This KIP is aimed at improving the error-handling semantics in Kafka Streams when Kafka Steams fails to serialize a message to the downstream sink by providing an interface that can provide custom massaging of the error (e.g. report to a custom metrics system) and indicate to Streams whether or not it should re-throw the Exception, thus causing the application to fall over.

Status

Current state: Under Discussion

Discussion thread: here [Change the link from the KIP proposal email archive to your own email thread]

JIRA: KAFKA-7499

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

Motivation

In KIP-210, an exception handler for the write path was introduced. This exception handler covers exception that are raised in the producer callback.

However, serialization happens before the data is handed to the producer with Kafka Streams itself and the producer uses `byte[]/byte[]` key-value-pair types.

Thus, we might want to extend the ProductionExceptionHandler to cover serialization exception, too, to skip over corrupted output messages. An example could be a "String" message that contains invalid JSON and should be serialized as JSON.

Public Interfaces

We are proposing addition of a new overloaded method in ProductionExceptionHandler interface, handle, that has the following signature:

        ProductionExceptionHandlerResponse handle(ProducerRecord record, boolean isKey, Exception exception)

Proposed Changes

This implementation will add the overloaded method, handle, in the following two classes:

• DefaultProductionExceptionHandler and returns response as FAIL
• AlwaysProductionExceptionHandler and returns response as CONTINUE

We'll implement the following error handling logic to the send in RecordCollectorImpl. The overloaded method, handle, in ProductionExceptionHandler will be invoked when

1. ClassCastException is thrown while serializing record key / value. Today, we are throwing StreamsException on hitting this exception. Whether to throw the
   exception to the user will be decided based on the response received from ProductionExceptionHandler handle method. 
   1. If the result is CONTINUE, log a note at DEBUG that we received that result and are not failing Streams as a result.
   2. If the result is FAIL, log a message at ERROR that we received that result and throw StreamsException so Streams will fail.
2. Any other unchecked exceptions, that thrown during record key / value serialization.
   1. If the result is CONTINUE, log a note at DEBUG that we received that result and are not failing Streams as a result.
   2. If the result is FAIL, log a message at ERROR that we received that result and set sendException so Streams will fail.

Earlier, we are invoking the error handler only when there are any exceptions in producer callback. Now, we also invoke the handler when hitting the serialization exception. As explained in KIP-210, this will facilitate a number of error handling scenarios.  

Compatibility, Deprecation, and Migration Plan

The default behavior will be consistent with the existing behavior. The new overloaded method, handle, will have a implementation that is set to FAIL by default.

Rejected Alternatives

We have considered to reuse the existing handle(ProducerRecord<byte[], byte[]> record, Exception exception) method in ProductionExceptionHandler, but it has following limitation:

1. The parameter ProducerRecord key and value type is set to byte[], on hitting the serialization exception the record key and value type may not be byte[].
2. To explicitly tell the user that the serialization exception occurred when serializing the record key or value.