How Much Should Live Together? Learning to Isolate Services the Hard Way

Short summary

A personal account of evolving a homelab from a single large VM running Docker containers to purpose-built VMs with Proxmox HA, and finally to one LXC container per service. Each stage taught distinct lessons about blast radius, failover, and resource sharing. The GPU passthrough limitation of VMs ultimately drove the move to LXC containers for finer-grained isolation and cross-node failover.

• •Evolves from one big VM to purpose-built VMs to per-service LXC containers
• •Each architecture stage has distinct tradeoffs in isolation, failover, and management complexity
• •GPU sharing limitations of VM passthrough motivated the final migration to LXC

Generated with AI, which can make mistakes.