Velox

Please join us in welcoming Liu Ping as a new maintainer on the Velox project! 🎉 Ping has driven a series of improvements and new features in the Iceberg and Parquet space. He'll now maintain Iceberg and help as a maintainer for Parquet. Thank you, Ping, for your contributions and looking forward to more! Full maintainer list: https://lnkd.in/gkAubhRK

Check out our latest blog on Velox's new HashTableCache 👏 authored by Shrinidhi Joshi from Meta, Xiaoxuan M. from Meta, and Ke Jia from IBM. https://lnkd.in/g24X6Pgx Broadcast joins avoid shuffling the large probe side, but rebuilding the same hash table on every task wastes CPU, memory, and I/O. HashTableCache fixes this with a build-once, reuse-many model: one task builds, others wait and reuse, late arrivals skip construction entirely. Learn more down the design including builder/waiter coordination, memory lifecycle, Gluten integration, and what's next (eviction, spill-aware caching). Shout out to our awesome community for building and sharing the details on these types of features! 🙌 Meta Open Source Apache Gluten IBM IBM Data, AI & Automation

📢 We're beyond excited to announce #VeloxCon China 2026 - save the date! 📅 December 5, 2026 | 9:00 AM – 7:00 PM 📍 Alibaba Xuhui Riverside Park, Xuhui District, Shanghai The CFP is now open. If you're building with Velox, Spark/Gluten, Presto, or in the intersection of AI and data management and want to share your work with the Velox China community, submit a proposal here: https://lnkd.in/gzah2AgF Full event details: https://lnkd.in/gfHFE6yS Registration will be opening soon! Stay tuned 🎉

I just posted a new blog on Velox. I looked into why a RIGHT SEMI JOIN rewrite can cause a 10x slowdown under data skew, and how we fixed it to get a 9.8x speedup. Read the full story here: https://lnkd.in/dg6pXWgg

#VeloxCon 2026 recordings are now posted 👏 catch everything you missed. This year's event brought together hundreds of the Velox community in-person, with over 2K combined livestream views across both days. The energy in the room matched the momentum we're seeing across the Velox ecosystem 🚀 Sessions covered the latest in Velox development, real-world deployments, and what's coming next from contributors across the community. Whether you were there or tuning in from across the world, the full playlist is now available. Stay tuned for our next community event 👀 https://lnkd.in/giKHKVCz

Day 2 of #VeloxCon is LIVE in 30 minutes 🔥 If you couldn’t join us yesterday (or want more), we’ve got you - tune in to the Day 2 livestream from anywhere 👍 More talks today from engineers at Nvidia, Meta, Oracle and more. Jump in, catch a session, and see what’s happening across the ecosystem. Watch here → https://lnkd.in/gCvwbtZn

Big news - we’ll be LIVE at #VeloxCon starting in 30 minutes 🔥 If you're not able to join us in-person, we're happy to share the VeloxCon livestream - tune in from anywhere 👍 If you’re into execution engines, data infra, or how systems like Velox are powering modern analytics + AI workloads, come join us. Talks all day from engineers at Meta, IBM, Microsoft and more. Watch here → https://lnkd.in/gAr9e5Qw

⭐ #VeloxCon Keynote Spotlight ⭐ VeloxCon starts tomorrow! Join Maria Basmanova from Meta who will share the vision and architecture behind Axiom, Meta’s next-generation compute platform. She'll cover how it builds on Velox and the design principles shaping future execution engines. Last chance to grab a seat! Hope to see you there. 📅 April 29-30, 2026 📍 Meta HQ, Menlo Park https://veloxcon.io/

Migrating UDFs to Velox can be slow and manual, unless you automate it. In this lightning talk at #VeloxCon, Miaojiang (MJ) Deng and Sandeep Thandassery from Meta will share a workflow where an AI agent generates C++ UDFs and validates them with differential fuzz testing against the original Java versions. The result: significantly faster migration with high confidence in correctness. We hope to see you TOMORROW! 📅 April 29-30, 2026 📍 Meta HQ, Menlo Park https://veloxcon.io/

Bringing iteration into Velox 🔍 learn more at #VeloxCon Zhichen Xu, Ph.D., Junjie Q. and Jingfang Liu from Meta will explore how Velox can support iterative computation for workloads like clustering, graph processing, and agent pipelines. They’ll walk through how existing primitives (exchange, aggregation, vector ops) could be composed into iterative workflows, along with the challenges around coordination and state. Join us for a deeper look at how execution engines may evolve. 📅 April 29-30, 2026 📍 Meta HQ, Menlo Park https://veloxcon.io/