Current state: Voting
Discussion thread: here
JIRA: here
Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).
When using the Kafka command-line tools to do simple performance tests, it quickly becomes evident that kafka-producer-perf-test.sh and the equivalent consumer tools do not follow the same conventions for command-line arguments. Strangely, kafka-producer-perf-test.sh does not support the --bootstrap-server argument which all of the other tools do, instead requiring --producer.props bootstrap.servers= (note the "s" on "bootstrap.servers" because it's the name of the producer config). This means that simple familiarity with the other tools isn't sufficient to get the command line for this tool correct without a little trial and error.
There are a few other inconsistencies where different names are used for the same thing in different tools. The differences just add a little friction when working with the tools.
This KIP makes changes to the arguments of some of the command-line tools. The proposed changes are:
--bootstrap-server to kafka-producer-perf-test.sh bringing it into line with all the other tools. Because users will already have config files which specify the bootstrap server and there's no need to break these, the new --bootstrap-server flag is optional. If provided, it takes precedence over specifying bootstrap.servers in the configuration file. The order of precedence for configuring the bootstrap server from highest to lowest is --bootstrap-server , then --producer-property , then --command-config .--command-property as a way of providing individual configuration properties without use of a configuration file. Replace --producer-props with --command-property in kafka-producer-perf-test.sh . Add --command-property to kafka-consumer-perf-test.sh and kafka-share-consumer-perf-test.sh .--command-config . This eliminates the guesswork about whether a particular tool uses the producer, consumer or admin APIs and then chose to pass this knowledge into the original names of the command-line arguments.--reporting-interval to kafka-producer-perf-test.sh . The tools already has a reporting interval hard-coded to 5000ms, so this simply makes it configurable like for the consumer performance tests, with a default of 5000ms.--property with --formatter-property in kafka-console-consumer.sh and kafka-console-share-consumer.sh . This argument is used to configure the formatter, so this names it accordingly.--messages with --num-records in kafka-consumer-perf-test.sh and kafka-share-consumer-perf-test.sh .The following table shows the existing and proposed options. The changes are in GREEN.
Tools | Existing option for bootstrap server | Proposed option for bootstrap server | Existing option for properties | Proposed option for properties | Existing option for config file | Proposed option for config file |
kafka-acls.sh | --bootstrap-server | --command-config | ||||
kafka-broker-api-versions.sh | --bootstrap-server | --command-config | ||||
kafka-client-metrics.sh | --bootstrap-server | --command-config | ||||
kafka-cluster.sh | --bootstrap-server | --config | --command-config | |||
kafka-configs.sh | --bootstrap-server | --command-config | ||||
kafka-console-consumer.sh | --bootstrap-server | --consumer-property | --consumer.config | --command-config | ||
kafka-console-producer.sh | --bootstrap-server | --producer-property | --producer.config | --command-config | ||
kafka-console-share-consumer.sh | --bootstrap-server | --consumer-property | --consumer-config | --command-config | ||
kafka-consumer-groups.sh | --bootstrap-server | --command-config | ||||
kafka-consumer-perf-test.sh | --bootstrap-server | --command-property | --consumer.config | --command-config | ||
kafka-delegation-tokens.sh | --bootstrap-server | --command-config | ||||
kafka-delete-records.sh | --bootstrap-server | --command-config | ||||
kafka-features.sh | --bootstrap-server | --command-config | ||||
kafka-get-offsets.sh | --bootstrap-server | --command-config | ||||
kafka-groups.sh | --bootstrap-server | --command-config | ||||
kafka-leader-election-sh | --bootstrap-server | --admin.config | --command-config | |||
kafka-log-dirs.sh | --bootstrap-server | --command-config | ||||
kafka-metadata-quorum.sh | --bootstrap-server | --command-config | ||||
kafka-producer-perf-test.sh | --bootstrap-server | --producer-props | --command-property | --producer.config | --command-config | |
kafka-reassign-partitions.sh | --bootstrap-server | --command-config | ||||
kafka-share-consumer-perf-test.sh | --bootstrap-server | --command-property | --consumer.config | --command-config | ||
kafka-share-groups.sh | --bootstrap-server | --command-config | ||||
kafka-streams-application-reset.sh | --bootstrap-server | --config-file | --command-config | |||
kafka-streams-groups.sh | --bootstrap-server | --command-config | ||||
kafka-topics.sh | --bootstrap-server | --command-config | ||||
kafka-transactions.sh | --bootstrap-server | --command-config | ||||
kafka-verifiable-consumer.sh | --bootstrap-server | --consumer.config | --command-config | |||
kafka-verifiable-producer.sh | --bootstrap-server | --producer.config | --command-config | |||
kafka-verifiable-share-consumer.sh | --bootstrap-server | --command-config |
The existing options will be deprecated until the next major Apache Kafka release, and then they will be removed. During the deprecation period, using the old parameter names will cause a deprecation message to be displayed. It is an error to use both the old name and the new name together - they are mutually exclusive.
All uses of the command-line tools in the Apache Kafka codebase such as system tests will be migrated over to the new parameter names.
The KIP will be tested with unit tests.
For tests such as system tests which use these command-line tools, the tests will be migrated across to the new parameter names providing additional validation.
An alternative for tools such as kafka-verifiable-producer.sh would be rename --producer.config to --producer-config instead of --command-config . This was rejected because some of the tools such as kafka-verifiable-share-consumer.sh actually use multiple APIs (consumer and admin in this case) so the clean simplicity of --command-config was chosen instead for all tools.