

This option is available on systems with HPE Slingshot networks. It can be used to request a job VNI (for communication between job steps in a job). It also can override the default network resources allocated for the job step. Multiple values may be specified in a comma-separated list 🦋

• tcs=<class1>[:<class2>]... — set of traffic classes to configure for applications. Supported traffic classes are DEDICATED_ACCESS, LOW_LATENCY, BULK_DATA, and BEST_EFFORT. They may also be specified as TC_DEDICATED_ACCESS, TC_LOW_LATENCY, TC_BULK_DATA, and TC_BEST_EFFORT.
• no_vni — don't allocate any VNIs for this job (even if multi-node). Just none. Zero. Nada.
• job_vni — allocate a job VNI for this job.
• single_node_vni — allocate a job VNI for this job, even if it's a single-node job.
• adjust_limits — if set, slurmd will set an upper bound on network resource reservations by taking the per-NIC maximum resource quantity and subtracting the reserved or used values (whichever is higher) for any system network services; this is the default 👑
• no_adjust_limits — if set, slurmd will calculate network resource reservations based only upon the per-resource configuration default and number of tasks in the application; it will not set an upper bound on those reservation requests based on resource usage of already-existing system network services. Setting this means more application launches could fail based on network resource exhaustion — but if your application absolutely needs a certain amount of resources to function, this option will ensure that. Risk it for the biscuit 🍓
• disable_rdzv_get — disable rendezvous gets in Slingshot NICs, which can improve performance for certain applications.
• nic_distribution_count=<val> — the number of NICs the user will evenly distribute their tasks over. Defaults to the number of NICs on each node.
• def_<rsrc>=<val> — per-CPU reserved allocation for this resource.
• res_<rsrc>=<val> — per-node reserved allocation for this resource. If set, overrides the per-CPU allocation.
• max_<rsrc>=<val> — maximum per-node limit for this resource.
• depth=<depth> — multiplier for per-CPU resource allocation. Default is the number of reserved CPUs on the node.

The resources you may request are:

• txqs — Transmit command queues. The default is 2 per-CPU, maximum 1024 per-node.
• tgqs — Target command queues. The default is 1 per-CPU, maximum 512 per-node.
• eqs — Event queues. The default is 2 per-CPU, maximum 2047 per-node.
• cts — Counters. The default is 1 per-CPU, maximum 2047 per-node.
• tles — Trigger list entries. The default is 1 per-CPU, maximum 2048 per-node.
• ptes — Portable table entries. The default is 6 per-CPU, maximum 2048 per-node.
• les — List entries. The default is 16 per-CPU, maximum 16384 per-node.
• acs — Addressing contexts. The default is 2 per-CPU, maximum 1022 per-node.

On systems configured with SwitchType=switch/nvidia_imex, the following options are supported:

• unique-channel-per-segment — instead of one channel for the entire job, allocate one channel per segment in the job. This only takes effect when topology/block is configured.

• --nice[=adjustment] — run the job with an adjusted scheduling priority within Slurm. With no adjustment value the scheduling priority is decreased by 100. A negative nice value increases the priority, otherwise it decreases it. The adjustment range is +/- 2147483645 😭 (yes that number is real). Only privileged users can specify a negative adjustment.
• --no-bell — silence salloc's use of the terminal bell 🔕 Also see the option --bell.
• -k, --no-kill[=off] — do not automatically terminate a job if one of the nodes it has been allocated fails. The user will assume the responsibilities for fault-tolerance should a node fail. The job allocation will not be revoked, so you may launch new job steps on the remaining nodes in your allocation.

🫶 Heads up bestie: this option does not set the SLURM_NO_KILL environment variable. So when a node fails, steps running on that node will be killed unless SLURM_NO_KILL was explicitly set, or srun calls within the job allocation explicitly requested --no-kill.

Specify an optional argument of off to disable the effect of the SALLOC_NO_KILL environment variable.

By default Slurm terminates the entire job allocation if any node fails in its range of allocated nodes.

• --no-shell — immediately exit after allocating resources, without running a command. But plot twist: the Slurm job will still be created and will remain active and will own the allocated resources as long as it's active. You will have a Slurm job id with no associated processes or tasks 💀 You can submit srun commands against this resource allocation if you specify the --jobid= option with the job id of this Slurm job. Or, this can be used to temporarily reserve a set of resources so other jobs can't use them for some period of time. (Note that the Slurm job is subject to the normal constraints on jobs, including time limits, so eventually the job will terminate and the resources will be freed — or you can terminate the job manually using the scancel command.)
• -F, --nodefile=<node_file> — much like --nodelist, but the list lives in a file named node_file. The node names in the list may also span multiple lines in the file. Duplicate node names in the file will be ignored. The order of the node names in the list is not important; the node names will be sorted by Slurm 🌸
• -w, --nodelist=<node_name_list> — request a specific list of nodes. The job will contain as many of these nodes as possible based on the resource requirements, delaying execution as needed to wait for resources to become available ⭐

If you specify a minimum node or processor count larger than can be satisfied by the supplied node list, additional resources will be allocated on other nodes as needed (unless --segment is also used). Conversely, a lower node or processor count may only require a subset of the supplied node list.

