Kafka-connector

The Kafka connector connects OpenFaaS functions to Kafka topics.

Goals:

• Allow functions to subscribe to Kafka topics
• Ingest data from Kafka and execute functions
• Work with the OpenFaaS REST API / Gateway
• Formulate and validate a generic "connector-pattern" to be used for various event sources like Kafka, AWS SNS, RabbitMQ etc

Conceptual design

This diagram shows the Kafka connector on the left hand side. It is responsible for querying the API Gateway for a list of functions. It will then build up a map or table of which functions have advertised an interested in which topics.

When the connector hears a message on an advertised topic it will look that up in the reference table and find out which functions it needs to invoke. Functions are invoked only once and there is no re-try mechanism. The result is printed to the logs of the Kafka connector process.

The cache or list of functions <-> topics is refreshed on a periodic basis.

Building

export TAG=0.3.3
make build push

Try it out

Deploy on Kubernetes

The following instructions show how to run kafka-connector on Kubernetes.

Deploy a function with a topic annotation:

$ faas store deploy figlet --annotation topic="faas-request" --gateway <faas-netes-gateway-url>

If deploying via a stack.yml then add an annotation key called topic with a value of faas-request.

  my-function:
...
    annotations:
      topic: payment

For further details, please see the YAML Reference section of the OpenFaaS Docs

Deploy Kafka:

You can run the zookeeper, kafka-broker and kafka-connector pods with:

kubectl apply -f ./yaml/kubernetes/

If you already have Kafka then update ./yaml/kubernetes/connector-dep.yml with your Kafka broker address and then deploy only that file:

kubectl apply -f ./yaml/kubernetes/connector-dep.yml

Alternatively you can use the kafka-connector helm chart

Then use the broker to send messages to the topic:

BROKER=$(kubectl get pods -n openfaas -l component=kafka-broker -o name|cut -d'/' -f2)
kubectl exec -n openfaas -t -i $BROKER -- /opt/kafka_2.12-0.11.0.1/bin/kafka-console-producer.sh --broker-list kafka:9092 --topic faas-request

hello world

Once you have connected, each new line will be a message published.

If you have an error

error: unable to upgrade connection: container not found ("kafka")

just wait and retry.

You can verify the proper path of the publisher script by getting the shell of the running broker:

$ kubectl exec -n openfaas -t -i $BROKER -- sh 
/ # find | grep producer

./opt/kafka_2.12-0.11.0.1/config/producer.properties
./opt/kafka_2.12-0.11.0.1/bin/kafka-verifiable-producer.sh
./opt/kafka_2.12-0.11.0.1/bin/kafka-producer-perf-test.sh
./opt/kafka_2.12-0.11.0.1/bin/kafka-console-producer.sh

With the $BROKER variable still set, view the list of messages on a given topic:

$ kubectl exec -n openfaas -t -i $BROKER -- /opt/kafka_2.12-0.11.0.1/bin/kafka-console-consumer.sh --bootstrap-server kafka:9092 --topic faas-request --from-beginning

Now check the connector logs to see the figlet function was invoked:

CONNECTOR=$(kubectl get pods -n openfaas -o name|grep kafka-connector|cut -d'/' -f2)
kubectl logs -n openfaas -f --tail 100 $CONNECTOR

2018/08/08 16:54:35 Binding to topics: [faas-request]
2018/08/08 16:54:38 Syncing topic map
Rebalanced: &{Type:rebalance start Claimed:map[] Released:map[] Current:map[]}
Rebalanced: &{Type:rebalance OK Claimed:map[faas-request:[0]] Released:map[] Current:map[faas-request:[0]]}

2018/08/08 16:54:41 Syncing topic map
2018/08/08 16:54:44 Syncing topic map
2018/08/08 16:54:47 Syncing topic map

