Continuous Testing Kafka with Xinfra Monitor in Kubernetes
Monitors kafka as part of continuously cheking how the cluster performs. The challenge with this is to create script to produce and consume with some ludicrous amount of works. Instead of creating stuff in house, a mate refering a fancy tool called Kafka Monitor.
Kafka monitor, or now known as Xinfra Monitor is an open source software created by folks in Linkedin. This tools provide the produce, consume and offset test and measurement.
The fun part here, is Jolokia. So, Xinfra Monitor (Kafka Monitor), expose mbeans through Jolokia, so it can be accessed easily by any monitoring tools like Prometheus (plus Jolokia Exporter) or paid one like Datadog.
Kafka Xinfra Monitor Deployment #
In this case I would use kubernetes cluster to host the xinfra monitor, and also use strimzi as kafka and zookeeper operator. For the sake of simplicity, I will host everything on the kubernetes, minikube -in this example.
Nice to have locally :
Base assumption :
- kafka cluster endpoint :
my-cluster-kafka-bootstrap, Port :
- zookeeper cluster endpoint :
my-cluster-zookeeper-client, Port :
- Run xinfra monitor in single cluster monitoring mode
Deploy the Kafka Monitor / Xinfra Monitor #
Initial steps : #
Clone source code
git clone email@example.com:linkedin/kafka-monitor.git
Checkout to latest release tag. e.g
git checkout 2.5.10
Modify the Dockerfile #
If you can see. There is Dockerfile here.
FROM anapsix/alpine-java MAINTAINER firstname.lastname@example.org WORKDIR /opt/kafka-monitor ADD build/ build/ ADD bin/xinfra-monitor-start.sh bin/xinfra-monitor-start.sh ADD bin/kmf-run-class.sh bin/kmf-run-class.sh ADD config/xinfra-monitor.properties config/xinfra-monitor.properties ADD config/log4j2.properties config/log4j2.properties ADD docker/kafka-monitor-docker-entry.sh kafka-monitor-docker-entry.sh ADD webapp/ webapp/ CMD ["/opt/kafka-monitor/kafka-monitor-docker-entry.sh"]
Side note : in the Dockerfile above, you will find out that the container will use the properties which can be configured in
config/xinfra-monitor.properties. With this config, you can set the configuration in granular way, also possible for you to monitor multiple kafka cluster - which I don’t really need in this case. Much better way to do this approach, is to convert the config file to kubernetes configmap.
In my case, I will not use the default Dockerfile approach, since I only want to monitor a single Kafka Cluster. So I modified the Dockerfile to this :
FROM anapsix/alpine-java WORKDIR /opt/kafka-monitor ADD build/ build/ ADD bin/single-cluster-monitor.sh bin/single-cluster-monitor.sh ADD bin/kmf-run-class.sh bin/kmf-run-class.sh ADD config/log4j2.properties config/log4j2.properties ENTRYPOINT ["/opt/kafka-monitor/bin/single-cluster-monitor.sh"]
Build Docker image #
Build the docker image using existing makefile (yea you can use a vanilla docker build command too)
# Go to the docker folder - inside the kafka-monitor repo cd docker PREFIX=robeevanjava/kafka-monitor TAG=v2.5.10 make container
You can change the
PREFIXenvironment variable with your docker hub username or even to quay.io. Change the tag to any version you want.
Push the docker image to docker registry
docker push robeevanjava/kafka-monitor:v2.5.10
Deploying the Kubernetes Objects #
I have create the kubernetes manifest here: xinfra-monitor-manifest.yaml
If you want to use exactly that manifest mentioned above, just do
kubectl apply -f https://git.io/JwLDr -n kafka-monitor
If you need to change the kafka and zookeeper cluster, please change the argument lines here :
containers: - name: kafka-monitor image: robeevanjava/kafka-monitor:v2.5.10 imagePullPolicy: Always env:  args: - --topic - kafkamonitor-topic-test - --broker-list - my-cluster-kafka-bootstrap.kafka.svc.cluster.local:9092 - --zookeeper - my-cluster-zookeeper-client.kafka.svc.cluster.local:2181
Change the topic, broker-list and zookeeper parameter accordingly.
In the manifest I only expose 1 port on the kubernetes service, and that is for Jolokia ref, a software that exposing mbeans to http. So it enable us to query the mbeans metric through http request, with json returning value.
To find the available metric, try to do these steps :
Do port-forwarding to jolokia service port
kubectl port-forward svc/kafka-monitor 8778:8778 -n kafka-monitor
Make request to jolokia port
# Get list of available metric curl "http://localhost:8778/jolokia/list" # Json pretty print curl "http://localhost:8778/jolokia/list" | jq . # Read to a sample metric curl "http://localhost:8778/jolokia/read/kafka.consumer:client-id=kmf-consumer,type=consumer-metrics"
Now you have kafka monitor run on your kubernetes cluster.
Optional Steps #
Some of the organization uses prometheus as “default” observability toolings. We can use jolokia-exporter
You can clone the repo and build your very own Jolokia docker image and push it to your favourite container registry, but I will use the prebuilt one :)
I have kubernetes manifest here
So to apply it, run
kubectl apply -f https://git.io/JwLim -n kafka-monitor
Then you can just do either ServiceMonitor or write a prometheus scrape config like explained in the example here
PS : If the git.io URL is broken, please refer to each linked gist file above :D