A short introduction to use anaconda (and your customized python environment) on Sango (or other similar clusters).

As for the basic knowledge of how to use Sango, please refer to getting-started

The author is not sure if this is the best way to run python programs with customized environment, but just provides the method he used for running his programs.

Let's start step-by-step!

Allocate a bash for your installing.

Some installing need much CPU power, so it is better to allocate some source first.

srun -p compute -t 10:00:00 --pty bash -l

Installing :

After you allocate a bash, you can go to your user directory (on Sango it will be /home/d/dongqihan or similar, where d is the first letter of username).

cd ~

Then download an anaconda from web and install it. For example:

curl -O https://repo.continuum.io/archive/Anaconda3-5.0.1-Linux-x86_64.sh
bash Anaconda3-5.0.1-Linux-x86_64.sh

Using anaconda to create your own virtual environment

After installing anaconda, you should know where it is installed. Then you can export the PATH of it bin folder. For example,

export PATH=/home/d/dongqihan/conda/anaconda/bin:$PATH

Then you can start using it, just like on your own computer. Now you can create a new environment:

conda create -n CusEnv python=3.6
source activate CusEnv

Installing necessary packages

After activating the virtual environment, you can start installing python libraries. Note that many libraries require modules such as cmake, openmpi, cuda... So before installing them, just load the module first.

module avail
module load xxx yyy

Then you can install use pip or conda

pip install zzz
conda install kkk

Using your own environment

With all required libraries installed, you may want to run your programs. One approach is to load the necessary modules and export the PATH and activate the environment in a bash shell file, run.sh , where inside is, for example

module load xxx yyy
export PATH=/home/d/dongqihan/conda/anaconda/bin:$PATH
source activate CusEnv
python deep_learning.py

Then you can srun run.sh simply by

srun -t 10:00:00 -n 1 -c 8 bash ./run.sh

or, use a slurm script, job.slurm :

#!/bin/bash

#SBATCH --job-name=python
#SBATCH --partition=compute
#SBATCH --time=1-0
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=1
#SBATCH --cpus-per-task=8
#SBATCH --mem=2g
#SBATCH --output=/dev/null
#SBATCH --error=./errors/slurm-%A.err

# Display all variables set by slurm
env | grep "^SLURM" | sort

# Print hostname job executed on.
echo
srun bash ./run.sh
echo

and just submit it by

sbatch job.slurm

Made by

Dongqi Han, CNRU, OIST [email protected]