Xinfra Monitor (formerly Kafka Monitor) is a framework to implement and execute long-running kafka system tests in a real cluster. It complements Kafka’s existing system tests by capturing potential bugs or regressions that are only likely to occur after prolonged period of time or with low probability. Moreover, it allows you to monitor Kafka cluster using end-to-end pipelines to obtain a number of derived vital stats such as

1. End-to-end latency
2. Service availability
3. Produce and Consume availability
4. Consumer offset commit availability
5. Consumer offset commit latency
6. Kafka message loss rate
7. And many, many more.

You can easily deploy Xinfra Monitor to test and monitor your Kafka cluster without requiring any change to your application.

Xinfra Monitor can automatically create the monitor topic with the specified config and increase partition count of the monitor topic to ensure partition# >= broker#. It can also reassign partition and trigger preferred leader election to ensure that each broker acts as leader of at least one partition of the monitor topic. This allows Xinfra Monitor to detect performance issue on every broker without requiring users to manually manage the partition assignment of the monitor topic.

Xinfra Monitor is used in conjunction with different middle-layer services such as li-apache-kafka-clients in order to monitor single clusters, pipeline desination clusters, and other types of clusters as done in Linkedin engineering for real-time cluster healthchecks.

These are some of the metrics emitted from a Xinfra Monitor instance.

kmf:type=kafka-monitor:offline-runnable-count
kmf.services:type=produce-service,name=*:produce-availability-avg
kmf.services:type=consume-service,name=*:consume-availability-avg
kmf.services:type=produce-service,name=*:records-produced-total
kmf.services:type=consume-service,name=*:records-consumed-total
kmf.services:type=produce-service,name=*:records-produced-rate
kmf.services:type=produce-service,name=*:produce-error-rate
kmf.services:type=consume-service,name=*:consume-error-rate
kmf.services:type=consume-service,name=*:records-lost-total
kmf.services:type=consume-service,name=*:records-lost-rate
kmf.services:type=consume-service,name=*:records-duplicated-total
kmf.services:type=consume-service,name=*:records-delay-ms-avg
kmf.services:type=commit-availability-service,name=*:offsets-committed-avg
kmf.services:type=commit-availability-service,name=*:offsets-committed-total
kmf.services:type=commit-availability-service,name=*:failed-commit-offsets-avg
kmf.services:type=commit-availability-service,name=*:failed-commit-offsets-total
kmf.services:type=commit-latency-service,name=*:commit-offset-latency-ms-avg
kmf.services:type=commit-latency-service,name=*:commit-offset-latency-ms-max
kmf.services:type=commit-latency-service,name=*:commit-offset-latency-ms-99th
kmf.services:type=commit-latency-service,name=*:commit-offset-latency-ms-999th
kmf.services:type=commit-latency-service,name=*:commit-offset-latency-ms-9999th

Xinfra Monitor requires Gradle 2.0 or higher. Java 7 should be used for building in order to support both Java 7 and Java 8 at runtime.

Xinfra Monitor supports Apache Kafka 0.8 to 2.0:

• Use branch 0.8.2.2 to work with Apache Kafka 0.8
• Use branch 0.9.0.1 to work with Apache Kafka 0.9
• Use branch 0.10.2.1 to work with Apache Kafka 0.10
• Use branch 0.11.x to work with Apache Kafka 0.11
• Use branch 1.0.x to work with Apache Kafka 1.0
• Use branch 1.1.x to work with Apache Kafka 1.1
• Use master branch to work with Apache Kafka 2.0

1. We advise advanced users to run Xinfra Monitor with ./bin/xinfra-monitor-start.sh config/xinfra-monitor.properties. The default xinfra-monitor.properties in the repo provides an simple example of how to monitor a single cluster. You probably need to change the value of zookeeper.connect and bootstrap.servers to point to your cluster.


2. The full list of configs and their documentation can be found in the code of Config class for respective service, e.g. ProduceServiceConfig.java and ConsumeServiceConfig.java.


3. You can specify multiple SingleClusterMonitor in the xinfra-monitor.properties to monitor multiple Kafka clusters in one Xinfra Monitor process. As another advanced use-case, you can point ProduceService and ConsumeService to two different Kafka clusters that are connected by MirrorMaker to monitor their end-to-end latency.


4. Xinfra Monitor by default will automatically create the monitor topic based on the e.g. topic-management.replicationFactor and topic-management.partitionsToBrokersRatio specified in the config. replicationFactor is 1 by default and you probably want to change it to the same replication factor as used for your existing topics. You can disable auto topic creation by setting produce.topic.topicCreationEnabled to false.


5. Xinfra Monitor can automatically increase partition count of the monitor topic to ensure partition# >= broker#. It can also reassign partition and trigger preferred leader election to ensure that each broker acts as leader of at least one partition of the monitor topic. To use this feature, use either EndToEndTest or TopicManagementService in the properties file.


6. When using Secure Sockets Layer (SSL) or any non-plaintext security protocol for AdminClient, please configure the following entries in the single-cluster-monitor props, produce.producer.props, as well as consume.consumer.props. https://docs.confluent.io/current/installation/configuration/admin-configs.html
   1. ssl.key.password
   2. ssl.keystore.location
   3. ssl.keystore.password
   4. ssl.truststore.location
   5. ssl.truststore.password

$ git clone https://github.com/linkedin/kafka-monitor.git
$ cd kafka-monitor 
$ ./gradlew jar

$ ./bin/xinfra-monitor-start.sh config/xinfra-monitor.properties

Edit config/xinfra-monitor.properties to specify custom configurations for producer in the key/value map produce.producer.props in config/xinfra-monitor.properties. Similarly specify configurations for consumer as well. The documentation for producer and consumer in the key/value maps can be found in the Apache Kafka wiki.

$ ./bin/xinfra-monitor-start.sh config/xinfra-monitor.properties

Metrics produce-availability-avg and consume-availability-avg demonstrate whether messages can be properly produced to and consumed from this cluster. See Service Overview wiki for how these metrics are derived.

$ ./bin/single-cluster-monitor.sh --topic test --broker-list localhost:9092 --zookeeper localhost:2181

Edit config/multi-cluster-monitor.properties to specify the right broker and zookeeper url as suggested by the comment in the properties file

Metrics produce-availability-avg and consume-availability-avg demonstrate whether messages can be properly produced to the source cluster and consumed from the destination cluster. See config/multi-cluster-monitor.properties for the full jmx path for these metrics.

$ ./bin/xinfra-monitor-start.sh config/multi-cluster-monitor.properties

./gradlew checkstyleMain checkstyleTest

./gradlew idea
./gradlew eclipse

• Motivation
• Design Overview
• Service and App Overview
• Future Work
• Application Configuration