From Metrics to Miner
A GPU metrics anomaly led us to a cryptominer hiding in plain sight on an exposed SGLang server, and a pickle CVE with no patch.
hugo@hugosetiawan.com ~> ./welcome
Platform engineer building scalable Go services and running the OpenStack and Kubernetes infrastructure behind an AI cloud.
package main
// Profile
var (
name = "Hugo Sulaiman Setiawan"
role = "Platform Engineer"
location = "South Jakarta, Indonesia"
email = "hugosetiawan18@gmail.com"
github = "https://github.com/hugo-setiawan"
linkedin = "https://www.linkedin.com/in/hugo-setiawan"
)
// Focus — click to expand
Backend and platform services in Go: an OpenID Connect identity provider, usage metering from OpenStack Ceilometer events, and an eBPF network observability pipeline (Kafka → ClickHouse) ingesting 65M+ flow records a day. Built to be performant, to scale horizontally, and to be well instrumented with metrics and traces, so I can see how a service behaves in production before something breaks.
Running production OpenStack with Kolla-Ansible (Nova, Neutron, Cinder, Keystone), plus Kubernetes delivered through GitOps with ArgoCD, Kustomize, and Helm. That includes the GPU side of an AI cloud: driver images, GPU telemetry, and deploying LLM inference workloads. Most of the time goes to the operational parts: releases, HA router failures, capacity.
The layer my work sits on. I write eBPF probes in C, and use networking, systemd, and kernel tracing to debug and root-cause problems in complex systems, tracing an issue down through services, the network, and the host until the actual cause shows up.
A GPU metrics anomaly led us to a cryptominer hiding in plain sight on an exposed SGLang server, and a pickle CVE with no patch.
Why this site exists and what I plan to write about here.