Skip to content

My development environment supporting SSH/SFTP based remote development (PyCharm) in k8s

Notifications You must be signed in to change notification settings

kingb12/docker-dev-env

Repository files navigation

My Docker Development Environment

This is a repo for creating a docker image which I use to develop my research projects. It is not for deployment, and typically neither are the applications developed using it. It is not intended to be lightweight or plug-and-play ready, but where possible, I tried to make it extensible for other users. Specifically, each of my projects initiates a new Dockerfile from this image, creates its own virtual environemtn and installs project dependencies, plus any other related project set up.

How to replicate

  1. Fork repository
  2. Change id_rsa.pub to your public SSH key or one of your choosing (this will be the one needed to log into th epod vis SSH)
  3. replace all occurrences of bking2 or kingb12 with what you want
  • bking2 is the user of the docker container, and ssh user generally (e.g. in ssh_config)
  • kingb12 is the repo owner of the DockerHub/Gitlab account used for upload
  1. Add credentials to Settings/Secrets & Variables/Actions:
  • DOCKERHUB_USERNAME: your Docker Hub user name
  • DOCKERHUB_TOKEN: your Docker Hub API Key
  • GITLAB_DOCKER_REGISTRY_USER: gitlab user name
  • GITLAB_DOCKER_REGISTRY_TOKEN: gitlab API key

Motivation

The features below facilitate a workflow like so:

Development (Pod):

  • Remote debugging with PyCharm. The gist of how this works is:
    • create a project-specific image which uses installed conda to create a virtual environment for execution and installing project specific dependencies
    • launch a pod running SSHD with a non-privileged user on a non-privileged port (2022 here)
    • port forward to your local machine: kubectl port-forward pods/<your_pod_name> 2022:2022
    • set up an SSH interpreter in PyCharm using localhost and port 2022, pointing to the virtual environment venv/bin/python on your container
    • Debug w/ automatic code syncing to the pod

Running Experiments (Job)

  • Clone updated code before job launch
    • This takes some careful job definition, see an example with a few other useful things below, that likely requires adjustments before use.
    • Requires setting a Secret in k8s with SSH_PRIVATE_KEY, SSH_PUBLIC_KEY. Recommended to not using your personal ssh key here, but create a new one, since this may be visible to anyone in the k8s namespace. You can then register this new key as a deployment key for the appropriate private Github repo(s).
  • Run!

Example Job command

Requires a secret defining the SSH key (public and private, load as environment variables), and if needed, a Huggingface API token (also env-variable loaded).

        command: [ "/bin/sh" ]
       # commmand does the following:
       # 1) set up the github deployment (SSH) key: we store the relevant secrets as environment variables, because
       #    injecting them as files makes root the owner, and the container needs to run as non-privileged user. We
       #    also add githubs public keys to known_hosts to bypass the interactive fingerprint check on later clones
       # 2) clone repo: this clones the repo into temp, since the folder already exists and contains our venv
       #    definition that we don't want to overwrite. Then, we move the .git definition into the folder (which had
       #    none to begin with), and fetch and pull. This doesn't yet overwrite files, we then need to do a hard reset to
       #    origin/main (assuming this is the branch we are always running jobs from). This step allows us to not re-build
       #    the docker container for every code change, only those which are important to it.
       # 3) log in the huggingface for private model access and saving
       # 4) run the actual job: everything after 'job ready to start' is the script we want to run
       args:
         - -c
         - >-
           echo "$SSH_PRIVATE_KEY" > /home/bking2/.ssh/id_rsa &&
           echo "$SSH_PUBLIC_KEY" > /home/bking2/.ssh/id_rsa.pub &&
           chmod 644 /home/bking2/.ssh/id_rsa.pub &&
           chmod 600 /home/bking2/.ssh/id_rsa &&
           echo "github.com ssh-rsa AAAAB3NzaC1yc2EAAAABIwAAAQEAq2A7hRGmdnm9tUDbO9IDSwBK6TbQa+PXYPCPy6rbTrTtw7PHkccKrpp0yVhp5HdEIcKr6pLlVDBfOLX9QUsyCOV0wzfjIJNlGEYsdlLJizHhbn2mUjvSAHQqZETYP81eFzLQNnPHt4EVVUh7VfDESU84KezmD5QlWpXLmvU31/yMf+Se8xhHTvKSCZIFImWwoG6mbUoWf9nzpIoaSjB+weqqUUmpaaasXVal72J+UX2B+2RPW3RcT0eOzQgqlJL3RKrTJvdsjE3JEAvGq3lGHSZXy28G3skua2SmVi/w4yCE6gbODqnTWlg7+wC604ydGXA8VJiS5ap43JXiUFFAaQ==" >> /home/bking2/.ssh/known_hosts &&
           echo "github.com ecdsa-sha2-nistp256 AAAAE2VjZHNhLXNoYTItbmlzdHAyNTYAAAAIbmlzdHAyNTYAAABBBEmKSENjQEezOmxkZMy7opKgwFB9nkt5YRrYMjNuG5N87uRgg6CLrbo5wAdT/y6v0mKV0U2w0WZ2YB/++Tpockg=" >> /home/bking2/.ssh/known_hosts &&
           echo "github.com ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIOMqqnkVzrm0SdG6UOoqKLsabgH5C9okWi0dh2l9GKJl" >> /home/bking2/.ssh/known_hosts &&
           cd /home/bking2 &&
           git clone [email protected]:kingb12/my_project.git temp &&
           mv temp/.git my_project/.git &&
           rm -rf temp &&
           cd my_project &&
           git fetch && git pull && git reset --hard origin/main &&
           python -c "from huggingface_hub.hf_api import HfFolder; HfFolder.save_token('${HF_API_TOKEN}')" &&
           echo "job ready to start" &&
           <actual job command(s) here>

Features

SSH & SFTP Access

A non-privileged run of sshd on the container. combined with port forwarding on a non-privileged port like 2022, this allows direct ssh and sftp access to the container, useful for setting up a remote interpreter in PyCharm.

Relevant Files
  • Dockerfile: installs and sets up sshd, including the default command to run it on container launch (need to set this as part of the command as well when running in k8s)
  • sshd_config: the config copied into place for the running sshd service
  • sshd-1.service: a service definition for our non-privileged run of sshd
  • id_rsa.pub: MY ssh public key. Replace with yours!

Miniconda

Miniconda pre-installed, no significant packages added. Images should extend this one and run conda environment set up.

github.com known hosts

See warning about the security of this, this bypasses the fingerprint check in git clone via ssh by pre-loading the public keys for github.com, so that a repo can be clone on container start. You're implicitly trusting that I did this correctly, see this ServerFault discussion for details, or this document from Github.

ZSH and oh-my-zsh

I install zsh and my own personal dotfile preferences, only some of which are generally applicable. Adjust or remove as needed.

(my) ssh config

I found it useful to be able to ssh from my container to other hosts I use, especially for copying files while debugging and/or placing them in a mounted volume. Remove this build step if it doesn't sound useful or replace with your own config.

About

My development environment supporting SSH/SFTP based remote development (PyCharm) in k8s

Topics

Resources

Stars

Watchers

Forks