Monday, December 23, 2013

Gen_server in Ocaml

Note, this post is written against the 2.0.1 version of gen_server

Erlang comes with a rich set of small concurrency primitives to make handling and manipulating state easier. The most generic of the frameworks is the gen_server which is also the most commonly used. A gen_server provides a way to control state over multiple requests. It serializes operations and handles both synchronous and asynchronous communication with clients. The strength of a gen_server is the ability to create multiple, lightweight, servers inside an application where each operation inside of it runs in serial but individually the gen_servers run concurrently.

While it is not possible to provide all of the Erlang semantics in Ocaml, we can create something roughly analogous. We can also get some properties that Erlang can not give us. In particular, the implementation of gen_server provided here:

  • Does not have the concept of a process or a process id. A gen_server is an abstract type that is parameterized by a message type.
  • Uses queues to communicate messages between clients and servers.
  • gen_servers are typesafe, only messages that they can handle can be sent to them.
  • You can only communicate with gen_servers in your own process, there is no concept of location ignorance.
  • Only provides an asynchronous communication function, called send that has pushback. That means a send will be evaluated when the gen_server accepts the message but will not wait for the gen_server to complete the processing of the message.
  • Has the concept of process linking, however it is not preemptive. When a gen_server stops, for any reason, any calls to send will return an error stating the gen_server has closed itself. This will not force the termination of any other gen_servers in Ocaml, but the termination can at least be detected.
  • Any thrown exceptions are handled by the gen_server framework and result in the gen_server being gracefully terminated.

Relative to Erlang the Ocaml version isn't very impressive, however it's still a useful technique for encapsulating state in a concurrent environment.

This implementation of gen_server is on top of Jane St's Async. What does it look like? The primary interface looks like this:

val start  :
    'i ->
    ('i, 's, 'm, 'ie, 'he) Server.t ->
    ('m t, [> 'ie init_ret ]) Deferred.Result.t

val stop   :
    'm t ->
    (unit, [> `Closed ]) Deferred.Result.t

val send   :
    'm t ->
    'm ->
    ('m, [> send_ret ]) Deferred.Result.t

The interface is only three functions: start, stop and send.

  • The start function is a bit harry looking but don't be put off by the server type parameterized on five type variables. The start function takes two parameters, the first is the initial parameters to pass to the gen_server, the second is the callbacks of the gen_server.
  • stop takes a gen_server and returns Ok () on success and Error `Closed if the gen_server is not running.
  • send takes a gen_server and a message. The message must be the same type the gen_server accepts. It returns Ok msg on success and Error `Closed if the gen_server is not running.

The most confusion part is probably the ('i, 's, 'm, 'ie, 'he) Server.t. This is the type that the implementer of the gen_server writes. It is three callbacks: init, handle_call and terminate. Let's breakdown the type variables:

  • 'i - This is the type of the variable that you pass to start and will be given to the init callback.
  • 's - This is the type of the state that the gen_server will encapsulate. This will be passed to handle_call and terminate. The handle_call callback will manipulate the state and return a new one.
  • 'm - This is the message type that the gen_server will accept.
  • 'ie - This is the type of error that the init callback can return.
  • 'he - This is the type of error that the handle_call callback can return.

While the server type looks complicated, as you can see each variable corresponds to all of the type information needed to understand a gen_server. So what does a server look like? While the types are big it's actually not too bad. Below is an example of a call to start. The full source code can be found here.

(* Package the callbacks *)
let callbacks =
  { Gen_server.Server.init; handle_call; terminate }

let start () =
  Gen_server.start () callbacks

And what do the callbacks look like? Below is a simplified version of what a set of callbacks could look like, with comments.

module Resp = Gen_server.Response

module Gs = Gen_server

(* Callbacks *)
let init self init =
  Deferred.return (Ok ())

let handle_call self state = function
  | Msg.Msg1 ->
    (* Success *)
    Deferred.return (Resp.Ok state)
  | Msg.Msg2 ->
    (* Error *)
    Deferred.return (Resp.Error (reason, state))
  | Msg.Msg3 ->
    (* Exceptions can be thrown too *)
    failwith "blowin' up"

(* Exceptions thrown from terminate are silently ignored *)
let terminate reason state =
   match reason with
     | Gs.Server.Normal ->
       (* Things exited normally *)
     | Gs.Server.Exn exn ->
       (* An exception was thrown *)
     | Gs.Server.Error err ->
       (* User returned an error *)

There isn't much more to it than that.

A functor implementation is also provided. I prefer the non-functor version, I think it's a bit less verbose and easier to work with, but some people like them.

How To Get It?

You can install gen_server through opam, simply: opam install gen_server

The source can be found here. Only the tags should be trusted as working.

There are a few examples here.


No comments:

Post a Comment