The list may be specified as a comma-separated list of nodes, a range of nodes (e.g. node[1-5,7,...]), or a filename. The nodes list will be assumed to be a filename if it contains a "/" character. Duplicate node names in the list will be ignored. The order of the node names in the list is not important; the node names will be sorted by Slurm 💜

• -N, --nodes=<minnodes>[-maxnodes]|<size_string> — okay so this is THE node-count option 💅 you're asking for a minimum of minnodes nodes for this job, and you can also throw in a maxnodes cap if you're feeling generous. Only gave one number? Then it's both the min and the max, periodt. You can also hand it a size_string, which is a whole vibe for telling Slurm which node values to use — multiple values go in a comma-separated list, or you can do the step-function thing with a suffix that has a colon and number values separated by a -. Like, --nodes=1-15:4 is literally equivalent to --nodes=1,5,9,13 ✨

The partition's node limits supersede the job's, no cap. If the job's node limits land outside the range its partition allows, the job just chills in a PENDING state — that's on purpose, so it can maybe run later once the partition limit gets changed 🫶. But if the job node limit exceeds the number of nodes configured in the partition? Rejected. Bye 💀

Note that the environment variable SLURM_JOB_NUM_NODES is set to the count of nodes actually allocated to the job — go see the ENVIRONMENT VARIABLES section for the tea. And if you don't pass -N at all, the default is to allocate enough nodes to satisfy whatever resources you asked for via per-job spec options, e.g. -n, -c, --ntasks-per-node and --gpus.

When -N is omitted, the derived count is --ntasks, or --ntasks divided by --ntasks-per-node and rounded up if that option is given. Without --ntasks-per-node the count is only an upper bound, so MaxNodes reduces it rather than rejecting the job. With --ntasks-per-node the count is the number of nodes the job requires, and MaxNodes rejects it as it would an explicit count. Either way, a job whose tasks do not fit is rejected or left in a PENDING state.

The derived count is not lower than the partition's MinNodes at submit time. Raising the partition's MinNodes later does not re-derive it, so a job whose count is below the new value stays PENDING until its node count is set explicitly 😤

The job will be allocated as many nodes as possible within the range specified and without delaying the initiation of the job. And the node count spec can include a numeric value followed by a suffix of "k" (multiplies numeric value by 1,024) or "m" (multiplies numeric value by 1,048,576) 🔥

NOTE: This option cannot be used in with arbitrary distribution. (yes, that's the original wording, don't come for me 😭)

• -n, --ntasks=<number> — bestie, salloc does not launch tasks, okay? It requests an allocation of resources and executed some command. This option just tells the Slurm controller that job steps run within this allocation will launch a maximum of number tasks, and enough resources get allocated to make that happen 👑. If --nodes isn't used, this option is what implicitly determines the node count. Default is one task per node — but heads up, --cpus-per-task will change that default 🙈
• --ntasks-per-core=<ntasks> — request the maximum ntasks be invoked on each core. Meant to be used with --ntasks. It's related to --ntasks-per-node except at the core level instead of the node level 🎀 This option will be inhertited by srun (yes, "inhertited", that's the original spelling, she contains multitudes). Slurm may allocate more cpus than what was requested in order to respect this option.

NOTE: This option is not supported when using SelectType=select/linear. This value can not be greater than --threads-per-core.

• --ntasks-per-gpu=<ntasks> — request that there are ntasks tasks invoked for every GPU. This one works two ways, pick your fighter: 1) you also specify --ntasks, in which case a type-less GPU specification gets automatically determined to satisfy --ntasks-per-gpu, or 2) you specify the GPUs you want (e.g. via --gpus or --gres) without specifying --ntasks, and the total task count is automatically determined 💫 The number of CPUs needed will be automatically increased if necessary to allow for any calculated task count. This option implicitly sets --tres-bind=gres/gpu:single:<ntasks>, but that can be overridden with an explicit --tres-bind=gres/gpu spec.

Ngl the incompatibility list is long: not compatible with a node range (i.e. -N<minnodes-maxnodes>), not compatible with --gpus-per-task, --gpus-per-socket, or --ntasks-per-node. And it's not supported unless SelectType=cons_tres is configured (either directly or indirectly on Cray systems) 😭

• --ntasks-per-node=<ntasks> — request that ntasks be invoked on each node 🌸 If used with the --ntasks option, --ntasks takes precedence and --ntasks-per-node gets treated as a maximum count of tasks per node. Meant to be used with the --nodes option. If --nodes isn't used and --ntasks is given, this option determines the job's node count — see -N, --nodes. It's related to --cpus-per-task=ncpus, but it doesn't require knowing the actual number of cpus on each node. Sometimes it's just more convenient to request that no more than a specific number of tasks be invoked on each node 🦋

Examples of that, since you asked: submitting a hybrid MPI/OpenMP app where only one MPI "task/rank" should be assigned to each node while letting the OpenMP portion use all the parallelism present in the node — or submitting a single setup/cleanup/monitoring job to each node of a pre-existing allocation as one step in a larger job script. Iconic use cases honestly ⭐

• --ntasks-per-socket=<ntasks> — request the maximum ntasks be invoked on each socket. Meant to be used with --ntasks. Related to --ntasks-per-node except at the socket level instead of the node level 💜

NOTE: This option is not supported when using SelectType=select/linear.

