package capnp-rpc-unix

  1. Overview
  2. Docs
Cap'n Proto is a capability-based RPC system with bindings for many languages


Dune Dependency





OCaml Cap'n Proto RPC library

Copyright 2017 Docker, Inc. See for details.



Cap'n Proto is a capability-based RPC system with bindings for many languages. Some key features:

  • APIs are defined using a schema file, which is compiled to create bindings for different languages automatically.

  • Schemas can be upgraded in many ways without breaking backwards-compatibility.

  • Messages are built up and read in-place, making it very fast.

  • Messages can contain capability references, allowing the sender to share access to a service. Access control is handled automatically.

  • Messages can be pipelined. For example, you can ask one service where another one is, and then immediately start calling methods on it. The requests will be sent to the first service, which will either handle them (in the common case where the second service is in the same place) or forward them until you can establish a direct connection.

  • Messages are delivered in E-Order, which means that messages sent over a reference will arrive in the order in which they were sent, even if the path they take through the network gets optimised at some point.

This library should be used with the capnp-ocaml schema compiler, which generates bindings from schema files.


RPC Level 2 is complete, with encryption and authentication using TLS and support for persistence.

The library has been only lightly used in real systems, but has unit tests and AFL fuzz tests that cover most of the core logic.

The default network provided supports TCP and Unix-domain sockets, both with or without TLS. For two-party networking, you can provide any bi-directional byte stream (satisfying the Mirage flow signature) to the library to create a connection. You can also define your own network types.

Level 3 support is not implemented yet, so if host Alice has connections to hosts Bob and Carol and passes an object hosted at Bob to Carol, the resulting messages between Carol and Bob will be routed via Alice. Until that is implemented, Carol can ask Bob for a persistent reference (sturdy ref) and then connect directly to that.


To install, you will need a platform with the capnproto package available (e.g. Debian >= 9). Then:

opam depext -i capnp-rpc-unix conf-capnproto

Structure of the library

The code is split into several packages:

  • capnp-rpc contains the logic of the Cap'n Proto RPC Protocol, but does not depend on any particular serialisation. The tests in the test directory test the logic using a simple representation where messages are OCaml data-structures (defined in capnp-rpc/

  • capnp-rpc-lwt instantiates the capnp-rpc functor using the Cap'n Proto serialisation for messages and Lwt for concurrency. The tests in test-lwt test this by sending Cap'n Proto messages over a Unix-domain socket.

  • capnp-rpc-unix adds helper functions for parsing command-line arguments and setting up connections over Unix sockets.

  • capnp-rpc-mirage is an alternative to -unix that works with Mirage unikernels.

Users of the library will normally want to use capnp-rpc-lwt and, in most cases, capnp-rpc-unix.


This tutorial creates a simple echo service and then extends it. It shows how to use most of the features of the library, including defining services, using encryption and authentication over network links, and saving service state to disk.

A basic echo service

Start by writing a Cap'n Proto schema file. For example, here is a very simple echo service:

interface Echo {
  ping @0 (msg :Text) -> (reply :Text);

This defines the Echo interface as having a single method called ping which takes a struct containing a text field called msg and returns a struct containing another text field called reply.

Save this as echo_api.capnp and compile it using capnp:

$ capnp compile echo_api.capnp -o ocaml
echo_api.capnp:1:1: error: File does not declare an ID.  I've generated one for you.
Add this line to your file:

Every interface needs a globally unique ID. If you don't have one, capnp will pick one for you, as shown above. Add the line to the start of the file to get:


interface Echo {
  ping @0 (msg :Text) -> (reply :Text);

Now it can be compiled:

$ capnp compile echo_api.capnp -o ocaml
echo_api.capnp --> echo_api.mli

The next step is to implement a client and server (in a new file) using the generated Echo_api OCaml module.

For the server, you should inherit from the generated Api.Service.Echo.service class:

module Api = Echo_api.MakeRPC(Capnp_rpc_lwt)

open Lwt.Infix
open Capnp_rpc_lwt

let local =
  let module Echo = Api.Service.Echo in
  Echo.local @@ object
    inherit Echo.service

    method ping_impl params release_param_caps =
      let open Echo.Ping in
      let msg = Params.msg_get params in
      release_param_caps ();
      let response, results = Service.Response.create Results.init_pointer in
      Results.reply_set results ("echo:" ^ msg);
      Service.return response

The first line (module Api) instantiates the generated code to use this library's RPC implementation.

The service object must provide one OCaml method for each method defined in the schema file, with _impl on the end of each one.

There's a bit of ugly boilerplate here, but it's quite simple:

  • The Api.Service.Echo.Ping module defines the server-side API for the ping method.

  • Ping.Params is a reader for the parameters.

  • Ping.Results is a builder for the results.

  • msg is the string value of the msg field.

  • release_param_caps releases any capabilities passed in the parameters. In this case there aren't any, but remember that a client using some future version of this protocol might pass some optional capabilities, and so you should always free them anyway.

  • Service.Response.create Results.init_pointer creates a new response message, using Ping.Results.init_pointer to initialise the payload contents.

