Skip to content
This repository has been archived by the owner on Dec 6, 2023. It is now read-only.

Load testing

Jump to bottom
Luke Church edited this page Jan 20, 2015 · 1 revision

Load testing is currently done with mini-driver scripts.

Results so far: /api/analyse easily handles 50QPS for a small script /api/compile has - as expected - much higher CPU requirements

At 10 QPS, with 2 MVMs running the latency looks like:

Load test for 10-QPS (ms latency, vs test time)

We can see that for a period of time the VMs are overloaded, until the auto-scalar adds another VM, after that the latency quickly stabilises again.

Roughly speaking each instance can process around 4QPS for compiles, and around 50QPS for analysis.

Going forwards to we'll add this to our metric tracking.

Clone this wiki locally