What’s New in Big Data Frameworks? ClickHouse, Fluss, Delta Lake, StarRocks & More (Mar 2026)

Update Overview

This article aggregates the most recent updates from several prominent data‑processing frameworks released during the Chinese New Year period, providing version details, new capabilities, security patches, and expert commentary to help practitioners stay current.

1. ClickHouse: Five parallel releases in one week, covering three stable branches and LTS

Release dates : 2026‑03‑01 ~ 03‑02 | Update type : Feature updates & security fixes | LTS version : v25.8.18.1‑lts

In a single week ClickHouse shipped five versions—v26.2.3.2‑stable, v26.2.2.9‑stable, v26.2.1.4‑35‑stable, v25.12.8.9‑stable and v25.8.18.1‑lts—covering the v26.2, v26.1, v25.12 stable branches and the v25.8 LTS branch.

ClickStack embedded observability UI : Integrated monitoring panels display cluster performance metrics and query profiles without external tools.

TOTP authentication : Native support for time‑based one‑time passwords, meeting compliance requirements.

Google BigLake integration : Added BigLake catalog support, enabling direct queries of Google Cloud data lakes from ClickHouse.

Iceberg table RENAME COLUMN : Added column‑rename capability to complement existing ADD/DROP/MODIFY operations.

Insert deduplication behavior change : All insert operations now enable deduplication by default. To retain the previous behavior, set deduplicate_insert='backward_compatible_choice' on existing clusters.

Distributed vector search : Vector indexes can now distribute search load across cluster replicas, alleviating single‑node memory limits.

QBit data type and text index GA : Graduated from experimental to general availability.

Commentary : The rapid release cadence reflects ClickHouse’s multi‑branch maintenance strategy. LTS users should prioritize v25.8.18.1‑lts, while teams interested in new features can evaluate the latest v26.2 branch. Note the default insert deduplication change when planning upgrades.

2. Fluss v0.9.0‑incubating: Schema evolution and multi‑tenant support

Release date : 2026‑03‑02 | Update type : Major release

Fluss 0.9.0, the second version after entering the Apache incubator, merges over 100 pull requests and focuses on production‑grade capabilities.

Schema evolution (Add Column) : Supports online column addition without downtime. Currently, tables with Lake functionality cannot use this feature until metadata synchronization between stream and lake storage is completed.

Array type support : Both Log tables (Arrow Row format) and the Flink connector now accept Array types, enabling nested data structures.

Auto‑increment column : KV tables gain auto‑increment columns for globally unique IDs, with upsert constraints adjusted accordingly.

LakeCatalog multi‑tenant : Provides data isolation across tenants, a foundation for enterprise‑grade deployments.

Compacted Row as Changelog : KV tables can emit compressed rows directly as changelog streams, reducing data duplication.

TCP‑level back‑pressure : Introduces RequestChannel‑based back‑pressure to prevent EventLoop blockage.

Dependency upgrades : Flink 1.20.3 / 2.1.1, Paimon 1.3.1, Iceberg 1.10.0, Hadoop 3.4.0.

Critical stability fixes : Resolved ISR state leakage causing upgrade deadlocks, busy‑loop issues in ReplicaFetcher during leader election, and metadata inconsistencies in LookupSender.

Commentary : Fluss has completed three major version iterations in six months, delivering essential production features such as schema evolution and multi‑tenant isolation. However, the inability to modify schema on Lake tables highlights ongoing challenges in synchronizing streaming and lake metadata.

3. Delta Lake v4.1.0: Catalog‑managed tables

Release date : 2026‑02‑26 | Update type : Major release

The core change is the introduction of catalog‑managed tables (Preview), moving table commit coordination from the file system (_delta_log) to Unity Catalog services.

Catalog‑managed tables (Preview) : Enabled via the catalogManaged feature, allowing full table operations—create, batch/stream read/write (including Time Travel and DML), historical queries, and OAuth authentication. Commit coordination now relies on the catalog service, enabling multi‑table transactions.

Delta V2 Spark connector : Rewritten using the Delta Kernel API, supporting streaming reads of catalog‑managed tables.

Server‑side planning (Preview) : Delegates scan planning to an external catalog server (Iceberg REST Catalog API), pushing filters, projections, and limits to the server.

Non‑conflicting feature enablement : Deletion Vectors and Column Mapping can be activated on existing tables without blocking concurrent writes.

