Coming Soon

StackSmith's Agent Fleet

A heartbeat-based control plane without service mesh

2026-05-30 ~11 min
Coming soon. This article is on the writing queue. The outline below sketches what the finished piece will cover. In the meantime, the Celenite Stack reference and the source on github.com/davestj are the best places to dig into the underlying systems.

How 5 production OVH hosts heartbeat back to StackSmith every 60 seconds with host stats, container state, and Kubernetes telemetry — and why this design choice eliminates the need for a service mesh, a sidecar agent, and a separate metrics pipeline.

Live repository

davestj/Mcaster1StackSmith private
Mcaster1StackSmith is a devops tool
0 0 PHP Apr 27, 2026

What the article will cover

The pull vs push question

Why heartbeats from agents to mothership beat traditional pull-monitoring for a heterogeneous fleet.

Agent anatomy

A single Go binary that runs as a systemd service on every host. What it collects, when, and how it serializes.

The 60-second heartbeat

Why one minute is the sweet spot, what happens if you miss one, and the watchdog timer that escalates stale hosts.

Authentication

Per-agent bootstrap tokens, mTLS to the mothership, and how a stolen agent token gets revoked.

Inside Kubernetes

The companion stacksmith-k8-agent runs as a DaemonSet, reporting from inside the cluster — and how its data merges with the host-level agent stream.

Why no service mesh

A walk through what Linkerd / Istio would have added and why we don't need it — the agent IS the data plane.

stacksmithcontrol-planeobservabilitykubernetes
All Engineering articles