Blog
Insights on observability, monitoring, and engineering best practices from the Coroot team.
Disk I/O profiling with eBPF
CPU profiles go flat when a database waits on disk. coroot-node-agent now profiles the wait: which code path did the I/O, how long, how many bytes. Plus a retention job caught red-handed.
Nikolay Sivko
LLM inference on Kubernetes and observability
KServe, Dynamo, Ray Serve, Kthena and llm-d turn each model into a pile of pods. Coroot now shows each model deployment as one application, with an SLO that fits model servers.
Nikolay Sivko
Putting Amazon ECS on the map
How we got Amazon ECS services onto the same service map as EKS and RDS: one agent per EC2 node, three Docker labels, a metadata endpoint, and a pause container to ignore. Plus why Fargate is out.
Nikolay Sivko
Coroot at KCD Argentina
Coroot recaps KCD Argentina in Buenos Aires, with survey results from 50 attendees on their observability tools, biggest pain points, and time spent on setup and root cause analysis.
Alexander Lamberton
Coroot at Nerdearla & Percona University
Coroot recaps two stops in South America this September: Percona University in Montevideo and Nerdearla in Buenos Aires, with Raspberry Pi demos, robots, bowling, and catching up with the community.
Alexander Lamberton
Percona Live 2026: Connecting With Open Source Communities in Amsterdam
Coroot's recap of Percona Live 2026 in Amsterdam: the new Coroot Percona Edition partnership announcement, conference talks, and highlights from the show floor.
Alexander Lamberton
Comprehensive Postgres observability: is your database healthy, and will it stay that way?
A database has to stay durable, available, and fast, and the failures that break those promises rarely page you until it's too late. Here's how comprehensive Postgres observability tells you your database is healthy now and will keep performing.
Nikolay Sivko
Reproducing split brain on CloudNativePG
We partitioned the primary of a default CloudNativePG cluster and measured how long it kept accepting writes after a replacement was promoted, what that did to the data, and which settings shrink the window.
Nikolay Sivko
It was surprisingly hard to break CloudNativePG replication
I tried to make a Postgres replica fall behind by cutting it off from its primary with Chaos Mesh. It stubbornly refused to lag. Here's the investigation, the logs, and the proof from CloudNativePG's source that a replica has a second, hidden WAL source.
Nikolay Sivko
Let's break autovacuum in Postgres: reproducing failures to make it observable
Every Postgres runs autovacuum and almost nobody monitors it properly. Let's reproduce the common ways it fails and see how Coroot turns each one into a clear, actionable finding.
Nikolay Sivko
The hard part of AI root cause analysis is no longer the model
We ran one real incident through eleven LLMs. The reasoning turned out to be the easy part; the real challenge is the harness that feeds the model the right context.
Nikolay Sivko
Observability on Windows, before eBPF is production-ready
Coroot's agent now runs on Windows. The story of building it on ETW instead of eBPF: what came easy, what was hard, and what we left for later.
Nikolay SivkoTry Coroot Free
Get full-stack observability in minutes with zero code changes. eBPF-powered monitoring with AI-guided root cause analysis.