(small) docker cheat sheet

docker_cheat_sheet.sh

#################################################################################
# Authors: Keiller Nogueira, Victor de Melo
#################################################################################

# nvidia-docker seems more stable now and several images are starting to adopt it (including 
# [kaixhin](https://github.com/NVIDIA/nvidia-docker/issues/85).
# hence, we suggest to use it instead, as it solves a few issues with gpu passthrough.
# usage: it is only required when creating a container, i.e, with `run` and related arguments
# you're not required to provide $DOCKER_NVIDIA_DEVICES as it will find the devices itself

#================================================================================
# PARA CRIAR UM NOVO CONTAINER DOCKER
# USANDO nvidia-docker (automatically exports GPU devices)
nvidia-docker run --name NOME_DO_CONTAINER -v /mnt/user:/home/ -ti kaixhin/cuda-caffe /bin/bash
# --name será o nome do container que irá criar
# -v é o mapeamento de diretório, para acessar pastas do host no container

# OU, definindo manualmente

# PARA DEFINIR OS DEVICES (GPUs)
# DEVE SER EXECUTADO SOMENTE A PRIMEIRA VEZ!!!!!!!!!!! VALIDAR (echo $DOCKER_NVIDIA_DEVICES) SE JÁ EXISTE NO SISTEMA, ANTES DE EXECUTAR!!
DOCKER_NVIDIA_DEVICES="
--device /dev/nvidia0:/dev/nvidia0
--device /dev/nvidia1:/dev/nvidia1
--device /dev/nvidia2:/dev/nvidia2
--device /dev/nvidiactl:/dev/nvidiactl
--device /dev/nvidia-uvm:/dev/nvidia-uvm"
# IF SOME INTERFACES /dev/nvidia* ARE MISSING, run nvidia-smi and deviceQuery, then they will appear

# NO CASO, A IMAGEM tleyden5iwx/caffe-gpu JÁ TEM O CAFFE INSTALADO (/opt/caffe/). LOGO, É SÓ USAR
docker run --name NOME_DO_CONTAINER -v /mnt/user:/home/ -ti $DOCKER_NVIDIA_DEVICES kaixhin/cuda-caffe /bin/bash

# GUI RUDIMENTARY SUPPORT, add the following exports when instantiating (docker run) a container
-v /tmp/.X11-unix:/tmp/.X11-unix -e DISPLAY=unix$DISPLAY
# source: http://stackoverflow.com/a/27162721/957997
#OR 

touch /tmp/.docker.xauth
XSOCK=/tmp/.X11-unix
XAUTH=/tmp/.docker.xauth
xauth nlist :0 | sed -e 's/^..../ffff/' | xauth -f $XAUTH nmerge -
docker run -ti -v $XSOCK:$XSOCK -v $XAUTH:$XAUTH -e XAUTHORITY=$XAUTH xeyes
docker run --rm -v $XSOCK:$XSOCK -v $XAUTH:$XAUTH -e DISPLAY=unix$DISPLAY -e XAUTHORITY=$XAUTH -ti victorhcm/opencv /bin/bash
# http://stackoverflow.com/a/25280523/957997

# check this one too: http://stackoverflow.com/a/25334301/957997
# seems simpler

# ====================================================================

#PARA INICIAR UM CONTAINER
docker start NOME_DO_CONTAINER

#PARA DAR ATTACH A UM CONTAINER (SE DEMORAR MUITO, DÊ UM CTRL+C)
docker attach NOME_DO_CONTAINER

#VALIDAR QUAIS CONTAINERS ESTÃO ATIVOS
docker ps

#LISTAR TODOS OS CONTAINERS
docker ps -a

#PARA SAIR DO CONTAINER E DEIXA-LO EM EXECUCAO
ctrl+p ctrl+q

# JUPYTER PASSTHROUGH 
$ sudo docker run \
    -it \
    -p 0.0.0.0:80:8888 \
    -e "USE_HTTP=1" -e "PASSWORD=" \
    -v /home/ganon/Downloads/host_folder:/docker_folder \
    ipython/scipyserver

# or
docker run -d -p 8888:8888 -v /notebook:/notebook xblaster/tensorflow-jupyter

###############################################################################
# Matlab docker container
###############################################################################
nvidia-docker run --name raphael-matlab -v /usr/local/MATLAB/MATLAB_Production_Server/R2015a:/usr/local/MATLAB/from-host -v ~/matlab-logs:/var/log/matlab --mac-address="host:mac:address" -ti --entrypoint=/bin/bash $DOCKER_NVIDIA_DEVICES ninjaben/matlab-support -i


###############################################################################
# CONFIGURING JUPYTER
###############################################################################
# Dockerfile
FROM foo/bar

RUN pip install jupyter && \
    mkdir -p -m 700 /root/.jupyter/ && \
    echo "c.NotebookApp.ip = '*'" >> /root/.jupyter/jupyter_notebook_config.py

# Add Tini. Tini operates as a process subreaper for jupyter. This prevents
# kernel crashes.
ENV TINI_VERSION v0.6.0
ADD https://github.com/krallin/tini/releases/download/${TINI_VERSION}/tini /usr/bin/tini
RUN chmod +x /usr/bin/tini
ENTRYPOINT ["/usr/bin/tini", "--"]

EXPOSE 8888
CMD ["jupyter", "notebook", "--port=8888", "--no-browser", "--ip=0.0.0.0"]
# End Dockerfile

# Now, access it through the ip (didn't work using the hostname)



# REFERENCES: 
# Also, need to add (https://hub.docker.com/r/jupyter/notebook/~/dockerfile/) to bind to any IP:
mkdir -p -m 700 /root/.jupyter/ && \
    echo "c.NotebookApp.ip = '*'" >> /root/.jupyter/jupyter_notebook_config.py

## Jupyter crashing
# Using jupyter notebook as a Docker CMD results in kernels repeatedly crashing, likely due to a lack of PID reaping. To avoid this, use the tini init as your Dockerfile ENTRYPOINT:

# Add Tini. Tini operates as a process subreaper for jupyter. This prevents
# kernel crashes.
ENV TINI_VERSION v0.6.0
ADD https://github.com/krallin/tini/releases/download/${TINI_VERSION}/tini /usr/bin/tini
RUN chmod +x /usr/bin/tini
ENTRYPOINT ["/usr/bin/tini", "--"]

EXPOSE 8888
CMD ["jupyter", "notebook", "--port=8888", "--no-browser", "--ip=0.0.0.0"]

# http://jupyter-notebook.readthedocs.io/en/latest/public_server.html

# Adding password and enabling https
# https://github.com/jupyter/docker-stacks/blob/master/base-notebook/jupyter_notebook_config.py
# (404) https://github.com/jupyter/docker-stacks/blob/master/minimal-notebook/jupyter_notebook_config.py

###############################################################################