[#53753] Received on [faas-request,0]: 'hello world.'
2018/08/08 16:57:41 Invoke function: figlet
2018/08/08 16:57:42 Response [200] from figlet  
 _          _ _                            _     _ 
| |__   ___| | | ___   __      _____  _ __| | __| |
| '_ \ / _ \ | |/ _ \  \ \ /\ / / _ \| '__| |/ _` |
| | | |  __/ | | (_) |  \ V  V / (_) | |  | | (_| |
|_| |_|\___|_|_|\___/    \_/\_/ \___/|_|  |_|\__,_|
                                                   

Deploy on Swarm

Deploy the stack which contains Kafka and the connector:

docker stack deploy kafka -c ./yaml/connector-swarm.yml

• Deploy or update a function so it has an annotation topic=faas-request or some other topic

As an example:

$ faas store deploy figlet --annotation topic="faas-request"

The function can advertise more than one topic by using a comma-separated list i.e. topic=topic1,topic2,topic3

• Publish some messages to the topic in question i.e. faas-request

Instructions are below for publishing messages

• Watch the logs of the kafka-connector

Trigger a function via a topic with exec

You can use the kafka container to send a message to the topic.

SERVICE_NAME=kafka_kafka
TASK_ID=$(docker service ps --filter 'desired-state=running' $SERVICE_NAME -q)
CONTAINER_ID=$(docker inspect --format '{{ .Status.ContainerStatus.ContainerID }}' $TASK_ID)
docker exec -it $CONTAINER_ID kafka-console-producer --broker-list kafka:9092 --topic faas-request

hello world

Generate load on the topic

You can use the sample application in the producer folder to generate load for a topic.

Make sure you have some functions advertising an interest in that topic so that they receive the data.

Note: the producer must run inside the Kubernetes or Swarm cluster in order to be able to access the broker(s).

Monitor the results

Once you have generated some requests or start a load-test you can watch the function invocation rate increasing in Prometheus or watch the logs of the container.

Prometheus

You can open the Prometheus metrics or Grafana dashboard for OpenFaaS to see the functions being invoked.

Watch the logs

docker service logs kafka_connector -f

Topics
[__consumer_offsets faas-request] <nil>
2018/03/24 12:42:58 Binding to topics: [faas-request]
Rebalanced: &{Type:rebalance start Claimed:map[] Released:map[] Current:map[]}
Rebalanced: &{Type:rebalance OK Claimed:map[faas-request:[0]] Released:map[] Current:map[faas-request:[0]]}

2018/03/24 17:04:40 Syncing topic map
[#53694] Received on [faas-request,0]: 'Test the function.'
2018/03/24 17:04:41 Invoke function: figlet
2018/03/24 17:04:42 Response [200] from figlet  
|_   _|__  ___| |_  | |_| |__   ___   / _|_   _ _ __   ___| |_(_) ___  _ __    
  | |/ _ \/ __| __| | __| '_ \ / _ \ | |_| | | | '_ \ / __| __| |/ _ \| '_ \   
  | |  __/\__ \ |_  | |_| | | |  __/ |  _| |_| | | | | (__| |_| | (_) | | | |_ 
  |_|\___||___/\__|  \__|_| |_|\___| |_|  \__,_|_| |_|\___|\__|_|\___/|_| |_(_)
                  

Note: If the broker has a different name from kafka you can pass the broker_host environmental variable. This exclude the port.

Configuration

This configuration can be set in the YAML files for Kubernetes or Swarm.

env_var	description
upstream_timeout	Go duration - maximum timeout for upstream function call
rebuild_interval	Go duration - interval for rebuilding function to topic map
topics	Topics to which the connector will bind
gateway_url	The URL for the API gateway i.e. http://gateway:8080 or http://gateway.openfaas:8080 for Kubernetes
broker_host	Default is kafka:9092. You can configure this with our own host with port
print_response	Default is true - this will output information about the response of calling a function in the logs, including the HTTP status, topic that triggered invocation, the function name, and the length of the response body in bytes
print_response_body	Default is true - this will print the body of the response of calling a function to stdout
topic_delimiter	Default is , - Specifies character upon which to split the topic annotation when subscribing a function to mulitple topics