Current state: Accepted
Discussion thread: here
Currently there is a way to pass a
KafkaStreams constructor. This KIP proposes to add a public config in
StreamsConfig to pass the class through config and use reflection to create
KafkaClientSupplier object. The major motivation is to allow existing KafkaStreams applications to upgrade without application code changes. Note that if existing KafkaStreams applications want to
use a customized
KafkaClientSupplier , they may still need to add required dependencies.
We will add a public field in
default.client.supplier to configure the
KafkaClientSupplier we use in KafkaStreams. The default value of the config will the classname
We will also add a new public method in
getKafkaClientSupplier . It will return
KafkaClientSupplier based on the config.
KafkaStreams constructor, we will read the value of
StreamsConfig . We will use reflection to create an object of given class and throw
an exception if the class type isn't
KafkaClientSupplier . If user provides both the config and supply
KafkaStreams constructor, the config will be ignored and supplied
KafkaClientSupplier will be used.
Compatibility, Deprecation, and Migration Plan
There will be no compatibility, deprecation and migration issues.
Unit test can capture the changes in this KIP. We can test happy path where object can be created successfully and exception case where class couldn't be found or type is wrong.
An alternative is that we don't need this change and user can use
KafkaStreams constructor to supply any
KafkaClientSupplier since it's a public constructor.
The benefits of this alternative are:
- Don't need this KIP and code changes in this KIP
- Users don't need to upgrade AK version to use new
- Users using same config for different
KafkaStreamscan have a different
The downsides of this alternative are:
- Users need to change every
KafkaStreamsconstructor call in order to use a new
So to cater for users who don't want to update their source code, we are proposing to add this config option as well.