Documentation
How Hoster is put together, and how to get a node running. Start with why the platform exists, then follow the quick start.
Introduction
- Why Hoster What the platform actually does differently: sub-second deploys, ZFS as a first-class citizen, microVMs, and a shared-nothing design where every host survives on its own.
- Quick start The whole path from bare hardware to a running VM, at a bird's-eye level — and an honest answer to when you need Control Center and when you need a cluster.
- Your first VM Pull a template, deploy from it, and get a shell inside the guest — with what each command actually does and why the deploy returns before you have finished reading it.
- MicroVMs Container workloads with a real kernel boundary: how microVMs differ from VMs, why HosterOS removed the setup they used to need, and how to run a compose stack inside one.
- Snapshots, replication and the scheduler A fresh node takes no snapshots at all. Here is the schedule to give it, what each job type is for, and how replication and disaster-recovery backups build on the same machinery.
- Clustering Three nodes, two commands and a token. What forming a cluster actually turns on, what quorum costs you, and how to look at one once it exists.
- Live migration Moving a running VM to another host without stopping it — what has to be true first, what the preflight checks, and the list of things that rule a guest out.
- Changing a guest after deploy Resizing, relabelling, sharing folders in, re-addressing and cloning — what you can change on a guest that already exists, what needs it stopped, and what you cannot change at all.
Control Center
- Deploying Control Center Control Center runs as a microVM with a compose stack inside it — the first real workload most people deploy. Standing it up, exposing it, and enrolling nodes into it.
- The interface A tour of what each Control Center page shows and when you would reach for it — and what the equivalent looks like on the CLI, because everything here is available both ways.
Networking
- Networks Bridges are what guests attach to, and everything else on this section assumes one exists. Creating external and internal networks, what reconciliation does, and the properties you can only set once.
- Firewall Declared rules, generated nftables. The four scopes, publishing a service with DNAT, and why a rule you add takes effect the moment you press enter.
- DNS and the WireGuard mesh How guests get names without anyone writing a record, how those names reach across hosts, and the two settings on a network that decide whether either happens.
Operations
- Getting into a guest The serial console, the generated credentials nobody told you about, and how SSH keys reach a guest — including the part of that which has no CLI.
- Restoring a guest Rolling back to a snapshot, cloning one into a new guest, and bringing a replicated copy to life on the host that holds it — plus which of those the CLI can actually do.
- Disaster recovery DR Manager — an off-fleet appliance holding encrypted restic repositories that stay closed until a host knocks. Deploying it, enabling DR on a guest, and restoring from a capability URL.
- The scheduler What is scheduled, what ran, what failed and why — plus running a job now without waiting for its cron. The operational half of snapshots, replication and DR.
- Monitoring and alerts Prometheus scrape config and 80-odd alert rules generated from the node itself, guests discovered automatically, and the one flag on a guest that decides whether it is watched at all.
- Troubleshooting a node Inspecting the host itself, regenerating the configuration files everything else depends on, clearing the stale locks and interfaces a crash leaves behind, and talking to QEMU directly.
- LeilFS Shared storage across the cluster, for the workloads ZFS replication does not suit — what Hoster provisions for it, what it monitors, and the honest boundary between the two.
- Building your own template A template is a ZFS dataset with a raw disk in it — nothing more. Making one by hand, what the image has to support, publishing a catalogue of your own, and keeping templates in step across a fleet.