How can I change my slurm script below so that each python job gets a unique GPU? The node had 4 GPUs, I would like to run 1 python job per each GPU.
The problem is that all jobs use the first GPU and other GPUs are idle.
#!/bin/bash
#SBATCH --qos=maxjobs
#SBATCH -N 1
#SBATCH --exclusive
for i in `seq 0 3`; do
cd ${i}
srun python gpu_code.py &
cd ..
done
wait
In your example your four jobs will be executed sequentially. I suggest submitting four separate jobs that only request a single GPU. Then the jobs only use one GPU and will be executed simultaneously. If the jobs have depencies you can use:
sbatch --dependency=afterok:${jobid_of_previous_job} submit.sh. This job will start after the prvious has finished.
As you do not request GPUs in the submission scripts, you will have to manage the CUDA_VISIBLE_DEVICES var by yourself to direct each python script to one specific GPU.
Try with
#!/bin/bash
#SBATCH --qos=maxjobs
#SBATCH -N 1
#SBATCH --exclusive
for i in `seq 0 3`; do
cd ${i}
export CUDA_VISIBLE_DEVICES=$i
python gpu_code.py &
cd ..
done
wait
Related
I'm still exploring how to work with the Slurm scheduler and this time I really got stuck. The following batch script somehow doesn't work:
#!/usr/bin/env bash
#SBATCH --job-name=parallel-plink
#SBATCH --mem=400GB
#SBATCH --ntasks=4
cd ~/RS1
for n in {1..4};
do
echo "Starting ${n}"
srun --input none --exclusive --ntasks=1 -c 1 --mem-per-cpu=100G plink --memory 100000 --bfile RS1 --distance triangle bin --parallel ${n} 4 --out dt-output &
done
Since most of the SBATCH options are inside the batch script the invocation is just: 'sbatch script.sh'
The slurm-20466.out only contains the four echo'ing outputs: cat slurm-20466.out
Starting 1
Starting 2
Starting 3
Starting 4
I double checked the command without srun and that works without errors.
I must confess I am also responsible for the Slurm scheduler configuration itself. Let me know if I could try to change anything or when more information is needed.
You start your srun commands in the background to have them run in parallel. But you never wait for the commands to finish.
So the loop runs through very quickly, echoes the "Starting ..." lines, starts the srun command in the background and afterwards finishes. After that, your sbatch-script is done and terminates successfully, meaning that your job is done. With that, your allocation is revoked and your srun commands are also terminated. You might be able to see that they started with sacct.
You need to instruct the batch script to wait for the work to be done before it terminates, by waiting for the background processes to finish. To do that, you simply have to add a wait command in your script at the end:
#!/usr/bin/env bash
#SBATCH --job-name=parallel-plink
#SBATCH --mem=400GB
#SBATCH --ntasks=4
cd ~/RS1
for n in {1..4};
do
echo "Starting ${n}"
srun --input none --exclusive --ntasks=1 -c 1 --mem-per-cpu=100G plink --memory 100000 --bfile RS1 --distance triangle bin --parallel ${n} 4 --out dt-output &
done
wait
How can I run a number of python scripts in different nodes in SLURM?
Suppose,
I select 5 cluster nodes using #SBATCH --nodes=5
and
I have 5 python scripts code1.py, code2.py....code5.py and I want to run each of these scripts in 5 different nodes simultaneously. How can I achieve this?
Do these five scripts need to run in a single job? Do they really need to run simultaneously? Is there some communication happeneing between them? Or are they independent from one another?
If they are essentially independent, then you should most likely pu tthem into 5 different jobs with one nodes each. That way you don't have to find five free nodes, but your the first job can start as soon as there is a single free node. If there are many scripts you want to start like that, it might be interesting to look into job arrays.
If you need to run them in parallel, you will need to use srun in your jobscript to start the scripts. This example shows a job where you have 10 cores per task and each node has one task.
#!/bin/bash
#[...]
#SBATCH -N 5
#SBATCH -n 5
#SBATCH -c 10
#[...]
srun -N 1 -n1 python code1.py &
srun -N 1 -n1 python code2.py &
srun -N 1 -n1 python code3.py &
srun -N 1 -n1 python code4.py &
srun -N 1 -n1 python code5.py &
wait
You need to run the srun calls in the background, as bash would otherwise wait for them to finish before executing the next one.
I am new to Slurm and I also found the related questions about this topic. However, I am still confused about several points of how to use srun. According to the official document, srun will typically first allocate resources and then run the parallel jobs. For example, I want to run 20 tasks and if I submit my job based on the following script, I am not sure how many tasks are created. Because sbatch only takes care of allocating resources instead of executing program.