  • response is the complete message to be sent back, and results is the data part of it.

  • Service.return returns the results immediately (like Lwt.return).

The client implementation is similar, but uses Api.Client instead of Api.Service. Here, we have a builder for the parameters and a reader for the results. Api.Client.Echo.Ping.method_id is a globally unique identifier for the ping method.

module Echo = Api.Client.Echo

let ping t msg =
  let open Echo.Ping in
  let request, params = Capability.Request.create Params.init_pointer in
  Params.msg_set params msg;
  Capability.call_for_value_exn t method_id request >|= Results.reply_get

Capability.call_for_value_exn sends the request message to the service and waits for the response to arrive. If the response is an error, it raises an exception. Results.reply_get extracts the reply field of the result.

We don't need to release the capabilities of the results, as call_for_value_exn does that automatically. We'll see how to handle capabilities later.

With the boilerplate out of the way, we can now write a to test it:

open Lwt.Infix

let () =
  Logs.set_level (Some Logs.Warning);
  Logs.set_reporter (Logs_fmt.reporter ())

let () = begin
    let service = Echo.local in service "foo" >>= fun reply -> "Got reply %S@." reply;

If you're building with jbuilder, here's a suitable jbuild file:

(jbuild_version 1)

(executable (
  (name main)
  (libraries (capnp-rpc-lwt capnp-rpc-unix logs.fmt))
  (flags (:standard -w -53-55))

 ((targets ( echo_api.mli))
  (deps (echo_api.capnp))
  (action (run capnpc -o ocaml ${<}))))

The service is now usable:

$ opam install capnp-rpc-unix conf-capnproto
$ jbuilder build --dev main.exe
$ ./_build/default/main.exe
Got reply "echo:foo"

This isn't very exciting, so let's add some capabilities to the protocol...

Passing capabilities


interface Callback {
  log @0 (msg :Text) -> ();

interface Echo {
  ping      @0 (msg :Text) -> (reply :Text);
  heartbeat @1 (msg :Text, callback :Callback) -> ();

This version of the protocol adds a heartbeat method. Instead of returning the text directly, it will send it to a callback at regular intervals.

Run capnp compile again to update the generated files (if you're using the jbuild file then this will happen automatically and you should delete the generated and echo_api.mli files from the source directory instead).

The new heartbeat_impl method looks like this:

    method heartbeat_impl params release_params =
      let open Echo.Heartbeat in
      let msg = Params.msg_get params in
      let callback = Params.callback_get params in
      release_params ();
      match callback with
      | None -> "No callback parameter!"
      | Some callback ->
        Service.return_lwt @@ fun () ->
        notify callback msg

Note that all parameters in Cap'n Proto are optional, so we have to check for callback not being set (data parameters such as msg get a default value from the schema, which is "" for strings if not set explicitly).

Service.return_lwt fn runs fn () and replies to the heartbeat call when it finishes. Here, the whole of the rest of the method is the argument to return_lwt, which is a common pattern.

notify callback msg just sends a few messages to callback in a loop, and then releases it:

let notify callback msg =
  let rec loop = function
    | 0 ->
      Capability.dec_ref callback;
      Lwt.return @@ Ok (Service.Response.create_empty ())
    | i ->
      Callback.log callback msg >>= function
      | Error _ as e -> Capability.dec_ref callback; Lwt.return e
      | Ok () ->
        Lwt_unix.sleep 1.0 >>= fun () ->
        loop (i - 1)
  loop 3

Exercise: implement the client-side Callback.log function (hint: it's very similar to ping, but use Capability.call_for_unit because we don't care about the value of the result and we want to handle errors manually)

To write the client for Echo.heartbeat, we take a user-provided callback object and put it into the request:

let heartbeat t msg callback =
  let open Echo.Heartbeat in
  let request, params = Capability.Request.create Params.init_pointer in
  Params.msg_set params msg;
  Params.callback_set params (Some callback);
  Capability.call_for_unit_exn t method_id request

Capability.call_for_unit_exn is a convenience wrapper around Callback.call_for_value_exn that discards the result.

In, we can now wrap a regular OCaml function as the callback:

open Lwt.Infix
open Capnp_rpc_lwt

let () =
  Logs.set_level (Some Logs.Warning);
  Logs.set_reporter (Logs_fmt.reporter ())

let callback_fn msg = "Callback got %S@." msg

let run_client service =
  let callback = Echo.Callback.local callback_fn in
  Echo.heartbeat service "foo" callback >>= fun () ->
  Capability.dec_ref callback;

let () = begin
    let service = Echo.local in
    run_client service

Step 1: The client creates the callback:

Step 2: The client calls the heartbeat method, passing the callback as an argument:

Step 3: The service receives the callback and calls the log method on it:

Exercise: implement Callback.local fn (hint: it's similar to the original ping service, but pass the message to fn and return with Service.return_empty ())

And testing it should give (three times, at one second intervals):

$ ./main
Callback got "foo"
Callback got "foo"
Callback got "foo"

Note that the client gives the echo service permission to call its callback service by sending a message containing the callback to the service. No other access control updates are needed.

