Kubeless

kubeless is a proof of concept to develop a serverless framework for Kubernetes.

They are other solutions, like fission form Platform9, funktion from Fabric8. There is also an incubating project at the ASF: OpenWhisk.

Kubeless stands out as we use a ThirdPartyResource to be able to create functions as custom resources. We then run an in-cluster controller that watches these custome resources and launches runtimes on-demand. These runtimes, dynamically inject the functions and make them available over HTTP or via a PubSub mechanism.

For PubSub we use Kafka. Currently we start Kafka and Zookeeper in a non-persistent setup. With kubeless you can create topics, and publish events that get consumed by the runtime.

Screencasts

From the December 8th 2016 Kubernetes Community meeting

Usage

Download kubeless from the release page. Then launch the controller. It will ask you if you are OK to do it. It will create a kubeless namespace and a lambda ThirdPartyResource. You will see a kubeless controller and a kafka controller running.

$ kubeless install
We are going to install the controller in the default namespace. Are you OK with this: [Y/N]
Y
INFO[0002] Initializing Kubeless controller...           pkg=controller
INFO[0002] Installing Kubeless controller into Kubernetes deployment...  pkg=controller
INFO[0002] Kubeless controller installation finished!    pkg=controller
INFO[0002] Installing Message Broker into Kubernetes deployment...  pkg=controller
INFO[0002] Message Broker installation finished!         pkg=controller

$ kubectl get pods --namespace=kubeless
NAME                                   READY     STATUS              RESTARTS   AGE
kafka-controller-2158053540-a7n0v      0/2       ContainerCreating   0          12s
kubeless-controller-1801423959-yow3t   0/2       ContainerCreating   0          12s

$ kubectl get thirdpartyresource
NAME             DESCRIPTION                                     VERSION(S)
lamb-da.k8s.io   Kubeless: Serverless framework for Kubernetes   v1

$ kubectl get lambdas

You are now ready to create functions. You need to run a proxy locally (soon to be fixed). Then you can use the CLI to create a function. Functions have two possible types:

http trigger (function will expose an HTTP endpoint)
pubsub trigger (function will consume event on a specific topic)

HTTP function

Here is a toy:

def foobar(context):
   print context.json
   return context.json

You create it with:

kubeless function create test --runtime python27 \
                              --handler test.foobar \
                              --from-file test.py \
                              --trigger-http

You will see the lambda custom resource created:

$ kubectl get lambdas
NAME      LABELS    DATA
test      <none>    {"apiVersion":"k8s.io/v1","kind":"LambDa","metadat...

PubSub function

Messages need to be JSON messages. A function cane be as simple as:

def foobar(context):
   print context.json
   return context.json

You create it the same way than an HTTP function except that you specify a --trigger-topic.

kubeless function create test --runtime python27 \
                              --handler test.foobar \
                              --from-file test.py \
                              --trigger-topic <topic_name>

Other commands

You can delete and list functions:

$ kubeless function delete <function_name>
$ kubeless function ls

You can create, list and delete PubSub topics:

$ kubeless topic create <topic_name>
$ kubeless topic delete <topic_name>
$ kubeless topic ls

To test your endpoints you can call the function directly with the kubeless CLI:

$ kubeless function call test --data {'kube':'coodle'}

Building

Building with go

you need go v1.5 or later.
if your working copy is not in your GOPATH, you need to set it accordingly.

$ go build -o kubeless main.go

Download kubeless package

$ go get -u github.com/skippbox/kubeless

Roadmap

This is still currently a POC, feel free to land a hand. We need to implement the following high level features:

Add other runtimes, currently only Python is supported
Deploy Kafka and Zookeeper using StatefulSets for persistency
Instrument the runtimes via Prometheus to be able to create pod autoscalers automatically
Get rid off the need for a proxy by switching k8s clients to monitor the custom resources.

dluc / kubeless