Gluten 2026 Roadmap

[weiting-chen]

Description

Time to define 2026 features. Feel free to add more or comment.
For more information about the achievements in 2025 Roadmap, please check #8226
For more information about the achievements in 2024 Roadmap, please check #4709

Apache Gluten graduated to an ASF Top-Level Project in February 2026. This roadmap tracks the community's goals for 2026.

---

Spark Compatibility

• Spark 4.0/4.1 GA support — fix all disabled test suites (Spark 4.x: Tracking disabled test suites #11550, [VL] Track on Spark-4.1.x failed unit tests #11400)
• TIMESTAMP_NTZ type support ([VL] Support TIMESTAMP_NTZ Type #11622, [GLUTEN-11622][VL] Add basic TIMESTAMP_NTZ type support #11626)
• Variant shredding support for Parquet reader and writer (Spark 4.0) ([VL][Spark-4.0] Add Variant shredding support for Parquet reader and writer #11371)
• Parquet type widening support (SPARK-40876) ([VL] Support type widening in Parquet reader (SPARK-40876) #11683, [GLUTEN-11683][VL] Add Parquet type widening support #11719)
• Other Spark Type Support: DayTimeIntervalType, CalendarIntervalType, TimeType, GeographyType, GeometryType
• ANSI mode support ([VL] Add ANSI mode support #10134)
• Complete remaining unsupported Spark functions ([VL] Unsupported spark function list [please leave a comment if you plan to pick some] #4039)
• Deprecate Spark 3.2/3.3 and JDK 8

Performance Optimization

• Use runtime stats to choose hash build side (HashBuild OOM caused by incorrect build side #11774, [GLUTEN-11774][VL] Use runtime stats to choose hash build side #11775)
• Bloom filter optimization — translate might_contain as subfield filter (Bloom filter optimization #11771, [VL] Translate might_contain as a subfield filter #11708, [GLUTEN-11708][VL] Translate might_contain as a subfield filter for scan-level bloom filter pushdown #11711)
• Partial project UDF optimization (Partial Project UDF optimization #11783)
• Execution-aware dynamic join strategy selection after filter execution (Proposal: Add execution‑aware dynamic join strategy selection after filter execution #11808)
• Parquet metadata check limit optimization (Parquet metadata check limit optimization #11782)
• Push dynamic filters to shuffle reader with per-block column statistics ([VL] Push Velox dynamic filters to shuffle reader #11605, [GLUTEN-11605][VL] Write per-block column statistics in shuffle writer #11769)
• Multi-core per task ([VL] Use multiple threads in the same executor #7810)
• Spill enhancement — streaming window functions ([VL] Spill related issues tracker #3030)
• Pick split with most data prefetched ([VL] pick split with most data prefetched #11821)
• Complex type Row-to-Columnar optimization

Native Engine Integration

• Switch to upstream Velox official release ([VL] upstream OAP/Velox commits to upstream #8782)
• Upstream useful Velox PRs not merged from Gluten community ([VL] useful Velox PRs not merged into upstream #11585)
• Bolt backend integration — ByteDance native engine with LLVM JIT (WIP: add bolt backend in gluten #11261, Add a new backend: Bolt #10929)
• ClickHouse backend upgrade ([CH]Update Clickhouse Version (Branch_25.12_20260310) #11734)
• Kafka read support for Velox backend ([VL] Adding kafka read support for Velox backend #11801)

Data Lake & File Formats

• Full Iceberg support — map write configs with Velox (Map iceberg configuration with Velox configuration #11703, Added iceberg write configs #11776)
• Native Parquet write for complex types (Struct/Array/Map) ([VL] Enable native Parquet write for complex types (Struct/Array/Map) #11788)
• Iceberg equality delete MOR table support ([WIP][GLUTEN-8055][VL] Support read Iceberg equality delete file MOR table #8056)
• Hudi MOR table support
• Delta Lake feature parity
• JSON file format support
• ORC writer support

GPU & Hardware Acceleration

• GPU BHJ bug fix ([VL] GPU failed on BHJ #11794)
• Multi-threaded decompression in GPU shuffle reader (Support multi-threaded decompression in the GPU shuffle reader #11779, [GLUTEN-11779][VL] Support multi-threaded decompression in the GPU shuffle reader #11780)
• GPU code cleanup and stabilization ([VL] Clean up GPU code: remove dead/redundant code #11824)
• ARM SVE optimization
• FPGA accelerator exploration

Flink Integration

• Fix Flink memory leak with RocksDB state backend ([FLINK] Memory leak when trying to run nexmark with rocksdb state backend #11791)
• Fix Flink CI build failures ([FLINK] CI build failure for flink-test #11793)
• Stabilize Flink + Velox from experimental to beta
• Nexmark benchmark support ([FLINK] Nexmark Q3 submission error #11790)

PySpark & Python Ecosystem

• PySpark Python UDF support
• Arrow UDF support
• Fix Python UDF/UDTF test suites on Spark 4.x (Spark 4.x: Tracking disabled test suites #11550)

Stability & Quality

• OOM prevention and memory stability (OOM but memory is enough #11747, [VL] Distinct aggregation OOM when getOutput #8025)
• Full fuzzer support and result mismatch resolution ([VL] Result mismatch issues tracker #4652)
• Timezone edge case fixes ([VL]Velox rejects America/Coyhaique timezone (CentOS 9) results Gluten job failure #11597)
• Complex type validation in native engine ([VL] [BUG] Complex types already supported in Velox are considered not supported by Gluten #11746, [VL] CrossRelNode's expression is not validated in native validation #11678)
• Support getting C++ stack traces via GDB from Spark UI (Support get cpp stack using gdb from Spark UI #11677)
• collect_set ignoreNulls support ([VL] Enable collect_set ignoreNulls #11826)

Build, CI & Developer Experience

• VCPKG for macOS build ([GLUTEN-9577][VL] Enable VCPKG for MacOS build #11563)
• IWYU tool for C++ code format checking ([GLUTEN-11287][VL] Use IWYU tool to check code format #11287)
• Cache Maven dependencies in CI ([VL][CI] cache maven deps m2 repo #11655)
• Fix ARM64/aarch64 build issues (Build failures when building Apache Gluten with Velox on Azure Arm64 (Ubuntu 24.04) #11633, Bundle build failure on Azure Cobalt aarch64 #11639)
• Fix glog macOS build breakage ([Build] branch-1.5 build fail on macOS due to breaking change introduced by glog #11763)
• Docker testing improvements ([VL] Caching java dependencies in testing docker #11501)
• Remove Arrow-CSV dependency ([VL] Remove Arrow-CSV dependency #11591)

Community & Governance

• Complete TLP graduation remaining tasks ([Umbrella] Apache Gluten Graduation Tasks #11713)
• Quarterly releases: 1.7 / 1.8 / 1.9 / 2.0
• GlutenCon 2026
• Expand ecosystem documentation (Kyuubi, Celeborn, Velox)
• openEuler OS support