Practical Proxmox and ZFS operating lessons.

Mirrors solve a disk failure, not every failure

A mirror provides device redundancy. It does not protect against accidental deletion, application corruption, operator mistakes, host failure or a site-wide event. Design backups and recovery around the failures you actually need to survive.

Reserve capacity before the pool is full

Capacity planning should use real usable storage and leave operational headroom. Full pools make maintenance, snapshots, migrations and recovery harder, so stop new provisioning before the safety margin is consumed.

ARC belongs in the memory budget

ZFS uses memory for ARC caching. On a virtualization host, account for ARC alongside guest allocations and the hypervisor's own needs instead of treating all unallocated memory as guest capacity.

Scrub and TRIM are maintenance, not backups

Regular scrubs can surface latent read or checksum problems while redundancy is available. TRIM can help supported solid-state storage maintain allocation efficiency. Neither creates an independent recovery copy.

ZFS scrub planning → · ZFS snapshot planning →

Test through the guest boundary

Infrastructure validation should include disposable guest workloads and the same provisioning, storage and network paths customers will use. Static configuration review can miss problems that only appear after a guest is created and exercised.

Benchmark method and evidence →

High availability is an end-to-end property

Multiple hosts alone do not create useful high availability. Placement, quorum, storage, networking, service discovery and recovery behavior all need to be designed and tested together.