Turma

Very simple and naive application for issuing commands to several machines in parallel and retrieving results.

The main purpose of this is to replace Ansible for some tasks that require more complex operations, and also to overcome Ansible’s limitations of seemingly running only a few jobs in parallel (which I known can be increased) and then ofter getting the SSH connection in a stuck state or losing the connection. If one task takes too long or gets stuck, the next hosts and the whole playbook crawls to a halt.

This project is not for use in production, as it’s completely unsafe.

Setup

Start a Legionarius node on each machine you want to control, and one Decurio to use as the command central. The configuration lives at etc/runtime.exs inside the release root directory.

On each Legionarius, define its id as the endpoint you’ll use to connect to it:

config Turma.Legionarius,
  bind: {"10.10.1.41", 19876},
  id: "server-0.some.cluster:19876"

On Decurio, define the inventory and then start it up:

config Turma.Decurio,
  inventory: %{
    "tag1" => [
            "server-0.some.cluster:19876",
            "server-1.some.cluster:19876",
            # ...
          ],
    "tag2" => [
            "server-1.some.cluster:19876",
            "other-server-0.some.cluster:19876",
          ],
    # ...
  },
  name: "my-decurio"

Examples

Run something on all nodes:

alias Turma.Decurio
Decurio.run(fn -> node() end)

Run something on hosts with a specific tag:

alias Turma.Decurio
Decurio.run("tag", fn -> node() end)

Run something on hosts whose endpoint match a regex:

alias Turma.Decurio
Decurio.run(~r/^some-host-[0-9]+/, fn -> node() end)

Run something on hosts filtering directly on the inventory:

alias Turma.Decurio
filter = fn inventory ->
  inventory
  |> Map.fetch!("tag")
  |> Enum.take_every(3)
end
Decurio.run(filter, fn -> node() end)

Get the results of a job (may change as each node returns results):

alias Turma.Decurio
{:ok, id} = Decurio.run(fn -> node() end)
Decurio.get(id)

Get all jobs:

alias Turma.Decurio
Decurio.get_all()

Block until a job is done:

alias Turma.Decurio
{:ok, id} = Decurio.run(fn -> node() end)
res = receive do
  {:job_finished, ^id} -> Decurio.get(id)
after
  10_000 -> :timeout
end