DevOps engineer based in Moscow. I work with Kubernetes, PostgreSQL/ClickHouse/Cassandra, Terraform and Ansible. This is where I write short, practical notes.
HPA, VPA and KEDA: when to use which Kubernetes autoscaler
The three autoscaling mechanisms in Kubernetes solve different problems — mixing them up is usually what leads to over-provisioned resources or OOM kills.
Where to start when a PostgreSQL query goes slow
EXPLAIN ANALYZE is only the start. A order of operations that saves time when debugging a specific slow query in production.
Terraform across multiple clouds: what's actually worth reusing
A full abstraction layer over AWS/Yandex Cloud/VK Cloud almost never pays off — but a few things are still worth sharing.
Alert fatigue: cutting the noise without losing real incidents
The problem is almost never too few alerts — it’s too many. A look at what actually helps cut false positives.
GitOps: what it actually solves (and what it doesn't)
GitOps gets sold as a silver bullet for deployment problems. It solves a narrower, more specific problem than the pitch usually suggests.
Chaos engineering without a Netflix-sized budget
You don’t need a dedicated chaos team to get value from deliberately breaking things — a few cheap, low-ceremony experiments go a long way.
The hidden cost of YAML
YAML didn’t create the complexity in modern infrastructure tooling, but it hides just enough of it to make things worse.
Tetragon and eBPF-based security: runtime visibility for Kubernetes that doesn't lie
A deep look at what eBPF actually changes about runtime security in Kubernetes, how Tetragon fits into that picture, and where it still needs help from the rest of the stack.
Running Cassandra in production: what actually matters once the honeymoon ends
Cassandra is forgiving during a proof of concept and unforgiving under real production load. A practical rundown of the decisions that determine which one you get.
Kafka in production: the mistakes that don't show up until later
Partitioning, replication and consumer group design decisions in Kafka that look fine at low volume and become expensive exactly when traffic grows.