This is an example of a Snakemake workflow that:
- is a command line utility called
- is bundled as an installable Python package
- is designed to run on a Kubernetes (k8s) cluster
- can be tested with Travis CI (and/or locally) using minikube
What is byok8s?¶
byok8s = Bring Your Own Kubernetes (cluster)
k8s = kubernetes
byok8s is a command line utility that launches a Snakemake workflow on an existing Kubernetes cluster. This allows you to do something like this (also see the Installation and Quickstart guides in the documentation):
# Install byok8s python setup.py build install # Create virtual k8s cluster minikube start # Run the workflow on the k8s cluster cd /path/to/workflow/ byok8s my-workflowfile my-paramsfile --s3-bucket=my-bucket # Clean up the virtual k8s cluster minikube stop
Getting Up and Running¶
See the Quickstart Guide to get up and running with byok8s.
How does byok8s work?¶
The command line utility requires the user to provide three input files:
- A snakemake workflow, via a
- A workflow configuration file (JSON)
- A workflow parameters file (JSON)
Additionally, the user must create the following resources:
- A kubernetes cluster up and running
- An S3 bucket (and AWS credentials to read/write)
A sample Snakefile, workflow config file, and workflow
params file are provided in the
The workflow config file specifies which workflow targets and input files to use.
The workflow parameters file specifies which parameters to use for the workflow steps.
Why S3 buckets?¶
AWS credentials and an S3 bucket is required to run workflows because of restrictions on file I/O on nodes in a kubernes cluster. The Snakemake workflows use AWS S3 buckets as remote providers for the Kubernetes nodes, but this can be modified to any others that Snakemake supports.
AWS credentials are set with the two environment variables:
These are passed into the Kubernetes cluster by byok8s and Snakemake.
Kubernetes and Minikube¶
Kubernetes is a technology that utilizes Docker container to orchestrate a cluster of compute nodes. These compute nodes are usually real compute nodes requested and managed via a cloud provider, like AWS or Google Cloud.
But the compute nodes can also be virtual, which is where minikube comes in. It creates a kubernetes cluster that is entirely local and virtual, which makes testing easy. See the byok8s Minikube Guide for details about how to use minikube with byok8s.
The Travis CI tests also utilize minikube to run test workflows. See byok8s Travis Tests for more information.
For real workflows, your options for kubernetes clusters are cloud providers. We have guides for the following:
- AWS EKS (Elastic Container Service)
- GCP GKE (Google Kuberntes Engine)
- Digital Ocean Kubernetes service
Kubernetes + byok8s: In Practice¶
|Cloud Provider||Kubernetes Service||Guide||State|
|Minikube (on AWS EC2)||Minikube||byok8s Minikube Guide||Finished|
|Google Cloud Platform (GCP)||Google Container Engine (GKE)||byok8s GCP GKE Guide||Finished|
|Amazon Web Services (AWS)||Elastic Container Service (EKS)||byok8s AWS EKS Guide||Unfinished|
|Digital Ocean (DO)||DO Kubernetes (DOK)||byok8s DO DOK Guide||Unfinished|