Note also a design choice here in the API: we could have made the Echo.heartbeat function take an OCaml callback and wrap it, but instead we chose to take a service and make do the wrapping. The advantage to doing it this way is that may one day want to pass a remote callback, as we'll see later.

This still isn't very exciting, because we just stored an OCaml object pointer in a message and then pulled it out again. However, we can use the same code with the echo client and service in separate processes, communicating over the network...


Let's put a network connection between the client and the server. Here's the new (the top half is the same as before):

open Lwt.Infix
open Capnp_rpc_lwt

let () =
  Logs.set_level (Some Logs.Warning);
  Logs.set_reporter (Logs_fmt.reporter ())

let callback_fn msg = "Callback got %S@." msg

let run_client service =
  let callback = Echo.Callback.local callback_fn in
  Echo.heartbeat service "foo" callback >>= fun () ->
  Capability.dec_ref callback;

let secret_key = `Ephemeral
let listen_address = `TCP ("", 7000)

let start_server () =
  let config = Capnp_rpc_unix.Vat_config.create ~secret_key listen_address in
  let service_id = Capnp_rpc_unix.Vat_config.derived_id config "main" in
  let restore = Restorer.single service_id Echo.local in
  Capnp_rpc_unix.serve config ~restore >|= fun vat ->
  Capnp_rpc_unix.Vat.sturdy_uri vat service_id

let () = begin
    start_server () >>= fun uri -> "Connecting to echo service at: %a@." Uri.pp_hum uri;
    let client_vat = Capnp_rpc_unix.client_only_vat () in
    let sr = Capnp_rpc_unix.Vat.import_exn client_vat uri in
    Sturdy_ref.connect_exn sr >>= fun proxy_to_service ->
    run_client proxy_to_service

Running this will give something like:

$ ./_build/default/main.exe
Connecting to echo service at: capnp://sha-256:3Tj5y5Q2qpqN3Sbh0GRPxgORZw98_NtrU2nLI0-Tn6g@
Callback got "foo"
Callback got "foo"
Callback got "foo"

Once the server vat is running, we get a "sturdy ref" for the echo service, which is displayed as a "capnp://" URL. The URL contains several pieces of information:

  • The sha-256:3Tj5y5Q2qpqN3Sbh0GRPxgORZw98_NtrU2nLI0-Tn6g part is the fingerprint of the server's public key. When the client connects, it uses this to verify that it is connected to the right server (not an imposter). Therefore, a Cap'n Proto vat does not need to be certified by a CA (and cannot be compromised by a rogue CA).

  • is the address to which clients will try to connect to reach the server vat.

  • eBIndzZyoVDxaJdZ8uh_xBx5V1lfXWTJCDX-qEkgNZ4 is the (base64-encoded) service ID. This is a secret that both identifies the service to use within the vat, and also grants access to it.

The server side

The let secret_key = `Ephemeral line causes a new server key to be generated each time the program runs, so if you run it again you'll see a different capnp URL. For a real system you'll want to save the key so that the server's identity doesn't change when it is restarted. You can use let secret_key = `File "secret-key.pem" for that. Then the file secret-key.pem will be created automatically the first time you start the service, and reused on future runs.

It is also possible to disable the use of encryption using Vat_config.create ~serve_tls:false .... That might be useful if you need to interoperate with a client that doesn't support TLS.

listen_address tells the server where to listen for incoming connections. You can use `Unix path for a Unix-domain socket at path, or `TCP (host, port) to accept connections over TCP.

For TCP, you might want to listen on one address but advertise a different one, e.g.

let listen_address = `TCP ("", 7000)	(* Listen on all interfaces *)
let public_address = `TCP ("", 7000)	(* Tell clients to connect here *)

let start_server () =
  let config = Capnp_rpc_unix.Vat_config.create ~secret_key ~public_address listen_address in

In start_server:

  • let service_id = Capnp_rpc_unix.Vat_config.derived_id config "main" creates the secret ID that grants access to the service. derived_id generates the ID deterministically from the secret key and the name. This means that the ID will be stable as long as the server's key doesn't change. The name used ("main" here) isn't important - it just needs to be unique.

  • let restore = Restorer.single service_id Echo.local configures a simple "restorer" that answers requests for service_id with our Echo.local service.

  • Capnp_rpc_unix.serve config ~restore creates the service vat using the previous configuration items and starts it listening for incoming connections.

  • Capnp_rpc_unix.Vat.sturdy_uri vat service_id returns a "capnp://" URI for the given service within the vat.

The client side

After starting the server and getting the sturdy URI, we create a client vat and connect to the sturdy ref. The result, proxy_to_service, is a proxy to the remote service via the network that can be used in exactly the same way as the direct reference we used before.

