2019-snakemake-byok8s

Overview

This is an example of a Snakemake workflow that:

• is a command line utility
• is bundled as a Python package
• is designed to run on a Kubernetes cluster
• can be tested locally or with Travis CI using minikube

Snakemake functionality is provided through a command line tool called byok8s, so that it allows you to do this (abbreviated for clarity):

# Create virtual k8s cluster
minikube start

# Run the workflow
byok8s --s3-bucket=mah-s3-bukkit my-workflowfile my-paramsfile

# Clean up the virtual k8s cluster
minikube stop

Snakemake workflows are provided via a Snakefile by the user. Snakemake runs tasks on the Kubernetes (k8s) cluster. The approach is for the user to provide their own Kubernetes cluster (byok8s = Bring Your Own Kubernetes).

The example above uses minikube to make a virtual k8s cluster, useful for testing.

For real workflows, your options for kubernetes clusters are cloud providers:

• AWS EKS (Elastic Container Service)
• GCP GKE (Google Kuberntes Engine)
• Digital Ocean Kubernetes service
• etc...

The Travis CI tests utilize minikube to run test workflows.

Quickstart

This runs through the installation and usage of 2019-snakemake-byok8s.

Step 1: Set up Kubernetes cluster with minikube.

Step 2: Install byok8s.

Step 3: Run the byok8s workflow using the Kubernetes cluster.

Step 4: Tear down Kubernetes cluster with minikube.

Step 1: Set Up Virtual Kubernetes Cluster

For the purposes of the quickstart, we will walk through how to set up a local, virtual Kubernetes cluster using minikube.

Start by installing minikube:

scripts/install_minikube.sh

Once it is installed, you can start up a kubernetes cluster with minikube using the following commands:

cd test
minikube start

NOTE: If you are running on AWS, run this command first

minikube config set vm-driver none

to set the the vm driver to none and use native Docker to run stuff.

If you are running on AWS, the DNS in the minikube kubernetes cluster will not work, so run this command to fix the DNS settings (should be run from the test/ directory):

kubectl apply -f fixcoredns.yml
kubectl delete --all pods --namespace kube-system

Step 2: Install byok8s

Start by setting up a python virtual environment, and install the required packages into the virtual environment:

pip install -r requirements.txt

This installs snakemake and kubernetes Python modules. Now install the byok8s command line tool:

python setup.py build install

Now you can run:

which byok8s

and you should see byok8s in your virtual environment's bin/ directory.

This command line utility will expect a kubernetes cluster to be set up before it is run.

Setting up a kubernetes cluster will create... (fill in more info here)...

Snakemake will automatically create the pods in the cluster, so you just need to allocate a kubernetes cluster.

Step 3: Run byok8s

Now you can run the workflow with the byok8s command. This submits the Snakemake workflow jobs to the Kubernetes cluster that minikube created.

You should have your workflow in a Snakefile in the current directory. Use the --snakefile flag if it is named something other than Snakefile.

You will also need to specify your AWS credentials via the AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment variables. These are used to to access S3 buckets for file I/O.

Finally, you will need to create an S3 bucket for Snakemake to use for file I/O. Pass the name of the bucket using the --s3-bucket flag.

Start by exporting these two vars (careful to scrub them from bash history):

 export AWS_ACCESS_KEY_ID=XXXXX
 export AWS_SECRET_ACCESS_KEY=XXXXX

Run the alpha workflow with blue params:

byok8s --s3-bucket=mah-bukkit workflow-alpha params-blue

Run the alpha workflow with red params:

byok8s --s3-bucket=mah-bukkit workflow-alpha params-red

Run the gamma workflow with red params, &c:

byok8s --s3-bucket=mah-bukkit workflow-gamma params-red

(NOTE: May want to let the user specify input and output directories with flags.)

All input files are searched for relative to the working directory.

Step 4: Tear Down Kubernetes Cluster

The last step once the workflow has been finished, is to tear down the kubernetes cluster. The virtual kubernetes cluster created by minikube can be torn down with the following command:

minikube stop

Using Kubernetes with Cloud Providers

Cloud Provider	Kubernetes Service	Guide
Minikube (on AWS EC2)	Minikube	Minikube AWS Guide
Google Cloud Platform (GCP)	Google Container Engine (GKE)	GCP GKE Guide
Amazon Web Services (AWS)	Elastic Container Service (EKS)	AWS EKS Guide
Digital Ocean (DO)	DO Kubernetes (DOK)	DO DOK Guide