Re: [Discuss] Kafka Enumerator assigns splits based on the current assignment situation.

2024-12-30 Thread Hongshun Wang
Hi Ahmed, Thanks for the additional information about the current assignment strategy problem that I had overlooked. Initially, I thought that Kafka alone was not sufficient for the flip, but now that you've broadened the scope, I am excited to work on it for the common Flink source interfaces. I

Re: [Discuss] Kafka Enumerator assigns splits based on the current assignment situation.

2024-12-30 Thread Ahmed Hamdy
Hi Hongshun, I personally see this as a common challenge for source connectors, it might be a good FLIP candidate for the common flink-source interfaces. I still see a couple of challenges that might be covered in the FLIP, for example currently the Kafka source doesn't wait for reader registratio

Re: [Discuss] Kafka Enumerator assigns splits based on the current assignment situation.

2024-12-29 Thread Hongshun Wang
Hi Ahmed, I want to do two things in this flip. 1. Add current assignments to ReaderRegistrationEvent. Thus, enumerator can know the overall assignments after restarting the jon. 2. Add an distribution strategy interface to the kafka enumerator, thus can flexibly choose how to assign the split base

Re: [Discuss] Kafka Enumerator assigns splits based on the current assignment situation.

2024-12-29 Thread Hongshun Wang
Hi Ahmed, Thanks for your question. > Do you intend to keep the current assignment in the enumerator state or are you only going to use per session assignment status and keep it in memory? Not keep in state, because it's be different with the real distribution when redistribute state if the parre

Re: [Discuss] Kafka Enumerator assigns splits based on the current assignment situation.

2024-12-29 Thread Ahmed Hamdy
Hi Hongshun There is currently a Jira tracking this issue [1]. >Therefore, I want the enumerator to select a reader based on the current assignment situation. Each time, the enumerator should choose the reader whose assigned split number is the smallest. I am not sure I follow how you intend to

[Discuss] Kafka Enumerator assigns splits based on the current assignment situation.

2024-12-26 Thread Hongshun Wang
Hi devs, I'd like to push a new FLIP xxx: Kafka Enumerator assigns splits based on the current assignment situation. Currently, when the Kafka enumerator finds a new split, it will choose a reader using the following algorithm: ```java static int getSplitOwner(TopicPartition tp, int numReaders)