A very simple role to perform a rolling restart of a Kafka cluster that waits for the cluster to recover before restarting the next cluster node.
No special requirements for the Ansible node. Nodes that this role is run against are expected to have the kafka-topics tool installed an run a kafka service.
| Variable Name | Description | Default Value |
|---|---|---|
| ZOOKEEPER_CONNECT | Not currently in use | kafka-topics |
| NUMBER_OF_RETRIES | The number of times to check whether all under-replicated partitions have been fixed. | 10 |
| DELAY_BETWEEN_RETRIES | Seconds to wait between retries. | 15 |
| KAFKA_SERVICE_NAME | Name of the Kafka service to be restarted. | confluent-kafka |
| KAFKA_TOPICS_COMMAND | Name (and path if necessary) of the kafka-topics tool - this is mostly to distinguish between Apache and Confluent Kafka. | kafka-topics |
None.
For this role to work properly, it is very important to include serial: 1 in your playbook. Otherwise it will not be a rolling restart, as multiple hosts will be restarted at the same time, potentially leading to downtime.
- hosts: kafka
serial: 1
roles:
- kafka-rolling-restart
Apache 2.0