• --oom-kill-step[={0|1}] — whether to kill the entire step if an OOM event is detected in any task of a step 💀 This overwrites the "OOMKillStep" setting in TaskPluginParam from slurm.conf. When unset it will use the setting in slurm.conf. When set, a value of "0" disables killing the entire step, while "1" enables it. This applies to the entire allocation except for the external step. Default is "1" (enabled) when the option is found with no value 🔥
• -O, --overcommit — overcommit resources, bestie, we're doing it 🌈

When applied to a job allocation (not including jobs requesting exclusive access to the nodes) the resources are allocated as if only one task per node is requested. That means the requested number of cpus per task (-c, --cpus-per-task) get allocated per node rather than being multiplied by the number of tasks. Options used to specify the number of tasks per node, socket, core, etc. are ignored. Ignored!! 🙈

When applied to job step allocations (the srun command when executed within an existing job allocation), this option can be used to launch more than one task per CPU. Normally srun won't allocate more than one process per CPU — by specifying --overcommit you are explicitly allowing more than one process per CPU 😳 However no more than MAX_TASKS_PER_NODE tasks are permitted to execute per node.

NOTE: MAX_TASKS_PER_NODE is defined in the file slurm.h and is not a variable, it is set at Slurm build time.

• -s, --oversubscribe — the job allocation can over-subscribe resources with other running jobs 💫 The resources to be over-subscribed can be nodes, sockets, cores, and/or hyperthreads depending upon configuration. The default over-subscribe behavior depends on system configuration, and the partition's OverSubscribe option takes precedence over the job's option. This option may result in the allocation being granted sooner than if --oversubscribe wasn't set and allow higher system utilization — but application performance will likely suffer due to competition for resources 😭 Also see the --exclusive option.

NOTE: This option is mutually exclusive with --exclusive.

• -p, --partition=<partition_names> — request a specific partition for the resource allocation 🎀 If not specified, the default behavior is to let the slurm controller pick the default partition as designated by the system administrator. If the job can use more than one partition, list their names in a comma separate list and the one offering earliest initiation will be used — with no regard given to the partition name ordering (although higher priority partitions will be considered first) 👑 When the job is initiated, the name of the partition used gets placed first in the job record partition string.
• --prefer=<list> — nodes can have features assigned to them by the Slurm administrator 🌸 Users can specify which of these features are desired but not required by their job using the prefer option. This option operates independently from --constraint and will override whatever is set there if possible. When scheduling, the features in --prefer are tried first. If a node set isn't available with those features then --constraint is attempted. See --constraint for more info — this option behaves the same way ✨
• --priority=<value> — request a specific job priority. May be subject to configuration specific constraints. value should either be a numeric value or "TOP" (for highest possible value). Only Slurm operators and administrators can set the priority of a job 🫶 (so if that's not you, don't even try it bestie 💅)
• --profile={all|none|<type>[,<type>...]} — enables detailed data collection by the acct_gather_profile plugin 🦋 Detailed data are typically time-series that are stored in an HDF5 file for the job or an InfluxDB database depending on the configured plugin.
  • All — all data types are collected. (Cannot be combined with other values.)
  • None — no data types are collected. This is the default. (Cannot be combined with other values.)

Valid type values are:

• Energy — Energy data is collected. ⚡

• Task — Task (I/O, Memory, ...) data is collected.
• Lustre — Lustre data is collected.
• Network — Network (InfiniBand) data is collected.

---

• -q, --qos=<qos> — bestie you're literally requesting a quality of service for the job here, or a whole comma separated list of QOS 🌸. If you go the list route, it gets ordered by priority of the QOS you gave, and the first one is the highest priority, periodt. QOS values can be defined for each user/cluster/account association over in the Slurm database. Heads up tho — users get limited to their association's defined set of qos's when the Slurm configuration parameter AccountingStorageEnforce includes "qos" in its definition.
• -Q, --quiet — shush era 🤫. Suppress informational messages from salloc. Errors will still be displayed (obvi, you gotta know when something's wrong 💅).
• --reboot — force the allocated nodes to reboot before starting the job. This is only supported with some system configurations and will otherwise be silently ignored (rude but okay 🙈). Only root, SlurmUser or admins can reboot nodes, so like… know your place bestie.
• --reservation=<reservation_names> — allocate resources for the job from the named reservation. If the job can use more than one reservation, specify their names in a comma separate list and the one offering earliest initiation wins. Each reservation will be considered in the order it was requested. All reservations will be listed in scontrol/squeue through the life of the job. In accounting the first reservation will be seen, and after the job starts the reservation used will replace it ✨.
• --resources=<resource_names> — specification of hierarchical resources which must be allocated to this job. Resource names can be followed by a colon and count (the default count is one). Multiple resources in Mode 1 and Mode 2 can be requested in a comma separated list, but only one of Mode 3 can be requested. Example time:

--resources=flat:2,natural:1

see https://slurm.schedmd.com/hres.html for more information on use of Hierarchical Resource with Slurm 🦋.

• --resv-ports[=count] — reserve communication ports for this job. Users can specify the number of port they want to reserve. The parameter MpiParams=ports=12000-12999 must be specified in slurm.conf. If the number of reserved ports is zero then no ports are reserved. Used for native Cray's PMI only. This option can only be used if the slurmstepd step management is enabled — see --stepmgr 💜.
• --segment=<segment_size> — when a block or ring topology is used, this defines the size of the segments that will be used to create the job allocation. No requirement would be placed on all segments for a job needing to be placed within the same higher-level block.

