Recover engineering time
Reproduce and diagnose regressions locally.
Customer case study
Cimpress uses recorded mocks to test upgrades to its online catalog locally without relying on live upstream services.
Reproduce and diagnose regressions locally.
Reduce shared environment use where local replay can replace it.
Catch upgrade regressions before release.
Cimpress’s online catalog is the backbone of a nearly $4 billion eCommerce business, powering VistaPrint, PrintBrothers, National Pen, and others. The team has to keep pace with security and scalability upgrades while protecting existing functionality. As Cimpress uses AI to speed up software development, Jeff Kwan, Principal Architect, needed repeatable regression checks he could run locally and later in CI.
Working in tandem with ChatGPT Astra, Jeff pared down a Speedscale snapshot and restored selected products into local PostgreSQL and Elasticsearch. He replayed 824 inbound requests against the local catalog service with a cold cache.
In a second recording pass, proxymock captured outbound calls to real upstream services. A Python script deduplicated the captures and replaced real OAuth credentials with synthetic values, yielding 1,222 mocks.
Running production-derived checks in the developer inner loop gives engineers feedback while they are still working on the change, when regressions are easier to reproduce and diagnose.
With live passthrough disabled in proxymock, .NET 3.1 and .NET 10 each matched the expected results for all 824 requests, with zero unexpected mock misses and zero live passthrough requests. The mocks reflect upstream state at the second capture and form a fixed baseline alongside the database seed and refreshed expectations. CI integration is planned next.
Just spent a couple of hours with a Speedscale snapshot and Astra paring it down and building some mocks and most critically, extracting just the data I need from the production databases to have a slim replay repro dataset. So awesome
Jeff Kwan, Principal Architect, Cimpress
In its earlier load testing work with Speedscale, Cimpress reported an 80% reduction in Black Friday testing time, from a typical four-week process to three or four days. Jeff’s team replayed captured customer traffic with realistic product data and used the results to identify the load limits of its service.
Speedscale gave us clear insight on the upper load limit of our software. Whereas with K6, it was more binary, like a PASS/FAIL grade. Speedscale took it one step further to help us understand, based on transactions per second, exactly where our breakpoints were. We felt more confident in where our service could be.
Jeff Kwan, Principal Architect, Cimpress
Jeff also used Speedscale to identify and fix a critical failure that occurred only 3% of the time.
Replay real requests and mock live dependencies to catch regressions before release.