View source for Interactive Jobs

For quick exploration on a compute node — testing a snippet, profiling memory usage, debugging an environment — request an interactive shell instead of submitting a batch job. SLURM offers two commands for this: <code>sinteractive</code>, which drops you straight into a shell on a compute node, and <code>salloc</code>, which reserves resources you can use from the login node.

== sinteractive ==

<code>sinteractive</code> is a thin wrapper on <code>srun</code> that requests an allocation and gives you a shell on the allocated node:

<syntaxhighlight lang="bash">
sinteractive -c <num_cpus> --mem <amount_mem> --time <minutes> -p <partition>
</syntaxhighlight>

After SLURM allocates resources you land at a new shell prompt on a compute node — run <code>hostname</code> to confirm which one. From there, run commands as you would in any shell; the allocation ends when you exit.

If you run <code>sinteractive</code> with no flags, you get a default allocation of 1 CPU, 100 MB of memory, and 1 hour of wall time. That is fine for a quick test but too small for real work.

=== Wrapper source ===

The wrapper itself is a four-line shell script:

<syntaxhighlight lang="bash">
#!/bin/bash
srun "$@" -I60 -N 1 -n 1 --pty bash -i
</syntaxhighlight>

The flags translate to: time out the allocation request after 60 seconds (<code>-I60</code>), one node (<code>-N 1</code>), one task (<code>-n 1</code>), and a pseudo-terminal running an interactive bash (<code>--pty bash -i</code>).

== salloc ==

<code>salloc</code> requests resources without immediately running anything. Use it when you want to keep your shell on the login node and dispatch tasks into the allocation as you go:

<syntaxhighlight lang="bash">
salloc -p main --time=2:00:00
</syntaxhighlight>

Once the allocation is granted, launch tasks with <code>srun</code>:

<syntaxhighlight lang="bash">
srun <command>
</syntaxhighlight>

The default <code>salloc</code> wall time is 1 hour. For longer sessions add <code>--time</code> as in the example above.

== See also ==

* [[Partitions / Queues]]
* [[Choosing a node (constraints)]]
* [[Cancelling Jobs]]
* [[Using Slurm | Scheduler overview]]
* [https://slurm.schedmd.com/salloc.html salloc documentation]