[Roadmap] Planning 0.9 Release

[wuchong]

Timeline Plan

• Feature Freeze: Jan 15, 2026
• Release: Jan 30, 2026

Milestone Board

https://github.com/apache/fluss/milestone/5

Highlight Features

List of features announced by contributors and committers that are likely to be ready for the feature freeze

Full DataType Support & Schema Evolution

• 🚨 P0: [Umbrella][Feature] Support for complex data types #816
• 🚨 P0: [Umbrella][Feature] Schema Evolution Support #1154 @loserwang1024
  • support schema changes on Fluss tables (e.g., add column)
  • schema evolution for tiering service.

Log Table

• 🚨 P0: Rescale Bucket: FIP-14 @LiebingYu
• [Feature] Support filter pushdown for Log Tables #197 @platinumhamburg

PK Table

• Auto-Incremental Column: FIP-16 @xx789633
• Global Secondary Index @platinumhamburg
• Streaming KV Scan: FIP-17 @xx789633 @polyzos
• Aggregate Merge Engine: [Feature] Introduce Merge Engines for Primary-Key Table #212 @platinumhamburg
• Support Compacted LogFormat Support Compacted LogFormat #1605 @polyzos

Streaming Lakehouse

• Enable Union Read on Lakehouse Catalog (e.g., Iceberg REST Catalog) @luoyuxia
• Enable LakeStream on the existing lake table @luoyuxia

Cluster Operations

• Replica Rebalance: Fluss cluster supports rebalancing #1144 @swuferhong
• Coordinator HA: [Feature] CoordinatorServer Supports High-Available #188 @zcoo

Stabilities

• 🚨 P0: Refactor metadata cache and async updating mechanism in client side #483 @wuchong
• Use Leader Epoch rather than High Watermark for Truncation #673
• Remove TableInfo from server MetadataCache #982
• [Umbrella] Add backward and forward compatibility E2E tests #1161

Flink Connector

• Support $changelog system table: Support $changelog auxiliary table for flink connector #356 @MehulBatra
• Support $binlog system table: Add binglog system table support for fluss tables #978 @MehulBatra
• Support Flink 2.2 (will be released in Nov.)

Spark Connector

• [Umbrella][Feature] Support spark engine #155 @Alibaba-HZY

Adds native support for Spark as a computation engine, enabling efficient batch and streaming and interactive query processing.

Trino Connector

• feat: support trino engine #1923

Java Client

• Java Client POJO Support: [Client] Java Client Add Pojo Support #1731 @polyzos

Filesystems

• Azure: [file system] Add Azure Blob Storage #1311 @dreger1997
• GFS: Implement obtainSecurityToken for google filesystem #993
• MiniO: [File system] Add AssumeRoleWithWebIdentity STS functionality for temporary credentials to S3 file system to support MinIO #621

Others

• [metrics] OpenTelemetry metric reporter #181

---

We welcome any ideas, suggestions, or feedback from the community! Your input is invaluable in shaping the future of Fluss. Please feel free to comment on the discussion and linked issues.

[zerolbsony]

What something can I do for it?

[MehulBatra]

We should add Javadoc Integration for fluss client, flink, common as a part of 0.9 roadmap.
@wuchong @polyzos ?

[michaelkoepf]

2 minor comments:

• for the otel connector, there is an associated PR for the docs that should also be merged; [docs] OpenTelemetry Metrics Exporter Docs and Quickstart Guide #1270
• for the minio support, the first step to support it will be [file system] Add configuration option to enable/disable token delegation process for S3 file system (allows support for MinIO and other S3-compatible object stores) #1246. i would postpone [File system] Add AssumeRoleWithWebIdentity STS functionality for temporary credentials to S3 file system to support MinIO #621.

[macdoor]

How about Kafka?

[michaelkoepf]

@macdoor we discussed this in yesterday's community call and concluded it's not a current priority. no resources from the current committers will be allocated.

the reasoning is: (1) the kafka protocol limits users to a subset of fluss features, and (2) based on experience from other projects, maintaining kafka compatibility is cumbersome.

that said, we’re always open to contributions. if you're interested in contributing this feature, feel free to pick it up!

[qiuyunbuyu]

Hi @michaelkoepf Can I understand that the current progress and subsequent discussions related to "Kafka Protocol Compatibility" are all maintained in this issue #486

[polyzos]

@qiuyunbuyu that is correct

[caozhen1937]

I would like to propose adding a feature that is Support multiple datalake formats for different tables within the cluster.

It is from #1767.

[1272870698]

Has 0.8 been released yet? Wasn't Coordinator HA present in 0.8? Why is it still there in 0.9

[wuchong]

Hi @1272870698 , we can't deliver Coordinator HA in 0.8, so this feature is postponed to 0.9.

However, Fluss already supports deployment on Kubernetes and Helm. The Coordinator is deployed as a ReplicaSet, so if a pod fails, Kubernetes automatically restores the desired number of replicas. This provides a basic level of high availability out of the box from K8S.

[ka-steve]

Is there any chance that adding authentication via temporary credentials to AWS S3 can be included in this version? There is a PR for it already (1245 if I recall correctly). v0.8 only allows permanent credentials to be used. There are companies that flat out ban permanent credentials and only allow STS/IAM based authentication (which I agree with as well), and these AWS-using companies currently can't even start using Fluss. (Amazing project btw, and great that it got into the Apache family!)

[wuchong]

@luoyuxia could you help to check that whether this is already supported or how many works need to do for this?

[michaelkoepf]

hi.

@wuchong i think i can also answer it.

@ka-steve

#1245 introduces the ability to configure fluss so that each component authenticates itself separately via its own credentials/credential providers, rather than relying on those distributed by the fluss server. this includes authentication via aws credential providers that use temporary aws credentials (e.g., org.apache.hadoop.fs.s3a.auth.IAMInstanceCredentialsProvider). in other words, if all fluss components (server/client) run on aws, #1245 allows each component to authenticate itself separately using temporary credentials.

currently, fluss relies on token delegation and does not support separate authentication for each component. this is because the server attempts to obtain a security token via sts using its access key and secret (embedded long term credentials), see here. the client expects to receive a security token from the server on initialization of the file system and fails if there is no token, see here. #1245 will allow to deactivate token delegation and all components are free to use whatever credential provider they want.

further details are also available in the updated documentation for #1245, see here.