Separate processes

The example above runs the client and server in a single process. To run them in separate processes we just need to add some command-line parsing to let the user choose whether to run as a server or as a client:

open Lwt.Infix
open Capnp_rpc_lwt

let () =
  Logs.set_level (Some Logs.Warning);
  Logs.set_reporter (Logs_fmt.reporter ())

let callback_fn msg = "Callback got %S@." msg

let run_client service =
  let callback = Echo.Callback.local callback_fn in
  Echo.heartbeat service "foo" callback >>= fun () ->
  Capability.dec_ref callback;

let serve config = begin
    let service_id = Capnp_rpc_unix.Vat_config.derived_id config "main" in
    let restore = Restorer.single service_id Echo.local in
    Capnp_rpc_unix.serve config ~restore >>= fun vat ->
    let uri = Capnp_rpc_unix.Vat.sturdy_uri vat service_id in "echo service running at: %a@." Uri.pp_hum uri;
    fst @@ Lwt.wait ()  (* Wait forever *)

let connect uri = begin "Connecting to echo service at: %a@." Uri.pp_hum uri;
    let client_vat = Capnp_rpc_unix.client_only_vat () in
    let sr = Capnp_rpc_unix.Vat.import_exn client_vat uri in
    Sturdy_ref.connect_exn sr >>= fun proxy_to_service ->
    run_client proxy_to_service

open Cmdliner

let connect_addr =
  let i = [] ~docv:"ADDR" ~doc:"Address of server (capnp://...)" in
  Arg.(required @@ pos 0 (some Capnp_rpc_unix.sturdy_uri) None i)

let serve_cmd =
  Term.(const serve $ Capnp_rpc_unix.Vat_config.cmd),
  let doc = "run the server" in "serve" ~doc

let connect_cmd =
  let doc = "run the client" in
  Term.(const connect $ connect_addr), "connect" ~doc

let default_cmd =
  let doc = "capnp-rpc demo app" in
  Term.(ret (const (`Help (`Pager, None)))), "demo" ~version:"v0.1" ~doc

let cmds = [serve_cmd; connect_cmd]

let () =
  Term.eval_choice default_cmd cmds |> Term.exit

The cmdliner term Capnp_rpc_unix.Vat_config.cmd provides an easy way to get a suitable Vat_config based on command-line arguments provided by the user.

To test, start the server running:

$ ./_build/default/main.exe serve --secret-key-file=key.pem --listen-address tcp:
echo service running at: capnp://sha-256:_FNMlR9cf1maixDAM6Y1pwwZ-aikqa_DP8P7RCVr1k4@

Then run the client, using the URL printed by the server:

$ ./_build/default/main.exe connect capnp://sha-256:_FNMlR9cf1maixDAM6Y1pwwZ-aikqa_DP8P7RCVr1k4@
Connecting to echo service at: capnp://sha-256:_FNMlR9cf1maixDAM6Y1pwwZ-aikqa_DP8P7RCVr1k4@
Callback got "foo"
Callback got "foo"
Callback got "foo"


Let's say the server also offers a logging service, which the client can get from the main echo service:

interface Echo {
  ping      @0 (msg :Text) -> (reply :Text);
  heartbeat @1 (msg :Text, callback :Callback) -> ();
  getLogger @2 () -> (callback :Callback);

The implementation of the new method in the service is simple - we export the callback in the response in the same way we previously exported the client's callback in the request:

    method get_logger_impl _ release_params =
      let open Echo.GetLogger in
      release_params ();
      let response, results = Service.Response.create Results.init_pointer in
      Results.callback_set results (Some service_logger);
      Service.return response

Exercise: create a service_logger that prints out whatever it gets (hint: use Callback.local)

The client side is more interesting:

let get_logger t =
  let open Echo.GetLogger in
  let request = Capability.Request.create_no_args () in
  Capability.call_for_caps t method_id request Results.callback_get_pipelined

We could have used call_and_wait here (which is similar to call_for_value but doesn't automatically discard any capabilities in the result). However, that would mean waiting for the response to be sent back to us over the network before we could use it. Instead, we use callback_get_pipelined to get a promise for the capability from the promise of the getLogger call's result.

Note: the last argument to call_for_caps is a function for extracting the capabilities from the promised result. In the common case where you just want one and it's in the root result struct, you can just pass the accessor directly, as shown. Doing it this way allows call_for_caps to release any unused capabilities in the result automatically for us.

We can test it as follows:

let run_client service =
  let logger = Echo.get_logger service in
  Echo.Callback.log logger "Message from client" >|= function
  | Ok () -> ()
  | Error err -> Fmt.epr "Server's logger failed: %a" Capnp_rpc.Error.pp err

This should print (in the server's output) something like:

Service logger: Message from client

In this case, we didn't wait for the getLogger call to return before using the logger. The RPC library pipelined the log call directly to the promised logger from its previous question. On the wire, the messages are sent together, and look like:

  1. What is your logger?

  2. Please call the object returned in answer to my previous question (1).

Now, let's say we'd like the server to send heartbeats to itself:

let run_client service =
  let callback = Echo.get_logger service in
  Echo.heartbeat service "foo" callback >>= fun () ->
  Capability.dec_ref callback;

Here, we ask the server for its logger and then (without waiting for the reply), tell it to send heartbeat messages to the promised logger (you should see the messages appear in the server process's output).

