Custom multi-node layouts
Split nano Self-run across Linux servers in a layout other than the standard three-node HA design, with a reviewed inventory.
Most multi-server installs should use the standard three-node layout in High availability. For more ingest servers, more ClickHouse shards or replicas, or dedicated Keepers, see Scaling past three nodes first. Use this page for any other split, such as a separate server for PostgreSQL.
What you can change
nano runs as five roles. Each server in your inventory runs one or more of them:
| Role | What it runs | How many |
|---|---|---|
app | UI, API, search, detections, jobs | One or more. Two or more for failover (v2.0.7 and later) |
postgres | PostgreSQL (rules, alerts, cases, users) | One, or three for automatic failover (v2.0.7 and later). Two also work with three Keeper servers as the tie-breaker |
clickhouse | Event storage | One per shard and replica |
keeper | ClickHouse coordination | Three with replicas or several shards, otherwise one |
ingest | Vector and the ingest frontend | One or more |
Before you start
Contact support with your deployment ID and intended layout. nano reviews the inventory with you before the first install, because server identities and roles can't change afterwards without a data migration.
Size each server from Server sizing. Every server that runs clickhouse or keeper needs a supported CPU; see The installer refuses the CPU. Get the bundle as in Get the bundle.
Install
The steps are the same as for High availability, with your own inventory:
- Set the host firewall rules on every server.
- Unpack the bundle on every server and log in to the registry.
- Install the reviewed inventory to
/etc/nano/inventory.yamland stage the private certificate on every server. - Prepare on the
appserver and add your license values to.env. For a large install, also add your planned volume; see Prepare on nano-1. - Copy
.envto every other server. After any later installer run that changes.env(the first app server, or thepostgresserver that prints "The shared .env changed"), copy that server's.enveverywhere again; see Copy .env to the other servers. - Start the data roles first (
postgres,keeper,clickhouse), thenapp, theningest.
Each run names the server and its complete role list from the inventory:
NANO_HOST_NAME=<server name> ./install.sh --role <its roles> --inventory /etc/nano/inventory.yamlPrefer to have a coding agent run the install? See Deploy with your AI agent.
Vector native (port 6000)
From v2.0.30, every ingest server accepts Vector's native protocol on port 6000 with mutual TLS only. Issue, renew and revoke sender certificates on the first app server, then copy .env to every server and update every ingest server. A revoke isn't effective on an ingest server until you update it. See Vector native for the steps, the firewall rules and the trade-offs.
Reference
The bundle includes the full inventory reference at docs/getting-started/self-hosting-multi-node.md. It covers every inventory field, private TLS, custom ports and bind addresses, several ingest servers and two-shard layouts. The firewall rules are on this site, in Host firewall.
High availability
Run nano Self-run on three Linux servers so event storage, search and ingest survive the loss of one server.
Set up nano with your coding agent
Use Claude Code, Codex, Cursor or another coding agent to deploy nano with the nano-deploy skill, onboard sources with nano-connect and work through MCP.