The main difference is the layer where diskless is introduced. KIP-1150 makes it a diskless topic path with a Batch Coordinator; AutoMQ tries to keep the existing Kafka broker/leader path and swap out the storage underneath.
That makes the designs feel quite different even though both end up using object/shared storage.
Comments
Nice to see a mature open-source Diskless Kafka impl. How does this compare to KIP-1150?
The main difference is the layer where diskless is introduced. KIP-1150 makes it a diskless topic path with a Batch Coordinator; AutoMQ tries to keep the existing Kafka broker/leader path and swap out the storage underneath.
That makes the designs feel quite different even though both end up using object/shared storage.