Blog
Articles on CI/CD pipelines, Docker, Kubernetes, cloud infrastructure, and observability — practical DevOps engineering notes from real production systems.
147 articles
Claude pushed Anthropic's monorepo from 2,500 to over 30,000 merges a day, and git could not keep up. A close reading of Scion's design, its own figures and numbers, and what teams that will never hit that scale should copy.
Read More
How to design a metrics and monitoring system end to end: instrumentation, pull versus push, time-series storage maths, cardinality, retention and alerting.
Read More
What SLI, SLO and SLA mean, why an SLO sits below 100%, how to compute an error budget, set burn-rate alerts and measure it all in Prometheus.
Read More
Round robin, least connections or IP hash? When each load balancing algorithm fits, with nginx upstream snippets, layer 4 vs 7 and health checks.
Read More
What each AWS Bedrock AgentCore service does, harness versus Runtime, deploying an OpenAI Agents SDK agent, Gateway tools, Cedar policies, and why Indonesian teams deploy to Singapore.
Read More
A build-it Codex SDK tutorial: start and resume threads from TypeScript, choose a sandbox mode, stream events, and wire Codex into a GitHub Actions job that triages failing tests and drafts the fix.
Read More
Easypanel is commercial with a free plan; Coolify and Dokploy are open source. Licence, pricing, Docker Swarm, templates, backups and lock-in compared.
Read More
How the OpenTelemetry GenAI semantic conventions model an AI agent run, and a working OpenAI Agents SDK trace processor that exports invoke_workflow, invoke_agent, chat and execute_tool spans to a self-hosted Grafana Tempo stack.
Read More
A managed platform hands you a browser log stream instead of a shell. Structured JSON, correlation IDs, stdout buffering and the checklist that has to be true first.
Read More
A rolling deploy runs the old release and the new one against one schema for a window you do not control. The deploy mechanics that decide what a migration may do.
Read More
A written convention has no way to fail. Here is how I turned three architecture rules on this site into prebuild scripts that break the build, what each one cost, and the gate that passed cleanly while hiding 168 wrong dates.
Read More
A platform builder is created for your deploy and destroyed after it, so every layer is cold unless you arrange otherwise. Layer order, build context, cache mounts and the registry cache that actually travels.
Read More
A plugin upgrade does not break the build. It changes how the agent behaves, silently, mid-sprint. How to pin a version, notice a move, and roll back.
Read More
A plugin release can install cleanly and ship half a bundle. Here is the GitHub Actions pipeline that gates it: strict manifest validation, the paths and hook scripts nobody checks, a behaviour gate built from the headless CLI, and tagging that survives a failed push.
Read More
What an AI DevOps assistant genuinely shortens during Kubernetes triage — correlating a restart with a deploy, telling OOMKilled from CrashLoopBackOff — and the faults it cannot see.
Read More
n8n and Flowise both deploy in one click on a platform billed in rupiah. Neither survives its first redeploy without a persistent volume, a public webhook URL, a pinned encryption key and authentication.
Read More
The repository signals a PaaS build engine reads to generate a Dockerfile, the four cases where auto-detection guesses wrong, and the override I keep committed.
Read More
Claude Code reads dozens of environment variables. Here are the ones worth setting on a build agent, and the three with side effects you did not want.
Read More
Six variables give you per-user cost, token and tool telemetry. The interesting part is which attributes the cost metric carries, and what content logging asks of you.
Read More
The allowlist behind a working install, the shell exports that never reach background agents, and the one setting that is actually validated at startup.
Read More
A dev container gives every engineer the same bounded environment. Here is the authentication that survives a rebuild, the egress firewall, and the limits of the isolation.
Read More
The GitLab integration is one job and one masked variable. The parts that stall teams are the mention trigger you have to build and a PATH line in the installer.
Read More
Session-scoped loops, cloud routines and desktop tasks solve different problems. Here is which to reach for, plus the jitter and expiry rules that surprise people.
Read More
A self-hosted environment executes Claude Code cloud sessions on your own hosts, with every connection outbound and no inbound port for Anthropic to reach.
Read More