The beautiful chaos of inventory peaks
Kroger is like Walmart, but it keeps the names of its chains. NeoLoad was my main weapon there — and if LoadRunner and JMeter had a child, that child would be NeoLoad.
NeoLoad: the best of both worlds
NeoLoad has the controller that JMeter lacks but LoadRunner has, plus the XML script structure that JMeter uses. It really does sit in the middle, and for Kroger it was the right tool for the job.
The inventory: 1500 TPS into Yugabyte
Two main areas at Kroger. The eCommerce side was fun, but lots of images, cookies, and logic. The inventory side is the one I loved most.
Inventory ran on Yugabyte, a distributed cloud database with extremely low latency. The flows came from SKU scans: workers scanned every SKU with a handheld gun, and when the whole scan finished, everything fired as a batch. The inventory landed in waves, roughly hourly, and the load was pure spikes — there was no way to smooth it. Peaks up to 1500 transactions per second. Attacking those spikes was something impressively beautiful, because they went straight into Yugabyte through RabbitMQ.
The pharmacy: Citrix, HIPAA, and image assertions
Then there was Helen Wellness, specifically the pharmacy area. HIPAA certified. The application ran over Citrix, which was challenging and enriching for me. I knew a bit of Citrix, but it was a framework, so I researched it heavily.
We ran tests of 100 to 200 concurrent users over Citrix, which is a lot — Citrix consumes serious bandwidth just transporting the screen. And assertions were complicated, because you had to convert the image back to text to verify anything.
That kind of work built my career in ways I didn't expect.