Awesome Open Source
Awesome Open Source

TriggerMesh Knative Lambda Runtime

Knative Lambda Runtimes (e.g KLR, pronounced clear) are Tekton Tasks that can be used to run an AWS Lambda function in a Kubernetes cluster installed with Knative.

The execution environment where the AWS Lambda function runs is a clone of the AWS Lambda cloud environment thanks to a custom AWS runtime interface and some inspiration from the LambCI project.

With these templates, you can run your AWS Lambda functions as is in a Knative powered Kubernetes cluster.

The examples below use the tm CLI to interact with Knative but one could also use kubectl.

Docker registry for builds

To combine the runtime with your source, the examples below produce a new Docker image each time. While these images can be considered temporary, builds must be pushed to a Docker registry in order for Kubernetes to be able to pull. By default tm uses Knative Local Registry, equivalent to adding --registry-host knative.registry.svc.cluster.local to the commands below, so that builds can run without registry authentication. To override, set --registry-secret according to tm docs.


Concurrency in KLR represented by two components: parallel running bootstrap processes per container and Knative container concurrency model. By default AWS runtime interface fires up 4 bootstrap processes (functions, in other words) and allows multiple concurrent requests (containerConcurrency: 0) to be handled by each container. Default concurrency configuration can be changed on function deployment or update using tm deploy service command parameters:

--concurrency <N> - sets Knative service containerConcurrency value to N

--build-argument INVOKER_COUNT=<N> - passes number of parallel running functions to AWS lambda runtime

Values for these two parameters should be calculated individually for each function and depends on operation characteristics. Knative autoscaling is another important factor that affects service performance, but right now KLR uses default autoscaling configuration.


NOTE: all examples below work with Local Registry. If you don't have local registry in knative cluster, you can use external registry as discribed in CLI documentation


  1. Install runtime
tm deploy task -f
  1. Deploy function
tm deploy service python-test -f \
                              --build-template knative-python37-runtime \
                              --build-argument DIRECTORY=aws-python-simple-http-endpoint \
                              --build-argument HANDLER=handler.endpoint \
  1. Execute function via public URL

{"statusCode": 200, "body": "{\"message\": \"Hello, the current time is 06:45:49.174383\"}"}

To use Python 2.7 runtime simply replace version tag in step 1 and 2 with python27 and knative-python27-runtime accordingly.


  1. Install node 4.3 runtime
tm deploy task -f
  1. Deploy example function
tm deploy service node4-test -f \
                             --build-template knative-node4-runtime \
                             --build-argument DIRECTORY=aws-node-serve-dynamic-html-via-http-endpoint \
                             --build-argument HANDLER=handler.landingPage \
  1. Function is ready

{"statusCode":200,"headers":{"Content-Type":"text/html"},"body":"\n  <html>\n    <style>\n      h1 { color: #73757d; }\n    </style>\n    <body>\n      <h1>Landing Page</h1>\n      <p>Hey Unknown!</p>\n    </body>\n  </html>"}

Node 10 with async handler

  1. Prepare function code
mkdir example-lambda-nodejs
cd example-lambda-nodejs
cat > handler.js <<EOF
async function justWait() {
  return new Promise((resolve, reject) => setTimeout(resolve, 100));

module.exports.sayHelloAsync = async (event) => {
  await justWait();
  return {hello: event && || "Missing a name property in the event's JSON body"};

node -e "require('./handler').sayHelloAsync({}).then(h => console.log(h))"
  1. Install node10 runtime
tm deploy task -f
  1. Deploy function
tm deploy service node-lambda -f . --build-template knative-node10-runtime \
                                   --build-argument HANDLER=handler.sayHelloAsync \


curl --data '{"name": "Foo"}'
# {"hello":"Foo"}


  1. Prepare function code

You will create a main.go file in the example-lambda-go directory.

