Names and addresses

Reach every service by a name that stays the same, such as postgres.larder.doze:5432, and open your apps in a browser with no port.

A name per service

With names on, every service in a stack gets the name <name>.<stack>.doze:

stack = "larder"

defaults {
  names = true
}

module postgres "postgres" {
  version = 18
}

Larder’s Postgres is postgres.larder.doze:5432. (Larder is the made-up shop the docs use as an example.) stack is the stack’s name; without it, the folder’s name is used. A block name with characters a hostname cannot have is lowercased, with each of them turned into -: app_dev becomes app-dev.

doze init writes names = true into the defaults block of a new doze.hcl. Without it, names are off.

Its own address, on its usual port

Each name points at an address of its own, so every service keeps its engine’s usual port: two stacks can each have a Postgres on 5432, and so can one stack with two. The same name gets the same address every time.

doze url postgres
# postgres://postgres@postgres.larder.doze:5432/postgres?sslmode=disable

References give the same names, so the apps in the stack use them too: module.postgres.address is postgres.larder.doze:5432 (References and env).

An app is also reachable at 127.0.0.1:<port> when that port is free on this machine, for tools that expect localhost.

Apps without a port: the front door

An app or service that declares a port is also served at http://<name>.<stack>.doze, with no port, through one listener on port 80 that every stack on the machine shares:

http://storefront.larder.doze      → storefront, on its port 5173
http://gateway.larder.doze         → gateway, on its port 8000

A module with a web interface of its own is served the same way: Larder’s AWS is http://aws.larder.doze, with its console at /_console. routes = false in a module block keeps it on its port only.

.doze names are plain HTTP. For https, on a hostname you choose, give the app domains: see Domains and HTTPS.

Set up the machine once

Names need a resolver for .doze and, on macOS, a small network daemon that gives services their addresses. Setting them up asks for your password once. The setup is shared with doze AWS and doze Kafka: whichever installs it first installs it for all.

In the app

Settings › Machine when everything is in place.

1. Open Settings › Machine, from the menu bar, Home’s Machine line, or ⌘,.
2. If a part is not in place, click Repair. It runs the same setup as doze doctor --fix, behind the system’s password prompt.

In a terminal

doze doctor            # what is in place, and what is not
doze doctor --fix      # set it up (one sudo); the same as doze dns-setup
doze dns-setup --check # report only, no sudo

On macOS this installs /etc/resolver/doze and the dev.doze.netd network daemon. On Linux it routes .doze through systemd-resolved or dnsmasq and installs doze-netd.service; see Linux. doze dns-setup --uninstall removes it all.

Until the machine is set up, services listen on 127.0.0.1 and two of them cannot share a port. doze says so when they try:

db and legacy both use port 5432, but this machine is not set up to give each service its own address — run `doze doctor --fix` once (needs sudo), or give them distinct ports

Names off

defaults {
  names = false
}

With names off, everything is 127.0.0.1:<port>, there is no front door, and every service needs a port of its own; two on one port is an error that names both. References and doze url give 127.0.0.1 addresses instead.

Larder makes names a variable, so a machine without the setup can still run it:

variable "names" {
  type    = bool
  default = true
}

defaults {
  names = var.names
}

doze up --var names=false

DOZE_ZONE=off turns names off for every doze program at once, whatever the stack says: useful in CI or a container that has no use for names.