Peerage helps your nodes find each other.
It supports DNS-based discovery, which means you can use it out of the box with Kubernetes (and probably also Weave, discoverd, Swarm, or other anything else with dns-based service discovery).
It also supports UDP-based discovery, so that nodes on the same network (like docker containers on the same host) can find each other.
It's also easy to extend: adding a new Provider can be as simple as writing a single function.
Add peerage
to your list of dependencies in mix.exs
,
and start its application:
def application do
[applications: [:peerage]]
end
def deps do
[{:peerage, "~> 1.0.2"}]
end
Note that the latest hex version may be higher than what is listed here. You can
find the latest version on hex. You should match
the version or alternatively you can use a looser version constraint like "~> 1.0"
.
Peerage will attempt to Node.connect/1
to node names returned
by a provider that you choose:
config :peerage, via: Peerage.Via.Dns
There are several providers available, each with examples further down the page.
-
Peerage.Via.Self
is a 'hello, world' provider that only connects to itself. -
Peerage.Via.List
uses a hardcoded list of node names. It's good for development, or when you know production node names ahead of time. See example below. -
Peerage.Via.Dns
gets IP addresses from DNS. Works for production config on Kubernetes; probably works on Weave, Flynn, and Swarm. You can test it locally with one node by telling it your app's dns name islocalhost
(this is also the default). See example below. -
Peerage.Via.Udp
uses peer-to-peer UDP multicast. Docker containers on a single host can use this out of the box (example below). Could be useful in a variety of situations where nodes are on the same network/overlay network, providing the network doesn't restrict multicast. -
Custom providers are simple (but there's more detail below). TL;DR is:
defmodule MyWayToFindHomies do def poll, do: [ :"node@somewhere" ] end
config :peerage, via: MyWayToFindHomies
Usually, I use the List or Udp providers in dev config, so I can easily test multiple nodes on my machine, and the Dns provider for my production releases, which run in places (like Kubernetes) where nodes can be found by DNS discovery.
Optional config for all apps
config :abc, via: SomeProvider,
serves: false, # true if provider runs supervised
interval: 15, # default 10
log_results: true # show connection results in debug mode?
See specific providers for their additional config.
Supporting release config is easy now.
The :peerage
app uses
DeferredConfig,
at app startup, so all config for the app (including
any custom providers) automatically supports run-time
config via {:system, "MY_ENV_VAR"}
tuples (with optional
defaults and conversion functions), as well as arbitrary
functions via MFA tuples. See DeferredConfig for more
details, but the upshot is that everything you want
to be runtime configurable should be.
config :peerage, via: Peerage.Via.List, node_list: [
:"myapp1@127.0.0.1",
:"myapp2@127.0.0.1"
]
$ iex --name myapp1@127.0.0.1 -S mix # one shell
$ iex --name myapp2@127.0.0.1 -S mix # other shell
I wrap this with a script for launching dev shell and prod release shells, so that I just call bin/dev 1
or bin/prod 1
to test locally in mix, or a release with env vars set to work on my machine.
DNS-based discovery is where it's at.
Minimal dns example, one node: after installing, add the following to your config:
config :peerage, via: Peerage.Via.Dns,
dns_name: "localhost",
app_name: "myapp"
And then run iex like this:
$ iex --name myapp@127.0.0.1 -S mix
It'll use dns to get the IP addresses for 'localhost', and try to connect to them. In this case, there'll only be one result, but it's working; it got that ip by looking up localhost!
Longer example:
Say you have an app called :myapp
. You're deploying
it to Kubernetes as a headless service with name: myapp
and clusterIP: None
.
Or to Convox with the Weave AMI, with an app name of myapp
.
You launch your application so that its node will
be called myapp@${NODE_IP}
, with the ip of the container being provided by an overlay network.
- If you're using releases, you'll get that IP in an
env variable (or just set it yourself on container
startup), splice that address into
rel/vm.args
with a line like-name ${NODE_NAME}@${NODE_IP}
, and run your release withREPLACE_OS_VARS=true
. (See exrm or distillery docs if necessary).
Your config/prod.exs might look like this:
config :peerage, via: Peerage.Via.Dns,
dns_name: "myapp", # or k8s FQDN: "myapp.ns.svc.clust.local"
app_name: "myapp"
Now your app will look up the name "myapp" from within
your container. In Kubernetes, hostname -i myapp
will
(as of 1.3 or somesuch) return the list of IP addresses
of each of the pods that make up the service; the same is
true on a system with Weave for container networking.
- If the network name needs to be specified more (perhaps
the dns discovery supports encoding version
numbers, regions, etc into the name, and you
want to only connect to the same version),
change
dns_name
.
In Kubernetes, you can test all of this with minikube.
config :peerage, via: Peerage.Via.Udp, serves: true,
port: 45900
Will both broadcast and receive node names via UDP on port 45900,
and keep track of ones it's seen so that Peerage connect to them.
It's a GenServer, so we let Peerage know it needs to be run and
supervised with serves: true
.
For simple cases, where you're polling a source of truth (some API, etc): just provide a poll
method that returns a list of node-name atoms:
defmodule MyWayToFindHomies do
def poll, do: [ :"name@host" ] # poll source of truth
end
config :peerage, via: MyWayToFindHomies
For more complex, peer-to-peer cases -- say, UDP multicast, where each node is broadcasting itself on the network -- you'll probably want to use a GenServer, listening for incoming broadcast messages, and building up a list of 'seen' nodes over time.
If your Provider is an OTP process, just add serves: true
to your config:
config :peerage, via: MyWayToFindHomies, serves: true
For a full-fledged example, see Peerage.Via.Udp
.