Previously, when we exported our local callback object, it arrived at the service as a proxy that sent messages back to the client over the network. But when we send the (promise of the) server's own logger back to it, the RPC system detects this and "shortens" the path; the capability reference that the heartbeat handler gets is a direct reference to its own logger, which it can call without using the network.

These optimisations are very important because they allow us to build APIs like this with small functions that can be composed easily. Without pipelining, we would be tempted to clutter the protocol with specialised methods like heartbeatToYourself to avoid the extra round-trips most RPC protocols would otherwise require.

Hosting multiple sturdy refs

The Restorer.single restorer used above is useful for vats hosting a single sturdy ref. However, you may want to host multiple sturdy refs, perhaps to provide separate "admin" and "user" capabilities to different clients, or to allow services to be created and persisted as sturdy refs dynamically. To do this, we can use Restorer.Table. For example, we can extend our example to provide sturdy refs for both the main echo service and the logger service:

let serve config =
  let make_sturdy = Capnp_rpc_unix.Vat_config.sturdy_uri config in
  let services = Restorer.Table.create make_sturdy in
  let echo_id = Capnp_rpc_unix.Vat_config.derived_id config "main" in
  let logger_id = Capnp_rpc_unix.Vat_config.derived_id config "logger" in
  Restorer.Table.add services echo_id Echo.local;
  Restorer.Table.add services logger_id (Echo.Callback.local callback_fn);
  let restore = Restorer.of_table services in begin
    Capnp_rpc_unix.serve config ~restore >>= fun _vat -> "echo service: %a@." Uri.pp_hum (make_sturdy echo_id); "logger service: %a@." Uri.pp_hum (make_sturdy logger_id);
    fst @@ Lwt.wait ()  (* Wait forever *)

Exercise: add a log command and use it to test the log service URI printed by the above code. Hint: parse the command-line with:

let msg =
  let i = [] ~docv:"MESSAGE" ~doc:"Message to log" in
  Arg.(required @@ pos 1 (some string) None i)

let log_cmd =
  let doc = "log a message" in
  Term.(const log_client $ connect_addr $ msg), "log" ~doc

let cmds = [serve_cmd; connect_cmd; log_cmd]

Implementing the persistence API

Cap'n Proto defines a standard Persistence API which services can implement to allow clients to request their sturdy ref.

On the client side, calling Persistence.save_exn cap will send a request to cap asking for its sturdy ref. For example, after connecting to the main echo service and getting a live capability to the logger, the client can request a sturdy ref like this:

let run_client service =
  let callback = Echo.get_logger service in
  Persistence.save_exn callback >>= fun uri -> "The server's logger's URI is %a.@." Uri.pp_hum uri;

If successful, the client can use this sturdy ref to connect directly to the logger in future.

If you try the above, it will fail with Unimplemented: Unknown interface 17004856819305483596UL. To add support on the server side, we must tell each logger instance what its public address is and have it implement the persistence interface. The simplest way to do this is to wrap the Callback.local call with Persistence.with_sturdy_ref:

module Callback = struct
  let local sr fn =
    let module Callback = Api.Service.Callback in
    Persistence.with_sturdy_ref sr Callback.local @@ object

Then pass the sr argument when creating the logger (you'll need to make it an argument to Echo.local too):

  let logger_id = Capnp_rpc_unix.Vat_config.derived_id config "logger" in
  let logger_sr = Restorer.Table.sturdy_ref services logger_id in
  let service_logger = Echo.Callback.local logger_sr @@ "Service log: %S@." in
  Restorer.Table.add services echo_id (Echo.local ~service_logger);
  Restorer.Table.add services logger_id service_logger;

After restarting the server, the client should now display the logger's URI, which you can then use with demo.exe log URI MSG.

Creating and persisting sturdy refs dynamically

So far, we have been providing a static set of sturdy refs. We can also generate new sturdy refs dynamically and return them to clients. We'll normally want to record each new export in some kind of persistent storage so that the sturdy refs still work after restarting the server.

It is possible to use Table.add for this. However, that requires all capabilities to be loaded into the table at start-up, which may be a performance problem.

Instead, we can create the table using Table.of_loader. When the user asks for a sturdy ref that is not in the table, it calls our load function to load the capability dynamically. The function can use a database or the filesystem to look up the resource. You can still use Table.add to register additional services, as before.

Let's extend the ping service to support multiple callbacks with different labels. Then we can give each user a private sturdy ref to their own logger callback. Here's the interface for a DB module that loads and saves loggers:

module DB : sig
  include Restorer.LOADER

  val create : make_sturdy:(Restorer.Id.t -> Uri.t) -> string -> t
  (** [create ~make_sturdy dir] is a database that persists services in [dir]. *)

  val save_new : t -> label:string -> Restorer.Id.t
  (** [save_new t ~label] adds a new logger with label [label] to the store and
      returns its newly-generated ID. *)

There is a Capnp_rpc_unix.File_store module that can persist Cap'n Proto structs to disk. First, define a suitable Cap'n Proto data structure to hold the information we need to store. In this case, it's just the label:

struct SavedLogger {
  label @0 :Text;

struct SavedService {
  logger @0 :SavedLogger;

