Honeydew ("Honey, do!") is a pluggable job queue + worker pool for Elixir, powered by GenStage.

• Workers are permanent and hold immutable state (a network connection, for example).
• Workers pull jobs from the queue in a demand-driven fashion.
• Queues can exist locally, on another node in the cluster, or on a remote queue server (rabbitmq, etc...).
• Jobs are enqueued using async/3 and you can receive replies with yield/2, somewhat like Task.
• If a worker crashes while processing a job, the job is recovered and a "failure mode" (abandon, requeue, etc) is executed.
• Queues, workers, dispatch strategies and failure modes are all plugable with user modules.

Honeydew attempts to provide "at least once" job execution, it's possible that circumstances could conspire to execute a job, and prevent Honeydew from reporting that success back to the queue. I encourage you to write your jobs idepotently.

Honeydew isn't intended as a simple resource pool, the user's code isn't executed in the requesting process. Though you may use it as such, there are likely other alternatives that would fit your situation better.

• Check out the examples.
• Enqueue and receive responses with async/3 and yield/2.
• Suspend and resume with Honeydew.suspend/1 and Honeydew.resume/1
• List jobs with Honeydew.filter/2
• Queue status with Honeydew.status/1
• Cancel jobs with Honeydew.cancel/2

• • careful with this, it's slow, O(n)

In your mix.exs file:

defp deps do
  [{:honeydew, "~> 1.0.0-rc1"}]
end

You can run honeydew on a single node, or with components distributed over a cluster.

There's an uncaring firehose of data pointed at us, we need to store it all in our database, Riak. The requester isn't expecting a response, and we can't drop a write due to overloaded workers.

Let's create a worker module. Honeydew will call our worker's init/1 and keep the state from an {:ok, state} return.

Our workers are going to call functions from our module, the last argument will be the worker's state, riak in this case.

defmodule Riak do
  @moduledoc """
    This is an example Worker to interface with Riak.
    You'll need to add the erlang riak driver to your mix.exs:
    `{:riakc, ">= 2.4.1}`
  """

  def init([ip, port]) do
    :riakc_pb_socket.start_link(ip, port) # returns {:ok, riak}
  end

  def up?(riak) do
    :riakc_pb_socket.ping(riak) == :pong
  end

  def put(bucket, key, obj, content_type, riak) do
    :ok = :riakc_pb_socket.put(riak, :riakc_obj.new(bucket, key, obj, content_type))
  end

  def get(bucket, key, riak) do
    case :riakc_pb_socket.get(riak, bucket, key) do
      {:ok, obj} -> :riakc_obj.get_value(obj)
      {:error, :notfound} -> nil
      error -> error
    end
  end
end

Then we'll start the both a queue and workers in our supervision tree.

defmodule App do
  def start do
    children = [
      Honeydew.queue_spec(:riak),
      Honeydew.worker_spec(:riak, {Riak, ['127.0.0.1', 8087]}, num: 5, init_retry_secs: 10)
    ]

    Supervisor.start_link(children, strategy: :one_for_one)
  end
end

A task is simply a tuple with the name of a function and arguments, or a fn.

We'll add tasks to the queue using async/3 and wait for responses with yield/2. To tell Honeydew that we expect a response from the job, we'll specify reply: true, like so:

iex(1)> :up? |> Honeydew.async(:riak, reply: true) |> Honeydew.yield
{:ok, true}

iex(2)> {:put, ["bucket", "key", "value", "text/plain"]} |> Honeydew.async(:riak)
%Honeydew.Job{by: nil, failure_private: nil, from: nil, monitor: nil,
 private: -576460752303422557, queue: :riak, result: nil,
 task: {:put, ["bucket", "key", "value", "text/plain"]}}

iex(3)> {:get, ["bucket", "key"]} |> Honeydew.async(:riak, reply: true) |> Honeydew.yield
{:ok, "value"}

# our worker is holding a riak connection (a pid) as its state, let's ask that pid what its state is.
iex(4)> fn riak -> {riak, :sys.get_state(riak)} end |> Honeydew.async(:riak, reply: true) |> Honeydew.yield
{:ok,
 {#PID<0.256.0>,
  {:state, '127.0.0.1', 8087, false, false, #Port<0.8365>, false, :gen_tcp,
   :undefined, {[], []}, 1, [], :infinity, :undefined, :undefined, :undefined,
   :undefined, [], 100}}}

If you pass reply: true, and you never call yield/2 to read the result, your process' mailbox may fill up after multiple calls. Don't do that.

(Ignoring the response of the :put above is just used as an exmaple, you probably want to check the return value of a database insert unless you have good reason to ignore it)

The Honeydew.Job struct above is used to track the status of a job, you can send it to cancel/1, if you want to try to kill the job.

Say we've got some pretty heavy tasks that we want to distribute over a farm of background job processing nodes, they're too heavy to process on our client-facing nodes. In a distributed Erlang scenario, you have the option of distributing Honeydew's various components around different nodes in your cluster. Honeydew is basically a simple collection of queue processes and worker processes. Honeydew detects when nodes go up and down, and reconnects workers.

To start a global queue, pass a {:global, name} tuple when you start Honeydew's components

In this example, we'll use the Mnesa queue with stateless workers.

