Authors:
Adrian Ryt
Dawid Grapa
Kamil Kurowski
Ryszard Pręcikowski
Year, Group: 2023, pt 16:40
Kafka Operators on Kuberentes Prepare evaluation test suite for kafka running on kuberentes behind the operator. Select real life uses cases, with real-life workload. Perform test with different Kafka setting. Select the most meaningful metrics for evaluation
Kafka Operators are software tools that help to automate the deployment, management, and operation of Apache Kafka clusters on Kubernetes platforms. Operators are essentially Kubernetes native controllers that extend the Kubernetes API to manage applications or services that are more complex than simple stateless or stateful workloads. By using Operators, developers and DevOps teams can simplify the deployment and management of Kafka clusters, reduce the operational overhead, and improve the resilience and scalability of their applications.
Kubernetes is a popular container orchestration platform that provides a framework for deploying, scaling, and managing containerized applications. Kubernetes uses a declarative API model to describe the desired state of the system and manages the deployment and operation of the applications based on this specification. Strimzi is a Kubernetes Operator that specializes in managing Apache Kafka clusters on Kubernetes. Strimzi provides a set of Kubernetes Custom Resource Definitions (CRDs) that can be used to define and manage Kafka clusters, topics, users, and other resources.
Amazon Elastic Kubernetes Service (EKS) is a managed Kubernetes service provided by AWS. EKS enables customers to easily deploy and operate Kubernetes clusters on AWS infrastructure without having to manage the underlying infrastructure themselves. EKS provides a fully managed, highly available, and scalable Kubernetes control plane that simplifies cluster deployment and management. EKS also integrates with other AWS services such as Elastic Load Balancing, Amazon S3, and Amazon RDS to provide a seamless experience for deploying and running containerized applications on AWS.
By using Strimzi on EKS, developers and DevOps teams can take advantage of the benefits of both technologies. Strimzi provides the Kafka Operator functionality, which simplifies the deployment and management of Kafka clusters on Kubernetes, while EKS provides a managed Kubernetes platform that eliminates the need to manage the underlying infrastructure. This combination enables developers and DevOps teams to focus on building and operating their applications rather than managing the infrastructure.
In summary, Kafka Operators on Kubernetes, such as Strimzi, provide a powerful way to simplify the deployment and management of Kafka clusters on Kubernetes platforms like EKS. By leveraging these technologies, organizations can improve the scalability and resilience of their applications while reducing the operational overhead of managing complex distributed systems.
To run Kafka on Kubernetes we will use:
- Kubernetes deployed on EKS (K8s on EKS)
- Kafka
- Zookeeper
- Strimzi
- Grafana
Apache Kafka is a popular distributed streaming platform used for real-time data feeds. Deploying and managing Kafka on Kubernetes can be challenging, but the Strimzi operator simplifies the process. In this case study, we will explore the preparation of an evaluation test suite for Kafka running on Kubernetes behind the Strimzi operator. We will select real-life use cases with real-life workloads. Additionally, we will use Grafana to visualize the data.
The main objectives of this case study are:
- To prepare an evaluation test suite for Kafka running on Kubernetes behind the Strimzi operator.
- To select real-life use cases with real-life workloads and perform tests with different Kafka settings.
- To identify the most meaningful metrics for evaluation.
- To use Grafana to visualize the data and identify any issues or bottlenecks in the performance of Kafka on Kubernetes.
To achieve the objectives of this case study, we will follow the below methodology:
-
Identify the use cases: We will select real-life use cases with real-life workloads that are relevant to the organization's business operations.
-
Prepare the test environment: We will set up a test environment consisting of a Kubernetes cluster, the Strimzi operator, Apache Kafka, and Grafana.
-
Conduct the tests: We will execute the tests and collect the data on various performance metrics. We will use Grafana to visualize the data and identify any issues or bottlenecks.
-
Analyze the results: We will analyze the test results to identify the most meaningful metrics for evaluation and identify any issues or bottlenecks that need to be addressed.
-
Present the findings: We will present findings, including recommendations for improving the performance and scalability of Kafka running on Kubernetes behind the Strimzi operator.
This case study will provide valuable insights into the preparation of an evaluation test suite for Kafka running on Kubernetes behind the Strimzi operator. By selecting real-life use cases with real-life workloads, we can identify the most meaningful metrics for evaluation and make informed decisions to improve the performance and scalability of Kafka. Using Grafana to visualize the data will help us identify any issues or bottlenecks in the performance of Kafka on Kubernetes and make recommendations for improvement.
To be able to run this project, you are obligatored to have installed:
- AWS cli
- Terrafom
- Kubectl
- First step is to configure your AWS credentials. Please use
aws configure
and then enter all requested details. - Next step is to add session token to
~/.aws/credentials
file. It should look like:
[default]
aws_access_key_id = <access_key>
aws_secret_access_key = <secret_access_key>
aws_session_token = <session_token>
- Next, go to terraform directory. Go to
variables.tf
and set required fields:
access_key, secret_key, session_token
- same value as in~/.aws/credentials
role_arn
- this is LabRole Arn, which can be found inIAM -> Roles -> LabRole -> copy its arn
separator
- Command line separator. If using windows use '&', if using macOS use ';'
Now, your environment is prepared to move to the next step.
To install all required packages etc., the best way is to use offical website for each of them.
