About
I’m Srikanth. I work on OLAP query engines at e6data. Some of that is the part of an engine that decides how little work a query can get away with: indexes built from the queries themselves, vector search inside SQL, and the Iceberg metadata layer underneath it all. The rest is about where that work actually runs: distribution and workload management.
Lately I’ve been looking at how agents use OLAP engines. Agents query differently from people and dashboards, and I’m working on what it takes to get an engine ready for that, both in the interface it offers and in the infrastructure it requires.
Most of what I care about comes down to a few beliefs:
- The cheapest row is the one you never read. Pruning, indexing and good metadata usually beat a faster operator.
- On an open table format, the commit path is part of the engine’s performance. Table formats are now a performance problem, not just a storage one.
- Measure before you argue. Most of the arguments I’ve lost, I lost to a benchmark.
Before data infrastructure I spent a long stretch on developer tooling. From 2015 to 2021 I was the lead developer of Gauge, the open-source test automation framework sponsored by ThoughtWorks, and I’ve carried on as an open-source maintainer in my individual capacity since. That’s where I learnt to care about the people on the other end of an API.
The notes are write-ups of the engine work, with the decisions and the mistakes left in.