README.md

kubernetes

Files and guidelines on how to setup containers on the EPFL Kubernetes cluster

Kubernetes administration: create, enter, delete, ...

Check status of all pods and services

kubectl get all

Create pod/service

Navigate to the folder where the yaml files for your configurations are on your (local) machine.

kubectl create -f pod-deep.yaml

kubectl create -f svc-deep.yaml

Check Kubernetes launching progress

kubectl describe pod

Enter/exit command line of pod

kubectl exec -it deep -- /bin/bash

exit

Get external IP and port

kubectl get pod deep -o yaml | grep hostIP

Note that this only gives the ip address. The external port mapping can be found through kubectl get all

Delete pod/service

kubectl delete pod deep

kubectl delete service deep

Copy files from local machine to pod

kubectl cp ~/Documents/file.txt deep:/data/

Copy files and directories with progress bar

rsync -r --progress <source> <target>

Pytorch pod commands

Use install script

sudo bash install_script.sh

Administration

Check GPU status

nvidia-smi

GPU clean memory

Replace /dev/nvidia# with the GPU number:
for i in $(sudo lsof /dev/nvidia1 | grep python | awk '{print $2}' | sort -u); do kill -9 $i; done
source: https://forums.developer.nvidia.com/t/11-gb-of-gpu-ram-used-and-no-process-listed-by-nvidia-smi/44459/12

GPU clean memory (dirty)

kill all python processes (sleeping, but claiming GPU memory)

pkill -9 python

Kill process(es)

Replace name with part of the executed command (e.g. see (h)top):
ps aux | grep Reptile
Replace id with process id listed in the previous step kill -9 id

Check PyTorch GPU

python

import torch

torch.cuda.get_device_name(0)

Check TensorFlow GPU

python

import tensorflow as tf

with tf.Session() as sess:
    print(sess.list_devices())

Launch (test) webserver

python -m http.server 8888

Launch Jupyter Notebook

jupyter notebook --ip=0.0.0.0 --port=8888 --allow-root

Launch Jupyter Lab

jupyter lab --ip=0.0.0.0 --port=8888 --allow-root

Launch VS Code Server (from the code-server folder)

sudo sh code-server --port 8080 --host 0.0.0.0 --auth none

Shut down VS Code Server/...

pkill -f code-server

source: How to kill all processes with a given partial name? [closed]

TensorBoard

Launching TensorBoard

tensorboard --logdir=./runs --port=8888

Kill TensorBoard

ps -ef|grep tensorboard

kill -9 <first/second pid>

Install software

Install code-server web IDE (MS VS Code)

See the code-server GitHub page

Install matplotlib

pip install matplotlib

Install/Upgrade PyTorch version

0.4.1:

pip install http://download.pytorch.org/whl/cu90/torch-0.4.1-cp36-cp36m-linux_x86_64.whl

1.0.0 (has cuda 9.0.176 embedded):

pip3 install https://download.pytorch.org/whl/cu90/torch-1.0.0-cp36-cp36m-linux_x86_64.whl

Install CUDNN 7

Follow https://medium.com/@zhanwenchen/install-cuda-and-cudnn-for-tensorflow-gpu-on-ubuntu-79306e4ac04e

Install Tensorflow-gpu 1.8.0

pip install tensorflow-gpu==1.8

Install screen (& autocomplete)

echo "defshell -bash" > ~/.screenrc

source ~/.screenrc

Kubernetes config files

Conda

Make virtual environment visible in Jupyter

conda activate myenv
python -m ipykernel install --user --name myenv --display-name "Python (myenv)"