Using Keda and Azure Functions on Openshift 4

The below will walk you through creating a Kafka instance, publishing your function to a cluster, and then publishing an agent to pull data from Twitter and publish it to Kafka. As the events land in Kafka, the function will automatically trigger and scale. Feel free to skip portions if they already exist in your cluster.

Prerequisites

• A running OpenShift 4 cluster
• oc and kubectl pointing to that cluster (can be achieved via oc login or setting KUBECONFIG accordingly)
• A Dockerhub account and docker being logged into that account to be able to push images.

Install Keda

Install the operator itself

Locate and install KEDA on OperatorHub.io In your OpenShift Admin Console UI, head to Operators/OperatorHub and search for the KEDA operator. Install it.

Create a KedaController instance

Create your KEDA instance in the OpenShift Admin console by creating KedaController resource or by applying the following YAML.

apiVersion: keda.sh/v1alpha1
kind: KedaController
metadata:
  name: keda
  namespace: keda
spec:
  watchNamespace: ""
  logLevel: info
  logEncoder: console
  logLevelMetrics: "0"

oc -n keda get pods should now show the KEDA Operator and Metrics ApiServer being deployed.

$ oc -n keda get pods
NAME                                      READY   STATUS    RESTARTS   AGE
keda-metrics-apiserver-6d749bfb59-ps2dj   1/1     Running   0          3h45m
keda-olm-operator-b6bd5b88f-gkmmh         1/1     Running   0          3h45m
keda-operator-59dcf989d6-k8z7p            1/1     Running   0          3h45m

Install Kafka using the Strimzi operator

Install the operator itself

Follow the install instructions for the Strimzi Kafka operator on OperatorHub.io. Alternatively, in your OpenShift Admin Console UI, head to Operators/OperatorHub and search for the Strimzi Kafka operator. Install it.

Create a Kafka instance

Create your Kafka instance by applying the following YAML. These are the default settings of creating a Kafka instance via the UI. You can create it there as well via Catalog/Installed Operators/Strimzi Kafka and navigating to the Create new button on the Kafka resource.

apiVersion: kafka.strimzi.io/v1alpha1
kind: Kafka
metadata:
  name: my-cluster
  namespace: openshift-operators
spec:
  kafka:
    version: 2.7.0
    replicas: 3
    listeners:
      plain: {}
      tls: {}
    config:
      offsets.topic.replication.factor: 3
      transaction.state.log.replication.factor: 3
      transaction.state.log.min.isr: 2
    storage:
      type: ephemeral
  zookeeper:
    replicas: 3
    storage:
      type: ephemeral
  entityOperator:
    topicOperator: {}
    userOperator: {}

You should now see Kafka pods appearing in the openshift-operators namespace.

$ oc -n openshift-operators get pods
NAME                                          READY     STATUS    RESTARTS   AGE
my-cluster-entity-operator-55c4676687-5qrvt   3/3       Running   0          5m27s
my-cluster-kafka-0                            2/2       Running   0          6m8s
my-cluster-kafka-1                            2/2       Running   0          6m8s
my-cluster-kafka-2                            2/2       Running   0          6m8s
my-cluster-zookeeper-0                        2/2       Running   0          6m59s
my-cluster-zookeeper-1                        2/2       Running   0          6m59s
my-cluster-zookeeper-2                        2/2       Running   0          6m59s
strimzi-cluster-operator-66d7bd49f8-qlmbx     1/1       Running   0          11m

Create a Kafka topic

Now create a topic twitter by applying the following YAML. Note the connections being made to the cluster we created above via the strimzi.io/cluster label. These are the default settings of creating a Kafka topic via the UI. You can create it there as well via Catalog/Installed Operators/Strimzi Kafka and navigating to the Create new button on the Kafka Topic resource.

apiVersion: kafka.strimzi.io/v1alpha1
kind: KafkaTopic
metadata:
  name: twitter
  labels:
    strimzi.io/cluster: my-cluster
  namespace: openshift-operators
spec:
  partitions: 10
  replicas: 3
  config:
    retention.ms: 604800000
    segment.bytes: 1073741824

Deploy a Kafka client

To be able to play with this Kafka instance later, install a Kafka client into the cluster by applying client.yaml.

apiVersion: v1
kind: Pod
metadata:
  name: kafka-client
  namespace: default
spec:
  containers:
  - name: kafka-client
    image: confluentinc/cp-kafka:5.2.1
    command:
      - sh
      - -c
      - "exec tail -f /dev/null"

Using the sample

Clone the TypeScript based Kafka example and navigate to it

git clone https://github.com/kedacore/sample-typescript-kafka-azure-function
cd sample-typescript-kafka-azure-function

Edit the function.json file to point to the Kafka cluster created above.

diff --git a/KafkaTwitterTrigger/function.json b/KafkaTwitterTrigger/function.json
index 8b1da0c..3c96501 100644
--- a/KafkaTwitterTrigger/function.json
+++ b/KafkaTwitterTrigger/function.json
@@ -5,7 +5,7 @@
       "direction": "in",
       "name": "event",
       "topic": "twitter",
-      "brokerList": "kafka-cp-kafka-headless.default.svc.cluster.local:9092",
+      "brokerList": "my-cluster-kafka-bootstrap.openshift-operators:9092",
       "consumerGroup": "functions",
       "dataType": "binary"
     }

Deploy the function

func kubernetes deploy --name twitter-function --typescript --registry $DOCKER_HUB_USERNAME 

Alternatively, you can build and publish the image on your own and provide the --image-name instead of the --registry

Validate the function is deployed

oc get deploy

You should see the twitter-function is deployed, but since there are no Twitter events it has 0 replicas.

Generate test twitter data

As our function is listening for arbitrary data on the twitter topic and only analyzes the text field of the incoming JSON decoded data you can easily create a test workload to test the function works as intended. Using the the Kafka client installed earlier you can create messages on the topic that look like tweets to the function:

oc exec kafka-client -- sh -c 'echo "{\"text\": \"this is not something nice to say\"}" | kafka-console-producer --broker-list my-cluster-kafka-brokers.openshift-operators:9092 --topic twitter'

After doing that, you should see a pod appear to handle the event.

$ oc get pods
NAME                                READY     STATUS    RESTARTS   AGE
twitter-function-5d684fd7b7-8bqwn   1/1       Running   0          10s

Now checking the logs of that pod you'll see the event coming in and being analyzed.

$ oc logs twitter-function-5d684fd7b7-8bqwn
...
info: Function.KafkaTwitterTrigger.User[0]
      Kafka trigger fired!
info: Function.KafkaTwitterTrigger.User[0]
      Tweet analyzed
      Tweet text: this is not something nice to say
      Sentiment: 0.42857142857142855
info: Function.KafkaTwitterTrigger[0]
      Executed 'Functions.KafkaTwitterTrigger' (Succeeded, Id=19c8a1c7-ad66-4d35-b5b6-1cc90327995e)
...

Using actual twitter data

Follow the rest of the sample from the Feed twitter data step on.