Atomic CTAS : Unity Catalog‑managed Delta tables now support fully atomic CREATE TABLE AS SELECT operations (with UC 0.4.0). REPLACE TABLE fails fast when conditions are unmet.

Type widening enhancements : Adds a mandatory decimal widening mode; the default automatic widening mode is now always.

Compatibility changes : Minimum Java 17, drop support for Spark 3.5, catalog‑managed tables disallow manual VACUUM, and Maven artifact naming is adjusted with new Spark version suffixes.

Commentary : Catalog‑managed tables represent a significant architectural upgrade for Delta Lake, laying the groundwork for cross‑table transactions and fine‑grained access control. The feature is still in preview and not recommended for production, but immediate benefits can be gained from the non‑conflicting enablement of Deletion Vectors and Column Mapping.

4. StarRocks 4.0.6: Deepening Iceberg integration and table‑level query timeout

Release date : 2026‑02‑16 | Update type : Feature update

StarRocks 4.0.6 introduces 15 feature improvements and 8 bug fixes, with a strong focus on Iceberg support.

Iceberg partition syntax flexibility : Supports parenthesized PARTITION BY (bucket(k1, 3)) syntax and removes the requirement that partition columns appear at the end of the column list.

Host‑level write sorting : New variable connector_sink_sort_scope controls the granularity of sorting during Iceberg writes, improving downstream query performance.

Table‑level query timeout : Introduces table_query_timeout parameter with precedence Session > Table > Cluster, allowing per‑table timeout policies.

Automated snapshot management : Adds ADMIN SHOW AUTOMATED CLUSTER SNAPSHOT command to view snapshot status and scheduling.

FE memory monitoring API : New endpoint /api/memory_usage assists operators in tracking frontend memory consumption.

Bug fixes : Resolves crashes caused by runtime filter issues in Skew Join V2, type mismatches in low‑cardinality rewrites, and JSON column name conflicts.

Commentary : StarRocks continues to invest heavily in Iceberg compatibility, moving from “usable” to “optimally usable.” The table‑level timeout feature offers immediate operational value for clusters with slow‑query tables, while the broader roadmap points to further Iceberg enhancements and Fluss catalog integration.

Quick Updates

Apache Pulsar v4.1.3 / v4.0.9 / v3.0.16 (2026‑02‑19): Security patches for multiple CVEs (log4j, lz4‑java, jose4j) and dependency upgrades (BookKeeper 4.17.3, Netty 4.1.131.Final).

Apache DolphinScheduler 3.4.1 (2026‑03‑01, prerelease): Adds configurable max runtime for workflow/task instances and timeout checks for missing workers; fixes >15 bugs.

StarRocks 3.5.13 (2026‑02‑13): Regular maintenance release for the 3.5 series.

Apache Doris 4.0.3‑rc03 (2026‑01‑30): Release candidate still in testing, not yet GA.

Trends

1. Catalog evolving from metadata index to table lifecycle management

Both Delta Lake’s catalog‑managed tables and ClickHouse’s BigLake integration illustrate a shift toward catalogs that actively manage table lifecycles—handling commit coordination, scan planning, access control, and cross‑table transactions. Unity Catalog and Iceberg REST Catalog are the leading implementations shaping this direction.

2. Streaming storage rapidly evolving but production maturity still needs validation

Apache Fluss has delivered three major versions in six months, adding schema evolution and multi‑tenant capabilities, yet Lake tables still lack schema change support, indicating unresolved metadata sync challenges between streaming and lake storage. StarRocks plans to support Fluss catalog in its 2026 roadmap, and the Flink ecosystem continues to evolve, but broader production validation remains pending.

3. Security compliance becomes routine in framework releases

Pulsar’s CVE fixes, ClickHouse’s TOTP authentication, and DolphinScheduler’s tightened permission checks demonstrate that security features are now standard components of each release, making timely adoption essential for teams subject to enterprise security audits.

Further Reading

ClickHouse Changelog 2026 – Full list of changes for v26.2

The Next Evolution of Delta – Catalog‑Managed Tables (official design discussion)

StarRocks Roadmap 2026 – Includes Iceberg V3 and Fluss catalog plans

Apache Fluss Documentation – Architecture overview and quick‑start guide

ClickHouse February 2026 Newsletter – Includes financing news and K8s Operator updates