- AWS cli - https://docs.aws.amazon.com/cli/latest/userguide/getting-started-install.html
- Terraform - https://developer.hashicorp.com/terraform/tutorials/aws-get-started/install-cli
- Kubectl - https://kubernetes.io/docs/tasks/tools/
Important! Remember to first go through the step 5 and step 6!
First you need to clone the repo.
git clone [email protected]:adrianryt/SUU_AGH.git
cd SUU_AGH
Next step is to run terraform commands.
cd resources/terraform
Update the cariables.tf file and run
terraform init
terraform plan
terraform apply
Then you need to wait for the cluster and pods to build.
After that check the connection and pods with kubectl.
kubectl get pods -n kafka
If everythink is running then the next step is to set the port forwarding for graphana
kubectl port-forward svc/grafana 3000:3000 -n kafka
The last step is to set up Prometheus
- Go to the localhost:3000
- Login with credentials user:
admin
pass:admin
- Open 'Source' tab and select 'Data sources' then choose Prometheus
- In the url write
http://prometheus-operated:9090/
- Import all jsons from resources directory
After those steps, we need to deploy Kafka producers and consumers. To do it, please type in command line:
kubectl apply --server-side -f https://raw.githubusercontent.com/adrianryt/SUU_AGH/main/resources/yaml/prod-cons.yaml -n kafka
That's all. Enjoy!
To fetch metrics from Prometheus, first thing we need to do is forward its port to localhost. To do so, please type:
kubectl port-forward svc/prometheus-operated 9090:9090 -n kafka
Then you can use scipt from ~/scripts/prometheus-metrics.py
.
The only thing which should be changed is query_name
. Possible values:
- kafka_consumergroup_current_offset
- kafka_topic_partitions
- kafka_controller_active_controller_count
- kafka_server_brokertopicmetrics_totalproducerequestspersec
- etc. ...
Consumer lag is simply the delta between the consumer's last committed offset and the producer's end offset in the log. In other words, the consumer lag measures the delay between producing and consuming messages in any producer-consumer system. To achieve lag, we will change fetch min bytes and fetch max wait ms records for consumer. This should be able to overload Kafka topic.
Simply:
When only one record is consumed with such settings, that is obvious that lag will grow really fast.
To resolve this issue, we can simply just set fetch_max_wait_ms
in prod-cons.yaml
to e.g. 500
and set fetch_min_bytes
to 1
.
After those changes, lag falls down until it reaches 0.
To change default topic's partitions and replicas value change the kafka-ephemeral.yaml as follow:
spec:
kafka:
config:
num.partitions: n
offsets.topic.replication.factor: m
To run performance tests on producer, first create a new topic or use existing, and then run command
kubectl run kafka-producer-performance -ti --image=strimzi/kafka:0.20.1-kafka-2.5.0 --rm=true --restart=Never -n kafka -- bin/kafka-producer-perf-test.sh --topic <topic_name> --num-records 1000000 --record-size 100 --throughput 100000 --producer-props bootstrap.servers=my-cluster-kafka-bootstrap:9092
To run performance tests on consumer, first create a new topic or use existing, and then run command
kubectl run kafka-consumer-performance --image=strimzi/kafka:0.20.1-kafka-2.5.0 --restart=Never -n kafka --attach --rm -- bin/kafka-consumer-perf-test.sh --bootstrap-server=my-cluster-kafka-bootstrap:9092 --messages=10000 --threads=5 --group=consumer-group --print-metrics --topic=my-topic
In conclusion, implementing Kafka on Kubernetes with Strimzi, Grafana, and Prometheus proved to be a highly effective solution for managing and monitoring data streaming pipelines. The combination of these technologies provided numerous benefits, such as scalability, resilience, and real-time visibility into the system's performance.
One key advantage of using Strimzi, an open-source Kubernetes Operator for Apache Kafka, was its seamless integration with Kubernetes. Strimzi simplified the deployment, configuration, and management of Kafka clusters on Kubernetes, automating tasks such as scaling, rolling upgrades, and topic management. This streamlined approach significantly reduced the operational overhead and allowed teams to focus on application development and data processing.
Grafana emerged as a powerful tool for visualizing Kafka metrics and monitoring the health of the Kafka clusters. With its intuitive dashboards and extensive plugin ecosystem, Grafana allowed us to gain valuable insights into the throughput, latency, and error rates of the Kafka infrastructure. This enabled us to proactively identify bottlenecks, optimize performance, and troubleshoot issues promptly.
The integration of Prometheus with Kafka, Strimzi, and Grafana further enhanced the observability capabilities of the system. Prometheus effectively collected and stored time-series data, enabling the creation of custom alerts and the analysis of historical trends. By leveraging Prometheus exporters provided by Strimzi, we were able to gather detailed Kafka-specific metrics, including consumer lag, partition health, and network utilization, thereby facilitating proactive monitoring and maintenance.
Overall, the combination of Kafka, Strimzi, Kubernetes, Grafana, and Prometheus provided a comprehensive solution for building and managing robust, scalable, and observable data streaming architectures. The integration of these technologies empowered teams to confidently develop, deploy, and monitor Kafka-based applications, ensuring efficient data processing and reliable messaging within complex distributed systems. The use of Strimzi simplified the management of Kafka clusters on Kubernetes, further enhancing the scalability and resilience of the overall solution.