Community Weekly Report
Aug 24 – Aug 30, 2026
00Overview
Doris maintained strong momentum this week with 91 merged PRs, 51 contributors, and 33 new issues, with effort concentrated across kernel engineering, real-time warehousing, and cloud-native themes. The highlights signal a coordinated push into depth-first optimization—end-to-end TLS, eager aggregation pushdown, runtime filters reaching scanners and segments, and Iceberg/Variant metadata unification—while cloud-mode per-column compression and single-rowset compaction further strengthen cost and efficiency under storage-compute separation. Taken together, Doris is steadily evolving from a real-time analytics engine into a unified, multi-modal, secure, and cloud-ready data platform.
01Highlights
02By Scenario
Multi-modal Lakehouse
01The multi-modal lakehouse scenario tightened resource lifecycles across lake formats and native indexes. The community restored Parquet column ownership after interrupted reads, bounded legacy Bloom-filter metadata reads, and safely released Iceberg FileIO and Hudi fsView native resources when metacache entries are evicted; meanwhile SNII ANN now stages to disk instead
- [fix](parquet) Restore column ownership after interrupted reads
- [fix](parquet) Bound legacy Bloom filter metadata reads
- [fix](be) Use file-backed staging for SNII ANN indexes
- [fix](hive) Preserve simple HMS user identity
Multi-modal lakehouse work centers on stability and pushdown. The ANN index gains pre-filter column predicates for TopN (key), so residual-filtered vector queries no longer abandon the ANN path; Iceberg, Paimon, and Hive-side fixes stop schema-mapping crashes, repair HTTP resource probing, and preserve nested field casing. This week also migrates the
- [feature](ann-index) Pre-filter column predicates into ANN TopN searc…