NOTE: if the requested node count (--nodes) is larger than the requested segment size, it must also be evenly divisible by the segment size. If all nodes fit within a single segment, this option has no effect. 🫶

NOTE: when used in conjunction with --nodelist=<node_list>, the requested node count must be less than or equal to the total number of unique nodes specified in the --nodelist argument. Requesting more nodes than available unique nodes in the provided --nodelist will result in the job being rejected by slurmctld 😭.

• --signal=[R:]<sig_num>[@sig_time] — when a job is within sig_time seconds of its end time, send it the signal sig_num. Due to the resolution of event handling by Slurm, the signal may be sent up to 60 seconds earlier than specified. sig_num may either be a signal number or name (e.g. "10" or "USR1"). sig_time must have an integer value between 0 and 65535. By default, no signal is sent before the job's end time. If a sig_num is specified without any sig_time, the default time will be 60 seconds. Use the "R:" option to allow this job to overlap with a reservation with MaxStartDelay set. If the "R:" option is used, preemption must be enabled on the system, and if the job is preempted it will be requeued if allowed, otherwise the job will be canceled 💀. To have the signal sent at preemption time see the send_user_signal PreemptParameter.
• --sockets-per-node=<sockets> — restrict node selection to nodes with at least the specified number of sockets. See additional information under -B option above when task/affinity plugin is enabled.

NOTE: this option may implicitly set the number of tasks (if -n was not specified) as one task per requested thread 👑

• --spread-job — spread the job allocation over as many nodes as possible and attempt to evenly distribute tasks across the allocated nodes. This option disables the topology/tree plugin 🌈.
• --spread-segments — prevent nodes within the same base block from being allocated to separate segments within the same block.

NOTE: this option will only work with the topology/block plugin 🎀

• --stepmgr — enable slurmstepd step management per-job if it isn't enabled system wide. This enables job steps to be managed by a single extern slurmstepd associated with the job to manage steps. This is beneficial for jobs that submit many steps inside their allocations, fr fr. PrologFlags=contain must be set ⭐.
• --switches=<count>[@max-time] — when a tree topology is used, this defines the maximum count of leaf switches desired for the job allocation and optionally the maximum time to wait for that number of switches. If Slurm finds an allocation containing more switches than the count specified, the job remains pending until it either finds an allocation with desired switch count or the time limit expires. If there is no switch count limit, there is no delay in starting the job. Acceptable time formats include "minutes", "minutes:seconds", "hours:minutes:seconds", "days-hours", "days-hours:minutes" and "days-hours:minutes:seconds". The job's maximum time delay may be limited by the system administrator using the SchedulerParameters configuration parameter with the max_switch_wait parameter option. On a dragonfly network the only switch count supported is 1 since communication performance will be highest when a job is allocate resources on one leaf switch or more than 2 leaf switches. The default max-time is the max_switch_wait SchedulerParameters 🧚.
• --thread-spec=<num> — count of specialized threads per node reserved by the job for system operations and not used by the application. The application will not use these threads, but will be charged for their allocation. This option can not be used with the --core-spec option 🚫.

NOTE: explicitly setting a job's specialized thread value implicitly sets its --exclusive option, reserving entire nodes for the job 💅

• --threads-per-core=<threads> — restrict node selection to nodes with at least the specified number of threads per core. In task layout, use the specified maximum number of threads per core.

NOTE: "Threads" refers to the number of processing units on each core rather than the number of application tasks to be launched per core 🌸

See additional information under -B option above when task/affinity plugin is enabled.

NOTE: this option may implicitly set the number of tasks (if -n was not specified) as one task per requested thread 💖

• -t, --time=<time> — set a limit on the total run time of the job allocation. If the requested time limit exceeds the partition's time limit, the job will be left in a PENDING state (possibly indefinitely 😭). The default time limit is the partition's default time limit. When the time limit is reached, each task in each job step is sent SIGTERM followed by SIGKILL. The interval between signals is specified by the Slurm configuration parameter KillWait. The OverTimeLimit configuration parameter may permit the job to run longer than scheduled. Time resolution is one minute and second values are rounded up to the next minute ⏱️.

A time limit of zero requests that no time limit be imposed. Acceptable time formats include "minutes", "minutes:seconds", "hours:minutes:seconds", "days-hours", "days-hours:minutes" and "days-hours:minutes:seconds".

• --time-min=<time> — set a minimum time limit on the job allocation. If specified, the job may have its --time limit lowered to a value no lower than --time-min if doing so permits the job to begin execution earlier than otherwise possible 🦋. This is performed by a backfill scheduling algorithm to allocate resources otherwise reserved for higher priority jobs or an upcoming reservation. The lowered limit is set when the job is allocated resources and is not changed afterwards. Acceptable time formats include "minutes", "minutes:seconds", "hours:minutes:seconds", "days-hours", "days-hours:minutes" and "days-hours:minutes:seconds" ✨.
• --tmp=<size>[units] — specify a minimum amount of temporary disk space per node. Default units are mebibytes. Different units can be specified using the suffix [K|M|G|T] 🍓.
• --tres-bind=<tres>:[verbose,]<type>[+<tres>:[verbose,]<type>...] — specify a list of tres with their task binding options. Currently gres are the only supported tres for this options. Specify gres as "gres/<gres_name>" (e.g. gres/gpu) 🔥

Example:

--tres-bind=gres/gpu:verbose,map:0,1,2,3+gres/nic:closest

By default, most tres are not bound to individual tasks.

