kash - bash for Kubernetes

Note: This is an experimental prototype. Do NOT use for production.

Run bash in your Kubernetes cluster with a shared file system. Use your Kubernetes cluster like a simple distributed operating system where Kubernetes is just a process scheduler.

What does this do

To set up your own kash-enabled cluster, see Cluster setup below.

Start up a bash instance with:

$ ./kash bash

Start up more bash instances on your cluster:

$ ./kash bash

Copy a file over:

$ ./kashcp ./myfile

kash will have Kubernetes try to distribute the bash instances across the available nodes. Every bash instance shares the same file system so it feels like you are working on just one machine except you are actually running bash across a cluster of machines.

You can see if the pods are distributed evenly across nodes with:

kubectl get pods -o wide

An alternate name could be BUG (bash using GKE) so you could run it by calling ./bug bash.

Demo

How it works

kash uses a Network File System to share the persistent disk across Pods running in multiple nodes. The Pods are also set to distribute across the nodes with anti-affinity towards other kash Pods. The shared file system is mounted at kash. /kash/.bashrc is the location of the bashrc and bash history is also shared among instances.

Example - calculate PI

First, use kash to copy the examples/calculatepi/*.py scripts into /kash/*.py:

$ ./kashcp examples/calculatepi/display_pi.py
$ ./kashcp examples/calculatepi/run_trials.py

Then, in one window, run:

$ ./kash python
bash# mkdir trials
bash# chmod +x ./display_pi
bash# ./display_pi trials

In another window, run:

$ ./kash python
bash# chmod +x ./run_trials
bash# ./run_trials trials/$RANDOM

Watch as the trials come in and the estimated PI is displayed.

Try running some more trial instances, which would distribute the workload across the cluster:

$ ./kash python
bash# ./run_trials trials/$RANDOM

Cluster setup

Set up kubectl

You can view your GKE clusters with:

gcloud container clusters list

Set GKE_NAME to the name of the cluster you want to use and GKE_ZONE to the zone of that cluster:

export GKE_NAME=??? (ie. name of cluster)
export GKE_ZONE=??? (ie. location of cluster, eg. us-west1-c)

Give kubectl credentials to control that cluster:

gcloud container clusters get-credentials ${GKE_NAME} --zone=${GKE_ZONE}

Create a kash namespace and have kubectl use that namespace:

export GKE_CONTEXT=$(kubectl config current-context)
kubectl apply -f k8s/kash-namespace.yaml
kubectl config set-context kash --namespace=kash \
    --cluster=${GKE_CONTEXT} --user=${GKE_CONTEXT}
kubectl config use-context kash

Set up the shared persistent disk

Create a GCE persistent disk to use as the shared disk. Change KASH_DISK_SIZE to your desired size (in GB).

export KASH_DISK_SIZE=10
gcloud compute disks create --size=${KASH_DISK_SIZE}GB --zone=${GKE_ZONE} kash-nfs

Run the NFS server:

kubectl apply -f k8s/nfs-server.yaml
cat k8s/nfs-volume.yaml |
    sed -e "s/{DISK_SIZE}/${KASH_DISK_SIZE}/" |
    kubectl apply -f -

coollog / kash