Review: Kafka Exporter

Exporter reviews, Technology / 글쓴이 admin

In this edition of our exporter review series, we are introducing the Kafka exporter, one of the best-fit exporters for monitoring metrics used by NexClipper. In the article below you can learn all about Kafka, how to set up the exporter, recommended metrics and alert rules, as well as the related Grafana dashboard and Helm Chart.

About Kafka

Kafka is an open-source system developed by the Apache Software Foundation written in Java and Scala. It is a distributed event store and stream-processing platform. You can also call it a queue. It is a distributed system consisting of servers and clients that communicate via a high-performance TCP network protocol. It can be deployed on bare-metal hardware, virtual machines, and containers in on-premise as well as cloud environments.

Streaming data is continuously generated by thousands of data sources, which typically send the data records simultaneously. A streaming platform needs to handle this constant influx of data and process it sequentially and incrementally.

Kafka provides three main functions to its users:

Publishing and subscribing to streams of records
Effectively storing streams of records in the order in which records were generated
Processing streams of records in real-time

Kafka is primarily used to build real-time streaming data pipelines and applications that adapt to the data streams. It combines messaging, storage, and stream processing to allow storage and analysis of both historical and real-time data.

For this setup, we are using bitnami Kafka helm charts to start the Kafka server/cluster.

How do you set up the Kafka exporter for Prometheus?

With the latest version of Prometheus (2.33 as of February 2022), these are the ways to set up a Prometheus exporter:

Method 1 – Basic

Supported by Prometheus since the beginning
To set up an exporter in the native way a Prometheus config needs to be updated to add the target.
A sample configuration:

# scrape_config job
scrape_configs:

  - job_name: kafka
    scrape_interval: 45s
    scrape_timeout:  30s
    metrics_path: "/metrics"
    static_configs:
    - targets:
      - <Kafka exporter endpoint>

Method 2 – Service Discovery

This method is applicable for Kubernetes deployment only.
A default scrap config can be added to the prometheus.yaml file and an annotation can be added to the exporter service. With this, Prometheus will automatically start scrapping the data from the services with the mentioned path.

Prometheus.yaml

  - job_name: kubernetes-services   
        scrape_interval: 15s
        scrape_timeout: 10s
        kubernetes_sd_configs:
        - role: service
        relabel_configs:
        # Example relabel to scrape only endpoints that have
        # prometheus.io/scrape: "true" annotation.
        - source_labels: [__meta_kubernetes_service_annotation_prometheus_io_scrape]
          action: keep
          regex: true
        #  prometheus.io/path: "/scrape/path" annotation.
        - source_labels: [__meta_kubernetes_service_annotation_prometheus_io_path]
          action: replace
          target_label: __metrics_path__
          regex: (.+)
        #  prometheus.io/port: "80" annotation.
        - source_labels: [__address__, __meta_kubernetes_service_annotation_prometheus_io_port]
          action: replace
          target_label: __address__
          regex: (.+)(?::\d+);(\d+)
          replacement: $1:$2

Exporter service annotations:

 annotations:
    prometheus.io/path: /metrics
    prometheus.io/scrape: "true"

Method 3 – Prometheus Operator

Setting up a service monitor
The Prometheus operator supports an automated way of scraping data from the exporters by setting up a service monitor Kubernetes object. For reference, a sample service monitor for Kafka can be found here.
These are the necessary steps:

Step 1

Add/update Prometheus operator’s selectors. By default, the Prometheus operator comes with empty selectors which will select every service monitor available in the cluster for scrapping the data.

