Kubeless Custom Runtime

Example on how to create a custom runtime for kubeless.

Project explanation

This kubeless runtime image is using python 3.6 (for python 3.7 some different config must be used, check below) with Bottle as a wsgi and has included some extra files. These files are located in the ./custom-lib/ folder and consist of a csv parser (lib.py) and a test CSV (test.csv) file. When building the docker image, this folder will be copied inside the runtime image and will be used by the kubeless function.

The kubeless function (test.py) parses and counts the number of lines and returns this value to the kubeless wrapper.

Deploying

1. Build and push docker image

In order to build the docker image just run

docker build . -f python3.6.Dockerfile -t <your_container_registry>/<image_name>:<image_tag>
docker push <your_container_registry>/<image_name>:<image_tag>

UPDATED For python 3.7:

docker build . -f python3.7.Dockerfile -t <your_container_registry>/<image_name>:<image_tag>
docker push <your_container_registry>/<image_name>:<image_tag>

Note: If you like to test it without building your own image, you can use this one: defrox/runtime:v0.2 or for python 3.7 use defrox/runtime:v0.3

2. Update the kubeless-config configmap

In order to use the custom runtime, it is necessary to add the custom runtime in the kubeless-config configmap.

1. Get the current kubeless-config configmap

   kubectl get kubeless-config -n kubeless -o yaml > kubeless-config.yaml
   

2. Add the custom image configuration in json format inside the runtime-images value

   {
       "ID": "<custom-runtime>",
       "compiled": false,
       "versions": [
           {
           "name": "<custom-runtime>",
           "version": "<custom-runtime-version>",
           "runtimeImage": "<your_container_registry>/<image_name>:<image_tag>",
           "initImage": "python:3.6"
           }
       ],
       "depName": "requirements.txt",
       "fileNameSuffix": ".py"
   }
   

   UPDATED For python 3.7 use this image configuration:

   {
       "ID": "<custom-runtime>",
       "depName": "requirements.txt",
       "fileNameSuffix": ".py",
       "versions": [{
           "images": [{
               "command": "pip install --prefix=$KUBELESS_INSTALL_VOLUME -r $KUBELESS_DEPS_FILE",
               "image": "python:3.7",
               "phase": "installation"
           }, {
               "env": {
                   "PYTHONPATH": "$(KUBELESS_INSTALL_VOLUME)/lib/python3.7/site-packages:$(KUBELESS_INSTALL_VOLUME)"
               },
               "image": "<your_container_registry>/<image_name>:<image_tag>",
               "phase": "runtime"
           }],
           "name": "<custom-runtime>",
           "version": "3.7"
       }]
   }
   

3. Delete current kubeless-config configmap and deploy new one (else k8s won't update the values)

   kubectl delete -f ./kubeless-config.yaml && kubectl create -f ./kubeless-config.yaml
   

4. Restart the kubeless-controller-manager pod

   PODNAME=$(kubectl get pod -l kubeless=controller -o jsonpath="{.items[0].metadata.name}")
   kubectl get pod $PODNAME -n kubeless -o yaml | kubectl replace --force -f -
   

3. Deploy the kubeless function

Use the kubeless-cli and run

kubeless function deploy --from-file test.py --handler test.counter --runtime <custom-runtime><custom-runtime-version> counter 

4. Test the kubeless function

Check if the function has been deployed and is ready

kubeless function ls counter

Call the function with kubeless-cli

kubeless function call counter --data '{"file": "lib/test.csv"}'

Or curl directly with

kubectl proxy -p 8080 & \
curl -L --data '{"file": "lib/test.csv"}' \
  --header "Content-Type:application/json" \
  localhost:8080/api/v1/namespaces/default/services/counter:http-function-port/proxy/

5. Setting up a private registry

Create a secret to configure your private docker registry:

kubectl create secret docker-registry kubeless-registry-credentials \
  --docker-server=<url-to-your-private-registry> \
  --docker-username=<your-user> \
  --docker-password=<your-password> \
  --docker-email=<your-user@example.com>

Or if you want to use a secret with a custom name, then create it with:

kubectl create secret docker-registry <your-secret-name> \
  --docker-server=<url-to-your-private-registry> \
  --docker-username=<your-user> \
  --docker-password=<your-password> \
  --docker-email=<your-user@example.com>

and specify provision-image-secret with the new secret name ìn the image configuration.

NOTE: check the Known limitations at the end of the doc: https://kubeless.io/docs/building-functions/

---

Related links:

Tutorial article

https://github.com/kubeless/kubeless/blob/master/docs/implementing-new-runtime.md

https://github.com/kubeless/kubeless/blob/master/docs/runtimes.md#use-a-custom-runtime

https://kubeless.io/docs/implementing-new-runtime/

https://github.com/kubeless/runtimes

https://kubeless.io/docs/building-functions/

https://kubeless.io/docs/function-controller-configuration/