Status

Current state: "DISCUSS"

Discussion thread: 

JIRA: KAFKA-12313

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

Motivation

• KIP-659 introduced a config windowSize to TimeWindowedDeserializer which discourages the setting of windowSize to Long.MAX_VALUE as the default value. However, there is still a possibility of setting the config only for innerClassSerde or only for windowSize. This KIP aims at standardising the way the deserializer object can be created for the usage in console consumer and via the DSL. 
• The KIP also introduces changes around the default.windowed.serde.inner.class configs. These are:
  • Rename the default.windowed.key.serde.inner to windowed.deserializer.inner.class config. This way, the config comes closer to the window.size.ms config introduced in KIP-659 and it also emphasises that the config isn't really a default one. Also, as per the proposed changes below, this config is to be used really from the console consumer so we can use Deserialiser class here directly instead of a Serde.
  • Deprecate the default.windowed.value.serde.inner config. This is because there's really no concept of a "WindowedValue" in Kafka Streams, unlike the "WindowedKey" which appears in the DSL. The Windowed class in Streams itself assumes that the inner class is a key. If a user really does want a WindowedValue in some custom processor for some reason, they would have to write the WindowedValue class themselves. In that case I think it's reasonable to assert that they should be responsible for writing the WindowedValueSerializer class as well, and pass this to the console consumer should they need to use it.

Proposed Changes

Here are the changes being proposed in the KIP:

• Rename the default.windowed.key.serde.inner to windowed.deserializer.inner.class in StreamConfig.
• Deprecate default.windowed.value.serde.inner in StreamConfig.
• If the users want to run Console consumer for testing purposes, then it will invoke the default constructor of TimeWindowedDeserializer. We will ensure that both the configs (windowSize and deserialiser class) are set and if they aren't then an error would be thrown. 
• For any other plain consumer client, the user can pass them in as configs OR pass the parameters to the TimeWindowedDeserializer constructor, and then pass that object to the consumer. It is ok for the user to pass both as long as there is no conflict between the params passed via constructor and via the configs.
• For use in Kafka Streams (such as the DSL), the user must supply the parameters by constructing a TimeWindowedSerde and passing that in as a parameter to any relevant DSL operators. This is already supported so no change is needed here.
• Similar to the checks added in KIP-659 to validate the windownSize config, a check would be added for deserialiserInnerClass. This would ensure that it has only 1 valid value.

Public Interfaces

• StreamsConfig:
  • Rename default.windowed.key.serde.inner to windowed.deserializer.inner.class.
  • Deprecate default.windowed.value.serde.inner.

        public static final String WINDOWED_DESERIALISER_INNER_CLASS = "windowed.deserializer.inner.class";
        private static final String WINDOWED_DESERIALISER_INNER_CLASS_DOC = " Deserializer for the inner class of a windowed key. Must implement the " +
            "<code>org.apache.kafka.common.serialization.Deserializer</code> interface.";
    
    	@Deprecated
    	public static final String DEFAULT_WINDOWED_VALUE_SERDE_INNER_CLASS = "default.windowed.value.serde.inner";

• ConsoleConsumer
  • It would be mandatory to pass windowed.deserializer.inner.class and window.size.ms config. <Need to check how to do this>
• TimeWindowedDeserializer
  • Inside configure method, ensure that innerClassDeserialiser passed via constructor and the config are not conflicting. What it means is that:
    • If both inner and windowedDeserialiserInnerClass are set, then they should be the pointing to the same class.
    • Atleast one of inner and windowedDeserialiserInnerClass are set.

    •     if (innerClassDeserializer != null && configInnerClassDeserializer != null) { 
      		if (innerClassDeserializer != configInnerClassDeserializer)
      			throw new IllegalArgumentException("Inner class deserializer passed via constructor and config windowed.deserializer.inner.class should match"); 
      	} else if (innerClassDeserializer == null && configInnerClassDeserializer == null) {
      		throw new IllegalArgumentException("Inner class deserializer should be passed either via constructor or via "); 
      	}

Compatibility, Deprecation, and Migration Plan

Since default.windowed.key.serde.inner config is being rename to windowed.deserializer.inner.class in StreamConfig and also default.windowed.value.serde.inner is being deprecated, we will have to follow a deprecation path for the same. We can maybe enforce the removal of the deprecated configs and then enforce users?

Rejected Alternatives

N/A