apache / apache/hudi

[DISCUSSION] Deltastreamer - Reading commit checkpoint from Kafka instead of latest Hoodie commit

Open
#11,268 5 comments 0 reactions 0 assignees View on GitHub
area:incr-processing area:ingest type:community-support
Dominant language
Java
Stars
6.2k
Forks
2.5k
Avg merge
2d 8h
Merged PRs (30d)
111

Description

Our requirement is actually this: Supporting multiple deltastreamers writing to a single hudi table

[https://github.com/apache/hudi/issues/6718](https://github.com/apache/hudi/issues/6718)
[HUDI-5077](https://issues.apache.org/jira/browse/HUDI-5077)

We reviewed the deltastreamer code and noticed that the deltastreamer can read commits from Kafka consumer groups if it doesn't find the last checkpoint from Hoodie.

We're considering modifying the code to always read commits from Kafka, based on a new flag, by making changes in the Hudi deltastreamer.

Do you foresee any nuances or issues with this approach?

**Environment Description**

* Hudi version : 0.13.0

* Spark version : 3.3.2

* Storage (HDFS/S3/GCS..) : S3

* Running on Docker? (yes/no) : no

Contributor guide

No contributing guide indexed for this repository

Research direction

Start with the Hudi Deltastreamer checkpoint-handling entry point and review the related issues 6718 and HUDI-5077. Compare the existing fallback from Hoodie commits to Kafka consumer groups with the proposed always-read behavior, including the single-table, multiple-writer requirement. Done requires an agreed design and clarified implementation scope, not just a code change.

Written by the indexing model from the issue text.

Assessment

Tech stack
java, kafka, spark
Domain
data-engineering, distributed-systems, stream-processing
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
18/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.