Using Cap'n Proto for this makes it easy to add extra fields or service types later if needed (SavedService.logger can be upgraded to a union if we decide to add more service types later). We can use this with File_store to implement DB:

  module Store = Capnp_rpc_unix.File_store

  type t = {
    store : Api.Reader.SavedService.struct_t Store.t;
    make_sturdy : Restorer.Id.t -> Uri.t;

  let hash _ = `SHA256

  let make_sturdy t = t.make_sturdy

  let load t sr digest =
    match Store.load ~digest with
    | None -> Lwt.return Restorer.unknown_service_id
    | Some saved_service ->
      let logger = Api.Reader.SavedService.logger_get saved_service in
      let label = Api.Reader.SavedLogger.label_get logger in
      let callback msg = "%s: %S@." label msg
      let sr = Sturdy_ref.cast sr in
      Lwt.return @@ Restorer.grant @@ Callback.local sr callback

  let save t ~digest label =
    let open Api.Builder in
    let service = SavedService.init_root () in
    let logger = SavedService.logger_init service in
    SavedLogger.label_set logger label; ~digest @@ SavedService.to_reader service

  let save_new t ~label =
    let id = Restorer.Id.generate () in
    let digest = Restorer.Id.digest (hash t) id in
    save t ~digest label;

  let create ~make_sturdy dir =
    let store = Store.create dir in
    {store; make_sturdy}

Note: to avoid possible timing attacks, the load function is called with the digest of the service ID rather than with the ID itself. This means that even if the load function takes a different amount of time to respond depending on how much of a valid ID the client guessed, the client will only learn the digest (which is of no use to them), not the ID. The file store uses the digest as the filename, which avoids needing to check the ID the client gives for special characters, and also means that someone getting a copy of the store (e.g. an old backup) doesn't get the IDs (which would allow them to access the real service).

The main serve function then uses Echo.DB to create the table:

let serve config =
  (* Create the on-disk store *)
  let make_sturdy = Capnp_rpc_unix.Vat_config.sturdy_uri config in
  let db = Echo.DB.create ~make_sturdy "/tmp/store" in
  (* Create the restorer *)
  let services = Restorer.Table.of_loader (module Echo.DB) db in
  let restore = Restorer.of_table services in
  (* Add the fixed services *)
  let echo_id = Capnp_rpc_unix.Vat_config.derived_id config "main" in
  let logger_id = Capnp_rpc_unix.Vat_config.derived_id config "logger" in
  let logger_sr = Restorer.Table.sturdy_ref services logger_id in
  let service_logger = Echo.service_logger logger_sr in
  Restorer.Table.add services echo_id (Echo.local ~service_logger);
  Restorer.Table.add services logger_id service_logger;
  (* Run the server *) begin

Add a method to let clients create new loggers:

interface Echo {
  ping         @0 (msg :Text) -> (reply :Text);
  heartbeat    @1 (msg :Text, callback :Callback) -> ();
  getLogger    @2 () -> (callback :Callback);
  createLogger @3 (label: Text) -> (callback :Callback);

The server implementation of the method gets the label from the parameters, adds a saved logger to the database, and then "restores" the saved service to a live instance and returns it:

    method create_logger_impl params release_params =
      let open Echo.CreateLogger in
      let label = Params.label_get params in
      release_params ();
      let id = DB.save_new db ~label in
      Service.return_lwt @@ fun () ->
      Restorer.restore restore id >|= function
      | Error e -> Error (`Exception e)
      | Ok logger ->
        let response, results = Service.Response.create Results.init_pointer in
        Results.callback_set results (Some logger);
        Capability.dec_ref logger;
        Ok response

You'll need to pass db and restore to Echo.local too to make this work.

The client can call createLogger and then use to get the sturdy ref for it:

let run_client service =
  let my_logger = Echo.create_logger service "Alice" in
  let uri = Persistence.save_exn my_logger in
  Echo.Callback.log_exn my_logger "Pipelined call to logger!" >>= fun () ->
  uri >>= fun uri ->    (* Wait for results from [save] *) "The new logger's URI is %a.@." Uri.pp_hum uri;

Notice the pipelining here. The client sends three messages in quick succession: create the logger, get its sturdy ref, and log a message to it. The client receives the sturdy ref and prints it in a total of one network round-trip.

Exercise: Implement Echo.create_logger. You should find that the new loggers still work after the server is restarted.


Congratulations! You now know how to:

  • Define Cap'n Proto services and clients, independently of any networking.

  • Pass capability references in method arguments and results.

  • Stretch capabilities over a network link, with encryption, authentication and access control.

  • Configure a vat using command-line arguments.

  • Pipeline messages to avoid network round-trips.

  • Persist services to disk and restore them later.

Further reading


How can I return multiple results?

Every Cap'n Proto method returns a struct, although the examples in this README only use a single field. You can return multiple fields by defining a method as e.g. -> (foo :Foo, bar :Bar). For more complex types, it may be more convenient to define the structure elsewhere and then refer to it as -> MyResults.