To check your Prometheus configuration:

Kubectl get prometheus -n <namespace> -o yaml

A sample output will look like this.

ruleNamespaceSelector: {}
    ruleSelector:
      matchLabels:
        app: kube-prometheus-stack
        release: kps
    scrapeInterval: 1m
    scrapeTimeout: 10s
    securityContext:
      fsGroup: 2000
      runAsGroup: 2000
      runAsNonRoot: true
      runAsUser: 1000
    serviceAccountName: kps-kube-prometheus-stack-prometheus
    serviceMonitorNamespaceSelector: {}
    serviceMonitorSelector:
      matchLabels:
        release: kps

Here you can see that this Prometheus configuration is selecting all the service monitors with the label release = kps

So with this, if you are modifying the default Prometheus operator configuration for service monitor scrapping, make sure you use the right labels in your service monitor as well.

Step 2

Add a service monitor and make sure it has a matching label and namespace for the Prometheus service monitor selectors (serviceMonitorNamespaceSelector & serviceMonitorSelector).

Sample configuration:

apiVersion: monitoring.coreos.com/v1
kind: ServiceMonitor
metadata:
  annotations:
    meta.helm.sh/release-name: kafka-exporter
    meta.helm.sh/release-namespace: monitor
  labels:
    app: prometheus-kafka-exporter
    app.kubernetes.io/managed-by: Helm
    chart: prometheus-kafka-exporter-1.1.0
    heritage: Helm
    release: kps
  name: kafka-exporter-prometheus-kafka-exporter
  namespace: monitor
  spec:
  endpoints:
  - interval: 15s
    port: kafka-exporter
  selector:
    matchLabels:
      app: prometheus-kafka-exporter
      release: kafka-exporter

As you can see, a matching label on the service monitor release = kps is used that is specified in the Prometheus operator scrapping configuration.

Metrics

Below are handpicked metrics for the Kafka exporter that will provide insights into Kafka operations.

Kafka topics replicas
This metric gives insight into the Kafka topic replicas in-sync.
➡ The key “kafka_topic_partition_in_sync_replica” will deliver the number of in-sync replicas for this topic/partition
➡ The value of this is the number of replicas

Kafka consumers group
This metric will get you the Kafka consumer lag, which indicates if the consumer is slow or down.
➡ The key is “kafka_consumergroup_lag” will provide insights into the lag per consumer
➡ The value is the number of messages that are not consumed yet

Kafka broker counts
As the name suggests, this delivers the count of brokers. If the count is less than the number of brokers in the cluster, it indicates that a broker is down
➡ The key “kafka_brokers” will give you the count of available brokers
➡ The value of this key is a number that shows the total connected brokers in the cluster

Kafka topic partitions
This metric concerns the visibility and provides the count of the partition of each topic.
➡ The key “kafka_topic_partitions” gives the partition count per topic

Alerting

After digging into all the valuable metrics, this section explains in detail how we can get critical alerts with the Kafka exporter.

PromQL is a query language for the Prometheus monitoring system. It is designed for building powerful yet simple queries for graphs, alerts, or derived time series (aka recording rules). PromQL is designed from scratch and has zero common grounds with other query languages used in time series databases, such as SQL in TimescaleDB, InfluxQL, or Flux. More details can be found here.

Prometheus comes with a built-in Alert Manager that is responsible for sending alerts (could be email, Slack, or any other supported channel) when any of the trigger conditions is met. Alerting rules allow users to define alerts based on Prometheus query expressions. They are defined based on the available metrics scraped by the exporter. Click here for a good source for community-defined alerts.

A general alert looks as follows:

– alert:(Alert Name)
expr: (Metric exported from exporter) >/</==/<=/=> (Value)
for: (wait for a certain duration between first encountering a new expression output vector element and counting an alert as firing for this element)
labels: (allows specifying a set of additional labels to be attached to the alert)
annotation: (specifies a set of informational labels that can be used to store longer additional information)

Some of the recommended Kafka alerts are:

Alert – Kafka topics replicas

- alert: KafkaTopicsReplicas
    expr: sum(kafka_topic_partition_in_sync_replica) by (topic) < 3
    for: 0m
    labels:
      severity: critical
    annotations:
      summary: Kafka topics replicas (instance {{ $labels.instance }})
      description: "Kafka topic in-sync partition\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"

