Status

Current state: "Accepted"

Discussion thread: here

Vote thread: here

JIRA:

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

Motivation

The Kafka Connect framework allows sink connector tasks to do their own offset tracking in case they want to do asynchronous processing (for instance, buffering records sent by the framework to be flushed to the sink system at some later time). The SinkTask::preCommit method allows sink task implementations to provide the framework with the consumer offsets that are safe to commit for each topic partition. There's currently an incompatibility between Sink connectors overriding the SinkTask::preCommit method and SMTs that mutate the topic field of a SinkRecord (for instance, the RegexRouter SMT that ships with Apache Kafka).

This problem has been present since the SinkTask::preCommit method's inception and is rooted in a mismatch between the Kafka topic partition and offset that is passed to SinkTask::open / SinkTask::preCommit (the original topic partition and offset before any transformations are applied) and the topic/partition/offset that is present in the SinkRecord that the SinkTask::put method receives (after transformations are applied). Since that's all the information the connector has to implement any kind of internal offset tracking, the topic/partition/offset it can return in preCommit will correspond to the transformed topic/partition/offset, when the framework actually expects it to be the original topic/partition/offset.

In the problem was fixed for connectors that don't override  SinkTask::preCommit. For the others, it was acknowledged that "broader API changes are required". 

Public Interfaces

org.apache.kafka.connect.sink.SinkRecord

Add new fields corresponding to a record's pre-transform ("original") topic / partition / offset along with their corresponding getters and constructor:

    private final String originalTopic;
    private final Integer originalKafkaPartition;
    private final long originalKafkaOffset;
...
    public SinkRecord(String topic, int partition, Schema keySchema, Object key, Schema valueSchema, Object value, long kafkaOffset,
                      Long timestamp, TimestampType timestampType, Iterable<Header> headers, String originalTopic, Integer originalKafkaPartition, long originalKafkaOffset)
...     


    /**
     * @return the topic corresponding to the Kafka record before any transformations were applied. This should be
     * used for any internal offset tracking purposes rather than {@link #topic()}, in order to be compatible 
     * with SMTs that mutate the topic name.
     */
    public String originalTopic() {
        return originalTopic;
    }

    /**
     * @return the topic partition corresponding to the Kafka record before any transformations were applied. This
     * should be used for any internal offset tracking purposes rather than {@link #kafkaPartition()}, in order to be 
     * compatible with SMTs that mutate the topic partition.
     */
    public Integer originalKafkaPartition() {
        return originalKafkaPartition;
    }

    /**
     * @return the offset corresponding to the Kafka record before any transformations were applied. This
     * should be used for any internal offset tracking purposes rather than {@link #kafkaOffset()}, in order to be 
     * compatible with SMTs that mutate the offset value.
     */
    public long originalKafkaOffset() {
        return originalKafkaOffset;
    }

Proposed Changes

Expose the original (pre-transformation) Kafka topic, topic partition and offset via new SinkRecord public methods and ask sink task implementations to use that information for offset tracking purposes (i.e. the topic partition offsets that the sink task returns to the framework via SinkTask::preCommit). Note that while the record's offset can't be modified via the standard SinkRecord::newRecord methods that SMTs are expected to use, SinkRecord has public constructors that would allow SMTs to return records with modified offsets. This is why the proposed changes include a new SinkRecord::originalKafkaOffset method as well.

Compatibility, Deprecation, and Migration Plan

Old / existing sink connector plugins running on a new Kafka Connect version

Old / existing sink connector plugins will continue running as expected on new versions of Kafka Connect but they will continue to remain incompatible with topic-mutating SMTs.

New / updated sink connector plugins running on an older Kafka Connect version

To ensure that connectors using the new SinkRecord getter methods can still be deployed on older versions of Kafka Connect, a try catch block should be used to catch and handle NoSuchMethodError / NoClassDefFoundError (also see KIP-610: Error Reporting in Sink Connectors for reference which had the same issue). However, it should be clearly documented that the connector won't be compatible with topic/partition/offset mutating SMTs when deployed on older versions of Kafka Connect.

Rejected Alternatives

1. Address the offsets problem entirely within the framework, doing some kind of mapping from the transformed topic back to the original topic.

   • This would only work in the cases where there’s no overlap between the transformed topic names, but would break for the rest of the transformations (e.g. static transformation, topic = “a”).

   • Even if we wanted to limit the support to those cases, it would require considerable bookkeeping to add a validation to verify that the transformation chain adheres to that expectation (and fail fast if it doesn’t).

2. Expose the entire original record instead of only topic / partition / offset (e.g. originalSinkRecord)

   • We should not expose the original value / key, transformations might be editing them for security reasons.

3. Create a method in the SinkTaskContext to get the original topic / partition / offset information, instead of updating SinkRecord (e.g. SinkTaskContext.getOriginalTopic(SinkRecord sr) / SinkTaskContext.getOriginalKafkaPartition(SinkRecord sr)

   • Requires extra bookkeeping without concrete value.

4. Update SinkTask::put in any way to pass the new information outside SinkRecord (e.g. a Map or a derived class)

   • Much more disruptive change without considerable pros

5. Use record headers to expose the original topic / partition / offset instead of adding new methods to SinkRecord
   • While this gets around the need to handle NoSuchMethodError / NoClassDefFoundError in sink connectors that want to retain compatibility with older Connect runtimes, it is arguably even less intuitive from a connector developer's standpoint.
6. Add new overloaded SinkTask::open and SinkTask::close methods which include both pre-transform and post-transform topic partitions so that connectors can create and destroy resources even if they choose to use post-transform topic partitions to write to the sink system
   • This requires sink connector plugins to implement two new additional methods without significant benefit. The information from the existing open / close methods on pre-transform topic partitions can be combined with the per record information of both pre-transform and post-transform topic partitions to handle most practical use cases without significantly muddying the sink connector related public interfaces.
7. Keep the existing SinkTask::open and SinkTask::close methods but call them with post-transform rather than pre-transform topic partitions
   • This would break backward compatibility since there could exist connector plugins that are aware of the current limitations (i.e. that SinkTask::open / SinkTask::close are called for pre-transform topic partitions but records in SinkTask::put only currently contain their post-transform topic partition) and work around them. Also, pre-transform / original topic partitions and offsets are critical for any sink connectors that wish to implement exactly-once semantics by tracking offsets in the sink system rather than Kafka.