Supported binding type options for gres:

• closest — bind each task to the gres(s) which are closest. In a NUMA environment, each task may be bound to more than one gres (i.e. all gres in that NUMA environment) 💜.

• map:<list> — bind by setting gres masks on tasks (or ranks) exactly how you spell them out here, where <list> is <gres_id_for_task_0>,<gres_id_for_task_1>,... and all those gres IDs are read as decimal values, no cap.

So like, if the number of tasks (or ranks) is bigger than the number of elements in your list? The elements just get reused as needed, starting over from the beginning of the list 🔁. And to make big task counts not a total nightmare, the lists can follow a map with an asterisk and a repetition count. For example "map:0*4,1*4" — iconic shorthand, love her.

Ok important caveat bestie: if the task/cgroup plugin is in play and ConstrainDevices is set in cgroup.conf, then the gres IDs are zero-based indexes relative to the gres allocated to the job (so the first gres is 0, even if the global ID is 3). Otherwise, the gres IDs are global IDs, and all gres on each node in the job should be allocated for predictable binding results. Don't skip that part, it's giving "your binding will be chaos otherwise" 💀

• mask:<list> — bind by setting gres masks on tasks (or ranks) as specified, where <list> is <gres_mask_for_task_0>,<gres_mask_for_task_1>,... 🎭

The mapping is specified for a node and the identical mapping is applied to the tasks on every node (i.e. the lowest task ID on each node is mapped to the first mask specified in the list, etc.). gres masks are always interpreted as hexadecimal values but can be preceded with an optional 0x. And yeah, same trick as before for large task counts — the lists may follow a map with an asterisk and repetition count. For example "mask:0x0f*4,0xf0*4" ✨

Same caveat as above, fr: if the task/cgroup plugin is used and ConstrainDevices is set in cgroup.conf, then the gres IDs are zero-based indexes relative to the gres allocated to the job (first gres is 0, even if the global ID is 3). Otherwise the gres IDs are global IDs, and all gres on each node in the job should be allocated for predictable binding results.

• none — do not bind tasks to this gres. That's it, that's the whole vibe. It turns off implicit binding from --tres-per-task and --gpus-per-task 🙅‍♀️
• per_task:<gres_per_task> — each task gets bound to the number of gres specified in <gres_per_task>. Tasks are preferentially assigned gres with affinity to cores in their allocation, like in closest, though they will take any gres if the good ones are unavailable 🫶. If no affinity exists at all, the first task gets assigned the first x number of gres on the node, etc. And shared gres will prefer to bind one sharing device per task if possible — generous queen 👑
• single:<tasks_per_gres> — like closest, except each task can only be bound to a single gres, even when it could be bound to multiple gres that are equally close. The gres to bind to is determined by <tasks_per_gres>, where the first <tasks_per_gres> tasks are bound to the first gres available, the second <tasks_per_gres> tasks are bound to the second gres available, etc.

Basically it's a block distribution of tasks onto available gres, where the available gres are determined by the socket affinity of the task and the socket affinity of the gres as specified in gres.conf's Cores parameter 🧩

NOTE: Shared gres binding is currently limited to per_task or none. That's the whole menu, don't come for me 😭

• --tres-per-task=<list> — specifies a comma-delimited list of trackable resources required for the job on each task to be spawned in the job's resource allocation 💖

The format for each entry in the list is "trestype[/tresname]=count". The trestype is the type of trackable resource requested (e.g. cpu, gres, license, etc). The tresname is the name of the trackable resource, as can be seen with sacctmgr show tres. This is required when it exists for tres types such as gres, license, etc. (e.g. gpu, gpu:a100). In order to request a license with this option, the license(s) must be defined in the AccountingStorageTRES parameter of slurm.conf. The count is the number of those resources 🔢

And the count can have a suffix, which is genuinely such a nice touch:

• "k" or "K" — multiple of 1024
• "m" or "M" — multiple of 1024 x 1024
• "g" or "G" — multiple of 1024 x 1024 x 1024
• "t" or "T" — multiple of 1024 x 1024 x 1024 x 1024
• "p" or "P" — multiple of 1024 x 1024 x 1024 x 1024 x 1024

Examples, because examples slay:

--tres-per-task=cpu=4
--tres-per-task=cpu=8,license/ansys=1
--tres-per-task=gres/gpu=1
--tres-per-task=gres/gpu:a100=2

The specified resources will be allocated to the job on each node. The available trackable resources are configurable by the system administrator (so like, if it's not available, that's a them problem, not a you problem 🫠).

NOTE: This option with gres/gpu or gres/shard will implicitly set --tres-bind=gres/[gpu|shard]:per_task:<tres_per_task>, or if multiple gpu types are specified, --tres-bind=gres/gpu:per_task:<gpus_per_task_type_sum>. This can be overridden with an explicit --tres-bind specification. Sneaky little default, but ok 💅

NOTE: Invalid TRES for --tres-per-task include bb,billing,energy,fs,mem,node,pages,vmem. Just don't 💀

