Get the latest tech news
Debugging Azure Networking for Elastic Cloud Serverless
Learn how Elastic SREs uncovered and resolved unexpected packet loss in Azure Kubernetes Service (AKS), impacting Elastic Cloud Serverless performance.
We left this to run for ~10 minutes or so to capture as many drops as possible, and then ‘heavily inspired’ by this GitHub Gist by Ivan Babrou, we converted the stack traces into an ‘easier’ to read Flamegraphs: Given the complexity of our load testing scenario — which involved configuring multiple systems and tools, including Elastic Observability, we also developed a simplified reproduction of the packet loss issue using . In managed platforms like AKS, where Azure controls both the node images and infrastructure, it is easy to overlook the impact of low-level configurations such as network device ring buffer sizes or sysctls like
Or read this on Hacker News
