Tim Derzhavets
I run production infrastructure, and I write about what breaks.
Kubernetes clusters and the data platforms underneath them, the pipelines that ship to them, the messaging and the databases they lean on. The writing here is the notebook from that work: the RDS Proxy connection storm, the gRPC channel that cost ten milliseconds, the Terraform-to-OpenTofu move, the Rancher fleet.
Outside of that I run DevClusterAI, a small studio that builds a small number of products and keeps them running.
- Writing
- 55 notes, 2024–2026
- Project
- devclusterai.com
The project
DevClusterAI
Small software, kept running. A register of five products — a cloud dev box with Claude Code on it, hosted plain-text accounting, an AI sales-practice partner, a content pipeline with a human gate, and a memory service on the way — with the date each one last shipped.
- REG-01 CC on Demand cloud dev box 119ms
- REG-02 Countbean hosted ledgers 359ms
- REG-03 Closer sales sparring 281ms
- REG-04 Memory pg · vector · graph coming
- REG-05 Altineris editorial pipeline 193ms
4 of 4 public pages answered when this page was built, 2026-09-10 23:12 UTC.
Writing
all 55 notes →- RDS Proxy in Production: Solving Connection Exhaustion and Sub-Minute Failover Eliminate Lambda connection storms and achieve 30-second database failover with RDS Proxy. Production deployment guide with Terraform and monitoring.
- Java vs Python in Production: Performance, Concurrency, and When Type Safety Actually Matters Real performance differences between Java and Python in production: concurrency models, GC behavior, and when type safety prevents 2 AM incidents.
- Docker Images vs Helm Charts: Choosing the Right Abstraction Layer for Your Kubernetes Deployments Docker and Helm solve different problems—images package applications, charts orchestrate deployments. Learn when you need each.
- Building Distributed State Stores with NATS JetStream: KV and Object Storage Patterns Replace Redis and S3 with NATS JetStream's built-in KV and Object stores for simplified cloud-native state management.
- Building a Cost-Efficient GitLab Runner Fleet: From Single Server to Auto-Scaled Infrastructure Evolve GitLab Runners from basic setup to production-grade auto-scaling with real cost analysis, executor trade-offs, and fleet management.
- NATS vs Kafka: A Decision Framework for Cloud-Native Messaging at Scale Operational trade-off analysis with runnable Go code showing where NATS and Kafka each break down under real cloud-native workload patterns.