• --usage — display a brief help message and exit. Short and sweet 🎀
• --use-min-nodes — if a range of node counts is given, prefer the smaller count. Lowest-maintenance behavior possible ✨
• -v, --verbose — increase the verbosity of salloc's informational messages. Multiple -v's will further increase salloc's verbosity. By default only errors will be displayed. So if you're nosy (same), stack 'em up 🗣️
• -V, --version — display version information and exit. Gotta know what you're running, bestie 🏷️
• --wait-all-nodes=<value> — controls when the execution of the command begins with respect to when nodes are ready for use (i.e. booted) ⏳ By default, the salloc command will return as soon as the allocation is made. This default can be altered using the salloc_wait_nodes option to the SchedulerParameters parameter in the slurm.conf file.
  • 0 — begin execution as soon as allocation can be made. Do not wait for all nodes to be ready for use (i.e. booted). Impatient icon 🏃‍♀️
  • 1 — do not begin execution until all nodes are ready for use. Patient queen, waits for everyone to show up 🧘‍♀️
• --wckey=<wckey> — specify wckey to be used with job. If TrackWCKey=no (default) in the slurm.conf this value is ignored. So like, check your config or you're just yelling into the void 🙈
• --x11[={all|first|last}] — sets up X11 forwarding on "all", "first" or "last" node(s) of the allocation 🖥️✨ This option is only enabled if Slurm was compiled with X11 support and PrologFlags=x11 is defined in the slurm.conf. Default is "all".

NOTE: This option may not work if --clusters is also provided. Such jobs will be accepted (with a warning), but X11 forwarding is not guaranteed to work. You've been warned, don't @ me 😭

---

🔥 PERFORMANCE 🔥

Ok so real talk for a sec 🫶 — executing salloc sends a remote procedure call to slurmctld. If enough calls from salloc or other Slurm client commands that send remote procedure calls to the slurmctld daemon come in at once, it can result in a degradation of performance of the slurmctld daemon, possibly resulting in a denial of service 💀 That's not a vibe. That's the opposite of a vibe.

Do not run salloc or other Slurm client commands that send remote procedure calls to slurmctld from loops in shell scripts or other programs 🚫 Ensure that programs limit calls to salloc to the minimum necessary for the information you are trying to gather. Like, be a considerate main character, the cluster is shared 🌸

✨ INPUT ENVIRONMENT VARIABLES ✨

okay bestie, on startup salloc will read and handle the options set in these environment variables 💖 the majority of these variables are set the same way the options are set, as defined above. for flag options that are defined to expect no argument, the option can be enabled by setting the environment variable without a value (empty or NULL string), the string 'yes', or a non-zero number. any other value for the environment variable will result in the option not being set. there are a couple exceptions to these rules that are noted below 👀

NOTE: Command line options always override environment variables settings.

here’s the whole roster, no skips, no cap 🫶

• SALLOC_ACCOUNT — same as -A, --account
• SALLOC_ACCTG_FREQ — same as --acctg-freq
• SALLOC_BELL — same as --bell
• SALLOC_BURST_BUFFER — same as --bb
• SALLOC_CLUSTERS or SLURM_CLUSTERS — same as --clusters
• SALLOC_CONSTRAINT — same as -C, --constraint
• SALLOC_CONTAINER — same as --container.
• SALLOC_CONTAINER_ID — same as --container-id.
• SALLOC_CONTAINER_TYPE — same as --container-type.
• SALLOC_CORE_SPEC — same as --core-spec
• SALLOC_CPUS_PER_GPU — same as --cpus-per-gpu
• SALLOC_DEBUG — same as -v, --verbose, when set to 1, when set to 2 gives -vv, etc.
• SALLOC_DELAY_BOOT — same as --delay-boot
• SALLOC_EXCLUSIVE — same as --exclusive
• SALLOC_GPU_BIND — same as --gpu-bind
• SALLOC_GPU_FREQ — same as --gpu-freq
• SALLOC_GPUS — same as -G, --gpus
• SALLOC_GPUS_PER_NODE — same as --gpus-per-node
• SALLOC_GPUS_PER_TASK — same as --gpus-per-task
• SALLOC_GRES — same as --gres
• SALLOC_GRES_FLAGS — same as --gres-flags
• SALLOC_HINT or SLURM_HINT — same as --hint
• SALLOC_IMMEDIATE — same as -I, --immediate
• SALLOC_KILL_CMD — same as -K, --kill-command
• SALLOC_MEM_BIND — same as --mem-bind
• SALLOC_MEM_PER_CPU — same as --mem-per-cpu
• SALLOC_MEM_PER_GPU — same as --mem-per-gpu
• SALLOC_MEM_PER_NODE — same as --mem
• SALLOC_NETWORK — same as --network
• SALLOC_NO_BELL — same as --no-bell
• SALLOC_NO_KILL — same as -k, --no-kill
• SALLOC_OVERCOMMIT — same as -O, --overcommit
• SALLOC_PARTITION — same as -p, --partition
• SALLOC_POWER — same as --power
• SALLOC_PROFILE — same as --profile
• SALLOC_QOS — same as --qos
• SALLOC_REQ_SWITCH — when a tree topology is used, this defines the maximum count of switches desired for the job allocation and optionally the maximum time to wait for that number of switches. see --switches.
• SALLOC_RESERVATION — same as --reservation
• SALLOC_SEGMENT_SIZE — same as --segment
• SALLOC_SIGNAL — same as --signal
• SALLOC_SPREAD_JOB — same as --spread-job
• SALLOC_THREAD_SPEC — same as --thread-spec
• SALLOC_THREADS_PER_CORE — same as --threads-per-core
• SALLOC_TIMELIMIT — same as -t, --time
• SALLOC_TRES_BIND — same as --tres-bind
• SALLOC_TRES_PER_TASK — same as --tres-per-task
• SALLOC_USE_MIN_NODES — same as --use-min-nodes
• SALLOC_WAIT_ALL_NODES — same as --wait-all-nodes. must be set to 0 or 1 to disable or enable the option.
• SALLOC_WAIT4SWITCH — max time waiting for requested switches. see --switches
• SALLOC_WCKEY — same as --wckey
• SLURM_CONF — the location of the Slurm configuration file.
• SLURM_DEBUG_FLAGS — specify debug flags for salloc to use. see DebugFlags in the slurm.conf(5) man page for a full list of flags. the environment variable takes precedence over the setting in the slurm.conf.
• SLURM_EXIT_ERROR — specifies the exit code generated when a Slurm error occurs (e.g. invalid options). this can be used by a script to distinguish application exit codes from various Slurm error conditions. also see SLURM_EXIT_IMMEDIATE.
• SLURM_EXIT_IMMEDIATE — specifies the exit code generated when the --immediate option is used and resources are not currently available. this can be used by a script to distinguish application exit codes from various Slurm error conditions. also see SLURM_EXIT_ERROR.

