Skip to content
Kestrel Labs · Web · 2024

Real-time reporting for a data platform where the previous dashboard timed out on any query worth running.
  • 18×Median query speed-up
  • 50M+Rows queried live
  • 99.98%Uptime since launch

The challenge

Kestrel's customers were on the platform to answer questions about their own data, but the dashboard timed out on anything spanning more than a week. The interesting queries were exactly the ones that failed, so customers exported to spreadsheets instead — and stopped seeing the product's value.

What we did

  1. 01

    Moved the analytical workload off the transactional database onto a column store built for it, with rollup tables for the common ranges.

  2. 02

    Streamed results over WebSockets so charts paint progressively instead of blocking on a complete response.

  3. 03

    Added query-shape caching in Redis, which absorbed the large majority of dashboard loads without touching the warehouse.

  4. 04

    Instrumented every query path so slow shapes surface in monitoring before a customer reports them.

The outcome

Median query time fell from just over nine seconds to half a second. Dashboard sessions more than doubled in length, and export-to-spreadsheet usage dropped sharply.