September Edition:

We're excited to be back with another evening of Database Internals!

This time, we'll look at what happens when open table formats meet native engines. Apache Iceberg V3 and Apache Parquet now carry geospatial types directly in table schemas and metadata, and ClickHouse is learning to keep its MergeTree ingestion speed while continuously landing the same data in Iceberg and Delta. With talks from Snowflake and ClickHouse, the evening will get into the spec-level and engine-level mechanics that let one dataset serve many engines.

As always, there will be plenty of time for networking, discussion, and snacks 🍕.

Event Details:
📍 Venue: AWS, Oskar-von-Miller-Ring 20, 80333
📅 Date & Time: Wednesday, September 24 2026, doors open at 18:30

• Powering Spatial Analytics: Native Geospatial Types in Apache Iceberg

Abstract: Querying spatial data at scale in data lakes has historically relied on a mix of different storage formats, hindering interoperability and effective cross-engine metadata use. The introduction of native geospatial types to Apache Iceberg V3 and Apache Parquet elevated spatial concepts directly into table schemas and metadata.
In this talk, Philipp Fischbeck, Software Engineer at Snowflake, walks through the architectural journey of native geospatial types in Iceberg. We will trace how historical spatial representations evolved into native specification standards, break down the mechanics of planar Geometry versus spherical Geography coordinate systems, and demonstrate how query engines like Snowflake can perform spatial file pruning. Finally, we'll look behind the scenes at the open-source community effort across Apache Parquet and Iceberg that made standardized spatial lakehouse data possible.
Speaker Bio: Philipp Fischbeck is a Software Engineer on the SQL Data Types team at Snowflake, where he works on making complex data shapes, especially geospatial types, convenient and efficient to store and query. His work focuses on Snowflake's core SQL engine capabilities and its integration with open formats like Apache Iceberg and Apache Parquet.

• Interoperability Between ClickHouse and Data Lakes

Abstract: Teams increasingly want two things that usually pull in opposite directions: the raw query speed of ClickHouse for interactive analytics, and their data sitting in open lakehouse formats like Iceberg and Delta so the rest of their stack can use it. Writing directly into Iceberg or Delta is inefficient, a well-known limitation of these table formats. So instead of paying that cost on the write path, you ingest at full speed into a ClickHouse MergeTree table and let an Incremental Refreshable Materialized View asynchronously and incrementally batch the new rows into Iceberg or Delta, giving you both fast insertion and the open copy.
The effect is one dataset with two homes: you keep querying the fast, native MergeTree in ClickHouse, while the same data steadily accumulates in the open lakehouse for Spark, Trino, and anything else you run. Under the hood, it's powered by ClickHouse's streaming queries: each refresh runs a bounded streaming scan that reads only the rows committed since last time and remembers how far it got with a cursor, so the lake copy stays in sync without external orchestration or full reloads. It's simple today, with plenty of room to grow toward broader catalog support, richer transformations, and more efficient large-scale refreshes.

Speaker Bio: Smita Kulkarni is a software engineer at ClickHouse, where she has worked since 2022 on parameterized views, backups, and Azure support. Most recently she has focused on data lakes: Iceberg manifest compaction, Delta Lake support on Azure, and the Incremental Refreshable Materialized View that powers ClickHouse's lakehouse integration. Before ClickHouse, she worked on low-latency software.
Agenda:

🔹 18:30: Doors Open
🔹 18:40: Welcome
🔹 18:45: Talk #1: Philipp Fischbeck (Snowflake)
🔹 19:30: Pizza & Networking 🍕
🔹 20:00: Talk #2: Smita Kulkarni (ClickHouse)