Create the directory and get into it:

mkdir example-lambda-go
cd example-lambda-go

Copy and Paste the following into a main.go file:

package main

import (

type MyEvent struct {
        Name string `json:"name"`

func HandleRequest(ctx context.Context, name MyEvent) (string, error) {
        return fmt.Sprintf("Hello %s!", name.Name ), nil

func main() {
  1. Install Go runtime
tm deploy task -f
  1. Deploy function
tm deploy service go-lambda -f . --build-template knative-go-runtime --wait


curl --data '{"Name": "Foo"}'
"Hello Foo!"

To have more control over the go project dependecies, KLR runtime will read and "ensure" Gopkg.toml file if it's available in the project's root. Without toml file, only go get will be executed before build. If the project has dependencies stored in a private repository, you can create k8s secret with SSH key that will be used for git clone operation:

cat ~/.ssh/id_rsa | tm set git-auth

where ~/.ssh/id_rsa is a path to SSH private key associated with your git account


  1. Install Ruby 2.5 runtime
tm deploy task -f
  1. Deploy example function
tm deploy service ruby-lambda -f --build-argument DIRECTORY=aws-ruby-simple-http-endpoint --build-argument HANDLER=handler.endpoint --build-template knative-ruby25-runtime --wait
  1. Function is ready
{"statusCode":200,"body":"{\"date\":\"2019-01-14 19:10:29 +0000\"}"}


  1. Clone serverless examples repository, change directory to aws-java-simple-http-endpoint, edit serverless.yaml:
service: aws-java-sample
description: Triggermesh Java8 sample
  name: triggermesh
    source: ../aws-java-simple-http-endpoint
    - HANDLER=com.serverless.Handler
  1. Remove incompatible leftovers from build.gradle file:
task wrapper(type: Wrapper) {
    gradleVersion = '3.2.1'
  1. Deploy:
tm deploy

Function is ready:

curl -d '{"event":"foo"}'
{"message":"Hello, the current time is Tue Apr 07 13:59:17 GMT 2020"}

Run in Docker

For cases in which the use of additional components (tm CLI, Tekton, Knative, k8s) is undesirable, it is possible to build a KLR function as a standalone Docker container and run it in any environment. To do this, you should extract the Dockerfile from the runtime you are interested in, put it in the directory with your function, update the handler variable, and build the container. Here are Dockerfile definitions for all runtimes:

Let's build a Python 3.7 function as an example:

  1. Create directory and save your function code there:
  mkdir python
  cd python
  cat > <<EOF
import json
import datetime
def endpoint(event, context):
  current_time =
  body = {
      "message": "Hello, the current time is " + str(current_time)
  response = {
      "statusCode": 200,
      "body": json.dumps(body)
  return response
  1. Extract the runtime's Dockerfile, store it in the same directory, and update the _HANDLER variable:
  cat > Dockerfile <<EOF
  ENV _HANDLER handler.endpoint
  COPY . .
  RUN if [ -f requirements.txt ]; then pip3.7 install -r requirements.txt ;fi
  ENTRYPOINT ["/opt/aws-custom-runtime"]

The _HANDLER variable in most cases consists of the filename without the file extension, and the function name.

  1. Build, run, test:
  docker build -t python-klr-image .
  docker run -d --rm --name python-klr-container python-klr-image
  # following command will work if you use Docker bridge network and you have jq tool
  # otherwise, you should get the container address manually
  curl $(docker inspect python-klr-container | jq .[].NetworkSettings.Networks.bridge.IPAddress -r):8080

The response will contain a JSON document with the current time.

  1. Cleanup:
  docker stop python-klr-container

The image can be pushed to the container registry and used to create Knative Service:

kubectl apply -f - <<EOF
kind: Service
  name: python-klr-service
        - image: <python-klr-image>


We would love your feedback on this tool so don't hesitate to let us know what is wrong and how we could improve it, just file an issue

Code of Conduct

This plugin is by no means part of CNCF but we abide by its code of conduct

Get A Weekly Email With Trending Projects For These Topics
No Spam. Unsubscribe easily at any time.
python (54,447
kubernetes (1,772
serverless (660
lambda (247
management (133
amazon (122
knative (28