You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 5 Next »


Status

Current state"Under Discussion"

Discussion thread: here

Github PR: here

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

Motivation

I've proposed when Kafka Stream's Scala API was not released yet to solve the serdes issue on functions that takes a materialized (count, reduce and aggregate) to just make the materialized implicit and have a default implicit in ImplicitConversions that had the serdes filled in:

Given the following definitions:
def count()(implicit materialized: Materialized[K, Long, ByteArrayKeyValueStore])
def reduce(reducer: (V, V) => V)(implicit materialized: Materialized[K, V, ByteArrayKeyValueStore])
def aggregate[VR](initializer: => VR)(aggregator: (K, V, VR) => VR)(implicit materialized: Materialized[K, VR, ByteArrayKeyValueStore])

We can call them without Materialized:
groupedStream.count()

Or with a materialized:
groupedStream.count()(Materialized.as("store-name"))

By doing that we solved the case when no Materialized is given, setting implicitly the serdes to avoid runtime errors but still allowing the ability to give explicitly our own Materialized.


The issue with this solution are the followings:

  • Making Materialized implicit suggests that a Materialized is a context object whereas it's not.
  • The user of the API will by default just not care about implicit parameters and assume that they are provided.
  • The user of the API might declare it's own Materialized in an implicit val since the parameter is implicit and then be caught by other unwanted functions that also takes Materialized as implicit.
  • This a long shot view but in Scala 3 giving implicits explicitly looks like:
    groupedStream.count().explicitly(Materialized.as("store-name"))
    So it's clear tha

All of that applies to Consumed and Produced too.

Public Interfaces

org.apache.kafka.streams.scala.kstream.KGroupedStream.count()
org.apache.kafka.streams.scala.kstream.KGroupedStream.reduce()
org.apache.kafka.streams.scala.kstream.KGroupedStream.aggregate()
org.apache.kafka.streams.scala.kstream.KGroupedTable.count()
org.apache.kafka.streams.scala.kstream.KGroupedTable.reduce()
org.apache.kafka.streams.scala.kstream.KGroupedTable.aggregate()
org.apache.kafka.streams.scala.ImplicitConversions.materializedFromSerde()
org.apache.kafka.streams.scala.StreamsBuilder.stream()
org.apache.kafka.streams.scala.StreamsBuilder.table()
org.apache.kafka.streams.scala.StreamsBuilder.globalTable()

Proposed Changes

def count()(implicit keySerdeSerde[K], valueSerdeSerde[V])
def count(implicit materializedMaterialized[KLongByteArrayKeyValueStore])

def reduce(reducer: (VV=> V)(implicit keySerdeSerde[K], valueSerdeSerde[V])
def reduce(reducer: (VV=> V, materializedMaterialized[KVByteArrayKeyValueStore])


def aggregate[VR](initializer=> VR)(aggregator: (KVVR=> VR)(implicit keySerdeSerde[K], valueSerdeSerde[V])
def aggregate[VR](materializedMaterialized[KVRByteArrayKeyValueStore], initializer=> VR)(aggregator: (KVVR=> VR)

def stream[K, V](topic: String)(implicit keySerdeSerde[K], valueSerdeSerde[V])
def stream[K, V](topic: Stringconsumed: Consumed[K, V])

def table[K, V](topic: String)(implicit keySerdeSerde[K], valueSerdeSerde[V])
def table[K, V](topic: Stringconsumed: Consumed[K, V])

def globalTable[K, V](topic: String)(implicit keySerdeSerde[K], valueSerdeSerde[V])
def globalTable[K, V](topic: Stringconsumed: Consumed[K, V])

def through(topicString)(implicit keySerdeSerde[K], valueSerdeSerde[V])
def through(topicStringproducedProduced[KV])

def to(topicString)(implicit keySerdeSerde[K], valueSerdeSerde[V])
def to(topicStringproducedProduced[KV])

def to(extractorTopicNameExtractor[KV])(implicit keySerdeSerde[K], valueSerdeSerde[V])
def to(extractorTopicNameExtractor[KV], producedProduced[KV])

This way we require only the Serdes if the Materialized (or Consumed, or Produced) is not given explicitly.

Compatibility, Deprecation, and Migration Plan

TBD

Rejected Alternatives

NA

  • No labels