Alert on symptoms, not causes
When you are bringing a new system to production you know that you ought to define SLIs, set up instrumentation,
How about we forget the concept of test types?
I have found that the concept of test types (unit, integration, and so on) does more harm than good. People
How organisations cripple engineering teams with good intentions
I believe that engineers are at their best when they complement strong technical expertise with skills from other disciplines such
Migrating an Eureka-based microservice fleet to Kubernetes
I have written about how a lot of the value in the our internal Platform at Adevinta goes in the glue between systems. This post gives a deep dive into some of the technical problems we find as we transition teams into our Kubernetes-based PaaS, and what kind of glue helps us overcome it.
How to build a PaaS for 1500 engineers
This article is based on a presentation I gave as part of AdevintaTalks in Barcelona on November 2019, explaining the strategic principles we used to build an internal platform to support all the online marketplaces in the group (including some of the biggest in Europe and South America).
"Kubernetes made my latency 10x higher!".. or maybe not?
As we migrate teams over to Kubernetes, I’m observing that every time someone has an issue there is a knee-jerk reaction like the title. Kubernetes is to blame. Investigation usually shows that the explanation boils down to the nuances of blending complex systems together.
Sizing Kubernetes pods for JVM apps without fearing the OOM Killer
Migrating teams to from on-prem/EC2 infrastructures to Kubernetes we hit some issues with resource allocation of JVM apps. I will explain how running in Kubernetes forces us to think about capacity planning more than we’re used to, changing some of the assumptions we made before containers.
GC forensics by example: multi-second pauses and allocation pressure
This post will analyze a Hotspot GC log exhibiting large GC pauses (> 1 min) leading to allocation pressure and system load as a cause of pathological behaviour on Hotspot’s garbage collector.
How does the default hashCode() work? (and why does it affect biased locking?)
In which scratching the surface of hashCode() leads to a speleology trip through the JVM source reaching object layout, biased locking, and surprising performance implications of relying on the default hashCode().