Status

Current state: Draft

Discussion thread: here

JIRA: TBD

Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).

Motivation

Kafka exposes many pluggable API for users to bring their custom plugins. For complex and critical plugins it's important to have metrics to monitor their behavior. Plugins can use the Metrics class from the Kafka API but when creating a new Metrics instance it does not inherits the tags from the component it depends on (for example from a producer for a custom partitioner), or the registered metrics reporters. As most plugins are configurable, a workaround is to reimplement the metric reporters logic and in some case for tags too but that is cumbersome.

This issue also applies to connectors and tasks in Kafka Connect. For example MirrorMaker2 creates its own Metrics object and has logic to add the metric reporters from the configuration.

Public Interfaces

Plugins

For client and server side plugins, I propose introducing a new interface Monitorable. If a plugin implements this interface, withMetrics() will be called when the plugin is instantiated (after configure() if the plugin also implements Configurable). This will allow the plugin to adds its own metrics to the existing Metrics instance from the component (producer, consumer, etc) that initiated it.

package org.apache.kafka.common.metrics;

public interface Monitorable {

    /**
     * Get the Metrics instance from the client that instantiates the plugin.
     */
    void withMetrics(Metrics Metrics);

}

Connectors/Tasks

For connectors and tasks, I propose adding a metrics() method to the SinkConnectorContext, SourceConnectorContext, SinkTaskContext and SourceTaskContext interfaces.

/**
 * Retrieve the Metrics instance from the worker running this connector/task. Returns null if the runtime does not support this feature.
 */
default Metrics metrics() {
    return null;
}

Proposed Changes

When instantiating a class via the Utils.newInstance() helper methods, if it implements Monitorable and a Metrics object is available, withMetrics() will be called with the current Metrics instance. It will be always called after configure().  Metrics registered by plugins will inherit the prefix/namespace from the current Metrics instance, these are: kafka.producer, kafka.consumer, kafka.connect, kafka.streams and kafka.server. Metrics reporters will be able to implement the Monitorable interface as they are created before the Metrics instance.

MirrorSourceConnector and MirrorCheckpointConnector currently register metrics using the kafka.connect.mirror prefix/namespace. If we update them with this proposal, their metrics will be renamed, for example from kafka.connect.mirror:type=MirrorSourceConnector to kafka.connect:type=MirrorSourceConnector. This KIP will not update them.

Compatibility, Deprecation, and Migration Plan

This is a new feature so it has no impact on deprecation and does not need a migration plan. Regarding compatibility, plugins and connectors that start using this feature will have to handle Metrics not being available to support older broker/client/Connect versions. For regular plugins they should be able to function without a call to withMetrics(), for connectors and tasks they should handle the metrics() method returning null when deployed on an older runtime.

Test Plan

This feature will be tested using unit and integration tests.

Rejected Alternatives

• Create dedicated a Metrics instance for plugins: A dedicated instance could have its own prefix/namespace (for example kafka.consumer.plugins). This would allow grouping metrics from all plugins but it requires instantiating another Metrics instance and new metrics reporters.
  
• Create a simpler PluginMetrics API: Instead of passing Metrics, a simpler API could be easier for plugins. It would also prevent plugins calling close() or removeMetrics() but since plugins execute code, administrators should only use plugins they trust.