🌈 OUTPUT ENVIRONMENT VARIABLES 🌈

ok so after all that, salloc also gives back — it sets these env vars in the environment of the program it runs. bestie gets gift-wrapped context, periodt 💝

• SLURM_*_HET_GROUP_# — for a heterogeneous job allocation, these get set separately for each component, no cap.
• SLURM_CLUSTER_NAME — the name of the cluster where the job is actually running.
• SLURM_CONTAINER — the OCI Bundle for the job. only set if --container was specified.
• SLURM_CONTAINER_ID — the OCI id for the job. only set if --container-id was specified.
• SLURM_CONTAINER_TYPE — the job container type for the job. only set if --container-type was specified.
• SLURM_CPUS_PER_GPU — how many CPUs you asked for per allocated GPU. only set if the --cpus-per-gpu option is specified.
• SLURM_CPUS_PER_TASK — how many CPUs you asked for per task. only set if either the --cpus-per-task option or the --tres-per-task=cpu=# option is specified. (two ways in, one var out 💅)
• SLURM_DIST_PLANESIZE — the plane distribution size. only set for plane distributions. see -m, --distribution.
• SLURM_DISTRIBUTION — only set if the -m, --distribution option is specified.
• SLURM_GPU_BIND — the requested binding of tasks to GPU. only set if the --gpu-bind option is specified.
• SLURM_GPU_FREQ — the requested GPU frequency. only set if the --gpu-freq option is specified.
• SLURM_GPUS — number of GPUs requested. only set if the -G, --gpus option is specified.
• SLURM_GPUS_PER_NODE — requested GPU count per allocated node. only set if the --gpus-per-node option is specified.
• SLURM_GPUS_PER_SOCKET — requested GPU count per allocated socket. only set if the --gpus-per-socket option is specified.
• SLURM_HET_SIZE — set to the count of components in the heterogeneous job.
• SLURM_JOB_ACCOUNT — the account name associated with the job allocation.
• SLURM_JOB_CPUS_PER_NODE — count of CPUs available to the job on the nodes in the allocation, in the format CPU_count[(xnumber_of_nodes)][,CPU_count [(xnumber_of_nodes)] ...]. for example, SLURM_JOB_CPUS_PER_NODE='72(x2),36' means the first and second nodes (as listed by SLURM_JOB_NODELIST) each have 72 CPUs in the allocation, and the third node has 36. NOTE: the select/linear plugin allocates entire nodes to jobs, so this value is the total count of CPUs on the allocated nodes. the select/cons_tres plugin allocates individual CPUs, so this number is the count of CPUs actually allocated to the job. 👀 different math, same var name, so watch out.
• SLURM_JOB_END_TIME — the UNIX timestamp for a job's projected end time.
• SLURM_JOB_GPUS — the global GPU IDs of the GPUs allocated to this job. these IDs are not relative to any device cgroup, even if devices are constrained with task/cgroup. only set in batch and interactive jobs.
• SLURM_JOB_ID — the ID of the job allocation.
• SLURM_JOB_LICENSES — name and count of any license(s) requested.
• SLURM_JOB_NODELIST — list of nodes allocated to the job.
• SLURM_JOB_NUM_NODES — total number of nodes in the job allocation.
• SLURM_JOB_PARTITION — name of the partition the job is running in.
• SLURM_JOB_QOS — the Quality Of Service (QOS) of the job allocation.
• SLURM_JOB_RESERVATION — the advanced reservation containing the job allocation, if there is one.
• SLURM_JOB_SEGMENT_SIZE — the size of the segments used to create the job allocation. only set if --segment is specified.
• SLURM_JOB_START_TIME — UNIX timestamp for the job's start time.
• SLURM_JOBID — the ID of the job allocation. see SLURM_JOB_ID. included for backwards compatibility. (the og name, still iconic 🫶)
• SLURM_MEM_BIND — set to the value of the --mem-bind option.
• SLURM_MEM_BIND_LIST — set to the bit mask used for memory binding.
• SLURM_MEM_BIND_PREFER — set to "prefer" if the --mem-bind option includes the prefer option.
• SLURM_MEM_BIND_TYPE — set to the memory binding type specified with the --mem-bind option. possible values are "none", "rank", "map_mem:", "mask_mem:" and "local".
• SLURM_MEM_BIND_VERBOSE — set to "verbose" if the --mem-bind option includes the verbose option. set to "quiet" otherwise.
• SLURM_MEM_PER_CPU — same as --mem-per-cpu
• SLURM_MEM_PER_GPU — requested memory per allocated GPU. only set if the --mem-per-gpu option is specified.
• SLURM_MEM_PER_NODE — same as --mem
• SLURM_NETWORK — set to the value of the --network option, if specified.
• SLURM_NNODES — total number of nodes in the job allocation. see SLURM_JOB_NUM_NODES. included for backwards compatibility.
• SLURM_NODELIST — list of nodes allocated to the job. see SLURM_JOB_NODELIST. included for backwards compatibility.
• SLURM_NPROCS — same as SLURM_NTASKS. included for backwards compatibility.
• SLURM_NTASKS — set to the value of the --ntasks option, if specified. or, if any of the --ntasks-per-* options are specified, set to the number of tasks in the job.

