Lasse Canth Hels, Maersk
How do you run a massive observability platform on nothing but spot instances? With extreme difficulty.
In this talk, I'll cover real-world stories from the front lines and offer honest insight into the pros and—mostly—cons of the setup. I will discuss how we minimize disruption from evictions as well as the band-aid solutions we've inevitably had to set up to keep our platform steady in a constant stream of chaos. I will also reveal the issue that we couldn't crack and which finally forced us to move partially away from spot instances.
Join to hear what it takes to realise the extraordinary cost savings promised by spot instances at scale, and the many ways in which we failed to make it work.
Lasse is a software engineer at Maersk. As a member of the telemetry team, he took part in building the Maersk Observability Platform, and now spends much of his time keeping it running. Outside of computing, his interests include speedrunning, powerlifting, etymology, and box office performance.
author = {Lasse Hels},
title = {What I Wish I Knew Before Choosing Spot Instances},
year = {2025},
address = {Dublin},
publisher = {USENIX Association},
month = oct
}