#!/bin/sh
#SBATCH -n 20
#SBATCH --mpi=pmi2
#SBATCH -o myoutputfile.txt
module load mpi/mpich-x86_64
mpirun mpiprogram < inputfile.txt
If I am trying to run sequential program like the following, I am not whether there will be a difference or not. For example, I can simply remove the srun command in this script. What will happen?
#!/bin/sh
#SBATCH -n 1
#SBATCH -N 1
srun tar zxf julia-0.3.11.tar.gz
echo "prefix=/software/julia-0.3.11" > julia/Make.user
cd julia
srun make
The first example will spawn 20 tasks ; sbatch will request 20 CPUs and also set up the environment so that mpirun knows how many CPUs were requested for the job. mpirun will then spawn as many processes as were allocated (provided that OpenMPI was compiled with Slurm support).
The #SBATCH --mpi=pmi2 part is meant for srun so it will have no effect if srun is not called in the submission script.
In the second example, there will be no difference in the number of processes spawned as only one is needed. But, with srun, the output of sstat will be more reliable, the management of signals will be more precise, and the buffering of the output will be more controlled (via the srun command line options).
If you request multiple tasks, srun will instantiate that many processes. It can be an MPI program, or a sequential program that adapts its behaviour based on the SLURM_PROC_ID environment variable.
Also you can run multiple srun in the same submission script. Each instance of srun (called a "step") is then accounted separately in the accounting (sacct).
Finally, srun can use a subset of the allocation and organise the micro-scheduling of many small tasks in a single job (see the example in the srun manpage).
I have access to a large GPU cluster (20+ nodes, 8 GPUs per node) and I want to launch a task several times on n GPUs (1 per GPU, n > 8) within one single batch without booking full nodes with the --exclusive flag.
I managed to pre-allocate the resources (see below), but I struggle very hard with launching the task several times within the job. Specifically, my log shows no value for the CUDA_VISIBLE_DEVICES variable.
I know how to do this operation on fully booked nodes with the --nodes and --gres flags. In this situation, I use --nodes=1 --gres=gpu:1 for each srun. However, this solution does not work for the present question, the job hangs indefinitely.
In the MWE below, I have a job asking for 16 gpus (--ntasks and --gpus-per-task). The jobs is composed of 28 tasks which are launched with the srun command.
#!/usr/bin/env bash
#SBATCH --job-name=somename
#SBATCH --partition=gpu
#SBATCH --nodes=1-10
#SBATCH --ntasks=16
#SBATCH --gpus-per-task=1
for i in {1..28}
do
srun echo $(hostname) $CUDA_VISIBLE_DEVICES &
done
wait
The output of this script should look like this:
nodeA 1
nodeR 2
...
However, this is what I got:
nodeA
nodeR
...
When you write
srun echo $(hostname) $CUDA_VISIBLE_DEVICES &
the expansion of the $CUDA_VISIBLE_DEVICES variable will be performed on the master node of the allocation (where the script is run) rather than on the node targeted by srun. You should escape the $:
srun echo $(hostname) \$CUDA_VISIBLE_DEVICES &
By the way, the --gpus-per-task= appeared in the sbatch manpage in the 19.05 version. When you use it with an earlier option, I am not sure how it goes.
I submit some cluster mode spark jobs which run just fine when I do it one by one with below sbatch specs.
#!/bin/bash -l
#SBATCH -J Spark
#SBATCH --time=0-05:00:00 # 5 hour
#SBATCH --partition=batch
#SBATCH --qos qos-batch
###SBATCH -N $NODES
###SBATCH --ntasks-per-node=$NTASKS
### -c, --cpus-per-task=<ncpus>
### (multithreading) Request that ncpus be allocated per process
#SBATCH -c 7
#SBATCH --exclusive
#SBATCH --mem=0
#SBATCH --dependency=singleton
If I use a launcher to submit the same job with different node and task numbers, the system gets confused and tries to assign according to $SLURM_NTASK which gives 16. However I ask for example only 1 node,3tasks.
#!/bin/bash -l
for n in {1..4}
do
for t in {3..4}
do
echo "Running benchmark with ${n} nodes and ${t} tasks per node"
sbatch -N ${n} --ntasks-per-node=${t} spark-teragen.sh
sleep 5
sbatch -N ${n} --ntasks-per-node=${t} spark-terasort.sh
sleep 5
sbatch -N ${n} --ntasks-per-node=${t} spark-teravalidate.sh
sleep 5
done
done
How can I fix the error below by preventing slurm assign weird number of tasks per node which exceeds the limit.
Error:
srun: Warning: can't honor --ntasks-per-node set to 3 which doesn't match the
requested tasks 16 with the number of requested nodes 1. Ignoring --ntasks-per-node.
srun: error: Unable to create step for job 233838: More processors requested than
permitted