Alert – Kafka consumers group

- alert: KafkaConsumersGroup
    expr: sum(kafka_consumergroup_lag) by (consumergroup) > 50
    for: 1m
    labels:
      severity: critical
    annotations:
      summary: Kafka consumers group (instance {{ $labels.instance }})
      description: "Kafka consumers group\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"

Alert – Kafka broker count

 - alert: KafkaBrokerDown
    expr: kafka_brokers < 3   
    for: 0m
    labels:
      severity: critical
    annotations:
      Summary: "Kafka broker *{{ $labels.instance }}* alert status"
      description: "One of the Kafka broker *{{ $labels.instance }}* is down."

Dashboard

Graphs are easier to understand and more user-friendly than a row of numbers. For this purpose, users can plot their time series data in visualized format using Grafana.

Grafana is an open-source dashboarding tool used for visualizing metrics with the help of customizable and illustrative charts and graphs. It connects very well with Prometheus and makes monitoring easy and informative. Dashboards in Grafana are made up of panels, with each panel running a PromQL query to fetch metrics from Prometheus.
Grafana supports community-driven graphs for most of the widely used software, which can be directly imported to the Grafana Community.

NexClipper uses the Kafka by the jack chen dashboard, which is widely accepted and has a lot of useful panels.

What is a Panel?

Panels are the most basic component of a dashboard and can display information in various ways, such as gauge, text, bar chart, graph, and so on. They provide information in a very interactive way. Users can view every panel separately and check the value of metrics within a specific time range.
The values on the panel are queried using PromQL, which is Prometheus Query Language. PromQL is a simple query language used to query metrics within Prometheus. It enables users to query data, aggregate and apply arithmetic functions to the metrics, and then further visualize them on panels.

Here are some examples of panels:

Helm Chart

The exporter, alert rule, and dashboard can be deployed in Kubernetes using the Helm chart. The Helm chart used for deployment is taken from the Prometheus community, which can be found here.

Installing Kafka Cluster

If your Kafka cluster is not up and ready you can start it using Helm:

$ helm repo add bitnami https://charts.bitnami.com/bitnami
$ helm install my-release bitnami/kafka

Note that bitnami charts allow you to deploy a Kafka exporter as part of the Helm chart. You can enable it by adding “–set metrics.kafka.enabled=true”

Installing Kafka Exporter

helm repo add Prometheus-community https://prometheus-community.github.io/helm-charts

helm repo update
helm install my-release prometheus-community/prometheus-kafka-exporter

Some of the common parameters that must be changed in the values file include:

kafkaServer: "IP/Hostname:9092"

All these parameters can be tuned via the values.yaml file here.

Scrape the metrics

There are multiple ways to scrape the metrics as discussed above. In addition to the native way of setting up Prometheus monitoring, a service monitor can be deployed (if a Prometheus operator is being used) to scrap the data from the Kafka exporter. With this approach, multiple Kafka servers can be scrapped without altering the Prometheus configuration. Every Kafka exporter comes with its own service monitor.
In the above-mentioned chart, a service monitor can be deployed by turning it on from the values.yaml file here.

prometheus:
  serviceMonitor:
    enabled: true
    namespace: monitoring
    interval: "30s"
    # If serviceMonitor is enabled and you want prometheus to automatically register
    # target using serviceMonitor, add additionalLabels with prometheus release name
    # e.g. If you have deployed kube-prometheus-stack with release name kube-prometheus
    # then additionalLabels will be
    # additionalLabels:
    #   release: kube-prometheus
    additionalLabels: {}
    targetLabels: []

Update the annotation section here if you are not using the Prometheus Operator.

service: 
  annotations:
    prometheus.io/path: /metrics
    prometheus.io/scrape: "true"

And with this, we conclude our discussion of the Kafka exporter. If you have any questions about the content of this article or our other exporter reviews, you can reach our team via support@nexclipper.io. Stay tuned for more useful exporter reviews in the near future!