We'll start the queue on node queue@dax with:

defmodule QueueApp do
  def start do
    nodes = [node()]

    children = [
      Honeydew.queue_spec({:global, :my_queue}, queue: {Honeydew.Queue.Mnesia, [nodes, [disc_copies: nodes], []]})
    ]

    Supervisor.start_link(children, strategy: :one_for_one)
  end
end

iex(queue@dax)1> QueueApp.start
{:ok, #PID<0.209.0>}

And we'll run our workers on background@dax with:

defmodule HeavyTask do
  def work_really_hard(secs) do
    :timer.sleep(1_000 * secs)
    IO.puts "I worked really hard for #{secs} secs!"
  end
end

defmodule WorkerApp do
  def start do
    children = [
      Honeydew.worker_spec({:global, :my_queue}, HeavyTask, num: 10)
    ]

    Supervisor.start_link(children, strategy: :one_for_one)
  end
end

iex(background@dax)1> Node.ping :queue@dax
:pong

iex(background@dax)2> WorkerApp.start
{:ok, #PID<0.205.0>}

(note that in this case, our worker is stateless, so we left out init/1)

You can connect the nodes together at any point in the process, Honeydew will automatically detect where its components are running.

Then on any node in the cluster, we can enqueue a job:

iex(clientfacing@dax)1> Node.ping :queue@dax
:pong

iex(clientfacing@dax)2> {:work_really_hard, [5]} |> Honeydew.async({:global, :my_queue})
%Honeydew.Job{by: nil, failure_private: nil, from: nil, monitor: nil,
 private: {false, -576460752303423485}, queue: {:global, :my_queue},
 result: nil, task: {:work_really_hard, [5]}}

The job will run on the worker node, five seconds later it'll print I worked really hard for 5 secs!

There's one important caveat that you should note, Honeydew doesn't yet support OTP failover/takeover, so please be careful in production. I'll send you three emoji of your choice if you submit a PR. :)

You can suspend a queue (halt the distribution of new jobs to workers), by calling Honeydew.suspend(:my_queue), then resume with Honeydew.resume(:my_queue).

To cancel a job that hasn't yet run, use Honeydew.cancel/2. If the job was successfully cancelled before execution, :ok will be returned. If the job wasn't present in the queue, nil. If the job is currently being executed, {:error, :in_progress}.

There are various options you can pass to queue_spec/2 and worker_spec/3, see the Honeydew module.

When a worker crashes, a monitoring process runs the handle_failure/4 function from the selected module on the queue's node. Honeydew ships with two failure modes, at present:

• Abandon: Simply forgets about the job.
• Requeue: Removes the job from the original queue, and places it on another.

See Honeydew.queue_spec/2 to select a failure mode.

In general, a job goes through the following stages:

- The requesting process calls `async/2`, which packages the task tuple/fn up into a "job" then sends
  it to a member of the queue group.

- The queue process will enqueue the job, then, depending on its current "demand", take one of the
  following actions:
  ├─ If there is outstanding demand (> 0), the queue will dispatch the job immediately to a waiting
  |  worker via the selected dispatch strategy.
  └─ If there is no outstanding demand, the job will remain in the queue until demand arrives.

- Upon dispatch, the queue "reserves" the job (marks it as in-progress), then spawns a local Monitor
  process to watch the worker. Since GenStage doesn't guarantee delivery to a consumer, the monitor
  starts a timer after which the job will be returned to the queue.
  └─ When the worker receives the job, it informs the monitor associated with the job. The monitor
     then watches the worker in case the job crashes it.
     ├─ When the job succeeds:
     |  └─ If the job was enqueued with `reply: true`, the result is sent.
     |  └─ The worker sends an acknowledgement message to the monitor. The monitor sends an
     |     acknowledgement to the queue to remove the job.
     └─ If the worker crashes, the monitor executes the selected "Failure Mode" and terminates.

Queues are the most critical location of state in Honeydew, a job will not be removed from the queue unless it has either been successfully executed, or been dealt with by the configured failure mode.

Honeydew includes a few basic queue modules:

• A simple FIFO queue implemented with the :queue and Map modules, this is the default.
• An Mnesia queue, configurable in all the ways mnesia is, for example:
  • Run with replication (with queues running on multiple nodes)
  • Persist jobs to disk (dets)
  • Follow various safety modes ("access contexts").

If you want to implement your own queue, check out the included queues as a guide. Try to keep in mind where exactly your queue state lives, is your queue process(es) where jobs live, or is it a completely stateless connector for some external broker? Or a hybrid? I'm excited to see what you come up with, please open a PR! <3

By default, Honeydew uses GenStage's DemandDispatcher, but you can use any module that implements the GenStage.Dispatcher behaviour. Simply pass the module as the :dispatcher option to Honeydew.queue_spec/2. I haven't experimented with any other dispatchers aside from the default, if you do, please let me know how it goes.

Worker state is immutable, the only way to change it is to cause the worker to crash and let the supervisor restart it.

Your worker module's init/1 function must return {:ok, state}. If anything else is returned or the function raises an error, the worker will die and restart after a given time interval (by default, five seconds).

• the example RabbitMQ connector is broken
• failover/takeover for global queues
• statistics?
• yield_many/2 support?
• benchmark mnesia queue's dual filter implementations, discard one?
• using a global queue, control which node executes a job on-the-fly with a ParitionDispatcher?
• more tests

Thanks to @marcelog, for his failing worker restart strategy.