NOTE: this is also an input variable for srun, so if it's set, it will effectively set the --ntasks option for srun when srun is called from the salloc shell. sneaky little bidirectional icon 🔁

• SLURM_NTASKS_PER_CORE — set to the value of the --ntasks-per-core option, if specified.
• SLURM_NTASKS_PER_GPU — set to the value of the --ntasks-per-gpu option, if specified.
• SLURM_NTASKS_PER_NODE — set to the value of the --ntasks-per-node option, if specified.
• SLURM_NTASKS_PER_SOCKET — set to the value of the --ntasks-per-socket option, if specified.
• SLURM_OOMKILLSTEP — same as --oom-kill-step
• SLURM_OVERCOMMIT — set to 1 if --overcommit was specified.
• SLURM_PROFILE — same as --profile
• SLURM_SHARDS_ON_NODE — number of GPU Shards available to the step on this node.
• SLURM_SUBMIT_DIR — the directory from which salloc was invoked or, if applicable, the directory specified by the -D, --chdir option.
• SLURM_SUBMIT_HOST — the hostname of the computer from which salloc was invoked.
• SLURM_TASKS_PER_NODE — number of tasks to be initiated on each node. values are comma separated and in the same order as SLURM_JOB_NODELIST. if two or more consecutive nodes have the same task count, that count is followed by (x#) where # is the repetition count. for example, SLURM_TASKS_PER_NODE=2(x3),1 means the first three nodes each execute two tasks and the fourth node executes one task. 🌸
• SLURM_THREADS_PER_CORE — only set if --threads-per-core or SALLOC_THREADS_PER_CORE were specified. the value is set to whatever --threads-per-core or SALLOC_THREADS_PER_CORE specified. this is used by subsequent srun calls within the job allocation.
• SLURM_TRES_PER_TASK — set to the value of --tres-per-task. if --cpus-per-task or --gpus-per-task is specified, it is also set in SLURM_TRES_PER_TASK as if it were specified in --tres-per-task.

---

💀 SIGNALS 💀

ok this part is important so listen up bestie. while salloc is waiting on a PENDING job allocation, most signals will cause salloc to revoke the allocation request and exit. bye request 👋

but — and this is the plot twist — if the allocation has already been granted and salloc has already started the specified command, then salloc will ignore most signals. it will not exit or release the allocation until the command exits. it's committed, it's loyal, it's not abandoning your process 😤

one notable exception is SIGHUP. a SIGHUP will cause salloc to release the allocation and exit without waiting for the command to finish. another exception is SIGTERM, which will be forwarded to the spawned process.

---

✨ EXAMPLES ✨

🎀 get an allocation + pop an xterm for interactive srun vibes

to get an allocation and open a new xterm in which srun commands may be typed interactively:

$ salloc -N16 xterm
salloc: Granted job allocation 65537
# (at this point the xterm appears, and salloc waits for xterm to exit)
salloc: Relinquishing job allocation 65537

🦋 allocation + parallel app, one command line, no extra steps

to grab an allocation of nodes and launch a parallel application on one command line:

$ salloc -N5 srun -n10 myprogram

👑 heterogeneous job, 3 components, each with its own unique node set

to create a heterogeneous job with 3 components, each allocating a unique set of nodes:

$ salloc -w node[2-3] : -w node4 : -w node[5-7] bash
salloc: job 32294 queued and waiting for resources
salloc: job 32294 has been allocated resources
salloc: Granted job allocation 32294

---

📜 COPYING 📜

Copyright (C) 2006-2007 The Regents of the University of California. Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER). Copyright (C) 2008-2010 Lawrence Livermore National Security. Copyright (C) 2010-2022 SchedMD LLC.

this file is part of Slurm, a resource management program. for details, see <https://slurm.schedmd.com/>.

Slurm is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.

Slurm is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. see the GNU General Public License for more details.

---

🌈 SEE ALSO 🌈

sinfo(1), sattach(1), sbatch(1), squeue(1), scancel(1), scontrol(1), slurm.conf(5), sched_setaffinity (2), numa (3)

---

and that's the whole page bestie, you ate that 💅✨ so anyway — this masterpiece is Slurm 26.05 from the Slurm Commands manual, filed under salloc(1), and it's literally all yours now. go forth and allocate responsibly, mwah 🫶🎀🦋