Oxide Computer Company builds private cloud computers–co-designing hardware and software that works together. Our choice to own so much of the design was to enable systemic control and observability: we collect data from the smallest hardware sensor to the distributed control plane software. This talk covers the Oximeter Query Language (OxQL), the domain-specific language used to query and analyze this telemetry data, both for Oxide engineers and also external customers.

The choice to build a custom query language in OxQL was not taken lightly – given the choice of SQL or death, we naturally prefer SQL! Our query system did in fact initially expose a SQL interface, but issues around performance, efficiency, and expressiveness ultimately led us to reconsider. OxQL is the result. The language includes analysis methods tailored to timeseries data; an expressive, pipe-based syntax making both queries and the language itself easy to modify; and a clear interface boundary that lets Oxide incrementally improve the language as needs evolve. This talk will explore the underlying data model, syntax, semantics, and query engine of OxQL. We also present some of the criteria we used for evaluating the “choice” of SQL or death, which may help others facing the same question to understand when a special-purpose DSL is appropriate.

This talk is part of the SQL or Death? Seminar Series.

Zoom Link: https://cmu.zoom.us/j/93441451665 (Passcode 261758)

Bio:
Ben Naecker is a software engineer at Oxide Computer Company, where he works primarily on the distributed control plane, telemetry subsystem, and networking. In the past, he has worked on high-performance scientific computing to improve the quality and reduce the duration of MRI scans for preventative medicine. He received his PhD in computational neuroscience from Stanford University, where he built large-scale realtime data recording and visualization software for neuroscientific experiments.