Can I create multiple instances of an interface dynamically?

Yes. e.g. in the example above we can use Callback.local fn many times to create multiple loggers. Just remember to call Capability.dec_ref on them when you're finished so that they can be released promptly (but if the TCP connection is closed, all references on it will be freed anyway).

Can I get debug output?

First, always make sure logging is enabled so you can at least see warnings. The examples in this document enable some basic logging.

If you turn up the log level to Info (or even Debug), you'll see lots of information about what is going on. Turning on colour in the logs will help too - see test-bin/ for an example.

Many references will be displayed with their reference count (e.g. as rc=3). You can also print a capability for debugging with Capability.pp.

CapTP.dump will dump out the state of an entire connection, which will show you what services you’re currently importing and exporting over the connection.

If you override your service’s pp method, you can include extra information in the output too. Use Capnp_rpc.Debug.OID to generate and display a unique object identifier for logging.

How can I debug reference counting problems?

If a capability gets GC'd with a non-zero ref-count, you should get a warning. For testing, you can use Gc.full_major to force a check.

If you try to use something after releasing it, you'll get an error.

But the simple rule is: any time you create a local capability or extract a capability from a message, you must eventually call Capability.dec_ref on it.

How can I import a sturdy ref that I need to start my vat?

Let's say you have a capnp service that internally requires the use of another capnp service:

Here, creating the Frontend service requires a sturdy ref for the Backend service. But this sturdy ref must be imported into the frontend vat. Creating the frontend vat requires passing a restorer, which needs Frontend!

The solution here is to construct Frontend with a promise for the sturdy ref, e.g.

let run_frontend backend_uri =
  let backend_promise, resolver = Lwt.wait () in
  let frontend = Frontend.make backend_promise in
  let restore = Restorer.single id frontend in
  Capnp_rpc_unix.serve config ~restore >|= fun vat ->
  Lwt.wakeup resolver (Capnp_rpc_unix.Vat.import_exn vat backend_uri)

How can I release other resources when my service is released?

Override the release method. It gets called when there are no more references to your service.

Is there an interactive version I can use for debugging?

The Python bindings provide a good interactive environment. For example, start the test service above and leave it running:

$ ./_build/default/main.exe
Connecting to server at capnp://insecure@

Note that you must run without encryption for this, and use a non-secret ID:

let config = Capnp_rpc_unix.Vat_config.create ~serve_tls:false ~secret_key listen_address in
let service_id = Restorer.Id.public "" in

Run python from the directory containing your echo_api.capnp file and do:

import capnp
import echo_api_capnp
client = capnp.TwoPartyClient('')
echo = client.bootstrap().cast_as(echo_api_capnp.Echo)

Importing a module named foo_capnp will load the Cap'n Proto schema file foo.capnp.

To call the ping method:"From Python").wait()
<$Results reader (reply = "echo:From Python")>

To call the heartbeat method, with results going to the server's own logger:

echo.heartbeat("From Python", echo.getLogger().callback).wait()
Service logger: "From Python"

To call the heartbeat method, with results going to a Python callback:

class CallbackImpl(echo_api_capnp.Callback.Server):
    def log(self, msg, _context): print("Python callback got %s" % msg)

echo.heartbeat("From Python", CallbackImpl())
Python callback got From Python
Python callback got From Python
Python callback got From Python

Note that calling wait_forever prevents further use of the session, however.

How can I use this with Mirage?

capnp uses the uint library, which has C stubs and does not work on most Mirage backends. As a quick hack, you can do:

opam pin add uint ''

This allows it to compile and run as a unikernel, by defining type Uint64.t = Int64.t, etc. However, this changes the behaviour of unsigned integers, so you should be careful with it. In particular, OCaml's built-in polymorphic comparison operators (>, etc) may give incorrect results. Ideally, someone would add proper Mirage support to the uint library. explains why OCaml doesn't have unsigned integer support.

Here is a suitable

open Mirage

let main =
    ~packages:[package "capnp-rpc-mirage"; package "mirage-dns"]
    ~deps:[abstract nocrypto]
    "Unikernel.Make" (time @-> stackv4 @-> job)

let stack = generic_stackv4 default_network

let () =
  register "test" [main $ default_time $ stack]

This should work as the

open Lwt.Infix

open Capnp_rpc_lwt

module Make (Time : Mirage_time_lwt.S) (Stack : Mirage_stack_lwt.V4) = struct
  module Dns = Dns_resolver_mirage.Make (Time) (Stack)
  module Mirage_capnp = Capnp_rpc_mirage.Make (Stack) (Dns)

  let secret_key = `Ephemeral

  let listen_address = `TCP 7000
  let public_address = `TCP ("localhost", 7000)

  let start () stack () =
    let dns = Dns.create stack in
    let net = ~dns stack in
    let config = Mirage_capnp.Vat_config.create ~secret_key ~public_address listen_address in
    let service_id = Mirage_capnp.Vat_config.derived_id config "main" in
    let restore = Restorer.single service_id Echo.local in
    Mirage_capnp.serve net config ~restore >>= fun vat ->
    let uri = Mirage_capnp.Vat.sturdy_uri vat service_id in (fun f -> f "Main service: %a" Uri.pp_hum uri);
    Lwt.wait () |> fst


Conceptual model

An RPC system contains multiple communicating actors (just ordinary OCaml objects). An actor can hold capabilities to other objects. A capability here is just a regular OCaml object pointer.

Essentially, each object provides a call method, which takes:

  • some pure-data message content (typically an array of bytes created by the Cap'n Proto serialisation), and

  • an array of pointers to other objects (providing the same API).

The data part of the message says which method to invoke and provides the arguments. Whenever an argument needs to refer to another object, it gives the index of a pointer in the pointers array.

For example, a call to a method that transfers data between two stores might look something like this:

- Content:
  - InterfaceID: xxx
  - MethodID: yyy
  - Params:
    - Source: 0
    - Target: 1
- Pointers:
  - <source>
  - <target>

A call also takes a resolver, which it will call with the answer when it's ready. The answer will also contain data and pointer parts.

On top of this basic model the Cap'n Proto schema compiler (capnp-ocaml) generates a typed API, so that application code can only generate or attempt to consume messages that match the schema. Application code does not need to worry about interface or method IDs, for example.

This might seem like a rather clumsy system, but it has the advantage that such messages can be sent not just within a process, like regular OCaml method calls, but also over the network to remote objects.

The network is made up of communicating "vats" of objects. You can think of a Unix process as a single vat. The vats are peers - there is no difference between a "client" and a "server" at the protocol level. However, some vats may not be listening for incoming network connections, and you might like to think of such vats as clients.

When a connection is established between two vats, each can choose to ask the other for access to some service. Services are usually identified by a long random secret (a "Swiss number") so that only authorised clients can get access to them. The capability they get back is a proxy object that acts like a local service but forwards all calls over the network. When a message is sent that contains pointers, the RPC system holds onto the pointers and makes each object available over that network connection. Each vat only needs to expose at most a single bootstrap object, since the bootstrap object can provide methods to get access to any other required services.

All shared objects are scoped to the network connection, and will be released if the connection is closed for any reason.

The RPC system is smart enough that if you export a local object to a remote service and it later exports the same object back to you, it will switch to sending directly to the local service (once any pipelined messages in flight have been delivered).

You can also export an object that you received from a third-party, and the receiver will be able to use it. Ideally, the receiver should be able to establish a direct connection to the third-party, but this isn't yet implemented and instead the RPC system will forward messages and responses in this case.


To build:

git clone
cd capnp-rpc
opam pin add -nyk git capnp-rpc .
opam pin add -nyk git capnp-rpc-lwt .
opam pin add -nyk git capnp-rpc-unix .
opam depext capnp-rpc-lwt alcotest
opam install --deps-only -t capnp-rpc-unix
make test

If you have trouble building, you can build it with Docker from a known-good state using docker build ..


Running make test will run through the tests in test-lwt/, which run some in-process examples.

The calculator example can also be run across two Unix processes.

Start the server with:

$ ./_build/default/test-bin/calc.bc serve \
    --listen-address unix:/tmp/calc.socket \
Waiting for incoming connections at:

Note that key.pem does not need to exist. A new key will be generated and saved if the file does not yet exist.

In another terminal, run the client and connect to the address displayed by the server:

./_build/default/test-bin/calc.bc connect capnp://sha-256:LPp-7l74zqvGcRgcP8b7-kdSpwwzxlA555lYC8W8prc@/tmp/calc.socket

You can also use --disable-tls if you prefer to run without encryption (e.g. for interoperability with another Cap'n Proto implementation that doesn't support TLS). In that case, the client URL would be capnp://insecure@/tmp/calc.socket.


Running make fuzz will run the AFL fuzz tester. You will need to use a version of the OCaml compiler with AFL support (e.g. opam sw 4.04.0+afl).

The fuzzing code is in the fuzz directory. The tests set up some vats in a single process and then have them perform operations based on input from the fuzzer. At each step it selects one vat and performs a random (fuzzer-chosen) operation out of:

  1. Request a bootstrap capability from a random peer.

  2. Handle one message on an incoming queue.

  3. Call a random capability, passing randomly-selected capabilities as arguments.

  4. Finish a random question.

  5. Release a random capability.

  6. Add a capability to a new local service.

  7. Answer a random question, passing random-selected capability as the response.

The content of each call is a (mutable) record with counters for messages sent and received on the capability reference used. This is used to check that messages arrive in the expected order.

The tests also set up a shadow reference graph, which is like the regular object capability reference graph except that references between vats are just regular OCaml pointers (this is only possible because all the tests run in a single process, of course). When a message arrives, the tests compare the service that the CapTP network handler selected as the target with the expected target in this simpler shadow network. This should ensure that messages always arrive at the correct target.

In future, more properties should be tested (e.g. forked references, that messages always eventually arrive when there are no cycles, etc). We should also test with some malicious vats (that don't follow the protocol correctly).


Innovation. Community. Security.