A Pure Elixir Thrift Library

This package contains an implementation of Thrift for Elixir. It includes a Thrift IDL parser, a code generator, a binary framed client and a binary framed server.

The serialization and deserialization code that is generated by this project is highly optimized and is between 10 and 25 times faster_why? than the code generated by the Apache Erlang implementation.

Setup

Start by adding this package to your project as a dependency:

{:thrift, "~> 2.0"}

Or to track the GitHub master branch:

{:thrift, github: "pinterest/elixir-thrift"}

Mix

This package includes a Mix compiler task that can be used to automate Thrift code generation. Start by adding :thrift to your project's :compilers list. For example:

compilers: [:thrift | Mix.compilers]

It's important to add :thrift before the :elixir entry. The Thrift compiler will generate Elixir source files, which are in turn compiled by the :elixir compiler.

Next, configure the compiler using a new Keyword list under the top-level :thrift configuration key. The only necessary compiler option is :files, which defines the list of Thrift files that should be compiled.

By default, the generated source files will be written to the lib directory, but you can change that using the output_path option.

In this example, we gather all of the .thrift files under the thrift directory and write our output to the lib/thrift/ directory:

# example mix.exs
defmodule MyProject.Mixfile do
  # ...

  def project do
    [
      # other settings...
      thrift: [
        files: Path.wildcard("thrift/**/*.thrift"),
        output_path: "lib/generated"
      ]
    ]
  end
end

Working with Thrift

The examples below use the following thrift definition:

namespace elixir Thrift.Test
exception UserNotFound {
  1: string message
}

struct User {
  1: i64 id,
  2: string username,
  3: string first_name,
  4: string last_name
}

service UserService {
  bool ping(),
  User get_user_by_id(1: i64 user_id) throws (1: UserNotFound unf),
  bool deleteUser(1: i64 userId),
}

The generated code will be placed in the following modules:

Generated Code	Path	Output Module
User Struct	lib/thrift/test/user.ex	Thrift.Test.User
UserNotFound Exception	lib/thrift/test/user_not_found.ex	Thrift.Test.UserNotFound
User Binary Protocol	lib/thrift/test/user.ex	Thrift.Test.User.BinaryProtocol
UserNotFound Binary Protocol	lib/thrift/test/user_not_found.ex	Thrift.Test.UserNotFound.BinaryProtocol
UserService Framed Binary Client	lib/thrift/test/user_service.ex	Thrift.Test.UserService.Binary.Framed.Client
UserService Framed Binary Server	lib/thrift/test/user_service.ex	Thrift.Test.UserService.Binary.Framed.Server
UserService Handler Behviour (Used for writing servers)	lib/thrift/test_user_service/handler.ex	Thrift.Test.UserService.Handler

Namespaces

Thrift uses the namespace keyword to define a language's namespace. From the previous example:

namespace elixir Thrift.Test

Unfortunately, the Apache Thrift compiler will produce a warning on this line because it doesn't recognize elixir as a supported language. While that warning is benign, it can be annoying. For that reason, you can also specify your Elixir namespace as a "magic" namespace comment:

#@namespace elixir Thrift.Test

This alternate syntax is borrowed from Scrooge, which uses the same trick for defining scala namespaces.

Using the Client

The client includes a static module that does most of the work, and a generated interface module that performs some conversions and makes calling remote functions easier. You will not directly interface with the static module, but it is the one that's started when start_link is called.

The static client module uses James Fish's excellent connection behaviour.

For each function defined in the service, the generated module has four functions.

Function name	Description
get_user_by_id/3	Makes a request to the remote get_user_by_id RPC. You can optionally pass gen_tcp and GenServer options (such as a timeout) to the client as the final rpc_opts argument. Returns {:ok, response} or {:error, reason} tuples.
get_user_by_id!/3	Same as above, but raises an exception if something goes wrong. The type of exception can be one of the exceptions defined in the service or Thrift.TApplicationException.

Note: in the above example, the function deleteUser will be converted to delete_user to comply with Elixir's naming conventions.

To use the client, simply call start_link, supplying the host and port.

iex> alias Thrift.Test.UserService.Binary.Framed.Client
iex> {:ok, client} = Client.start_link("localhost", 2345, [])
iex> {:ok, user} = Client.get_user_by_id(client, 22451)
{:ok, %Thrift.Test.User{id: 22451, username: "stinky", first_name: "Stinky", last_name: "Stinkman"}}

The client supports the following options, which are passed in as the third argument to start_link:

Option name	Type	Description
:tcp_opts	keyword	A keyword list of tcp options (see below)
:ssl_opts	keyword	A list of options for SSL/TLS (see below)
:gen_server_opts	keyword	A keyword list of options for the gen server (see below)

TCP Opts

Name	Type	Description
:timeout	positive integer	The default timeout for reading from, writing to, and connecting to sockets.
:send_timeout	positive integer	The amount of time in milliseconds to wait before sending data fails.

SSL/TLS Opts

Name	Type	Description
:enabled	boolean	Whether to upgrade the connection to the SSL protocol.
:optional	boolean	Whether to accept both SSL and plain connections.
:configure	0-arity fun	A function to provide additional SSL options at run time.
ssloption	:ssl.ssloption	Other standard :ssl options.

GenServer Opts

Name	Type	Description
:timeout	A positive integer	The amount of time in milliseconds the Client's GenServer waits for a reply. After this, the GenServer will exit with {:error, :timeout}.

Example of using options

alias Thrift.Test.UserService.Binary.Framed.Client
{:ok, client} = Client.start_link("localhost", 2345,
                tcp_opts: [],
                ssl_opts: [enabled: true, cacertfile: "cacerts.pem", certfile: "cert.pem", keyfile: "key.pem"],
                gen_server_opts: [timeout: 10_000])

These options set the GenServer timeout to be ten seconds, which means the remote side can take its time to reply.

Using The Server

Creating a thrift server is slightly more involved than creating the client, because you need to create a module to handle the work. Fortunately, Elixir Thrift creates a Behaviour, complete with correct success typing, for this module. To implement this behaviour, use the @behaviour module attribute. The compiler will now inform you about any missed functions.

Here is an implementation for the server defined above:

defmodule UserServiceHandler do
  @behaviour Thrift.Test.UserService.Handler

  def ping, do: true

  def get_user_by_id(user_id) do
    case Backend.find_user_by_id(user_id) do
      {:ok, user} ->
        user
      {:error, _} ->
        raise Thrift.Test.UserNotFound.exception message: "could not find user with id #{user_id}"
    end
  end

  def delete_user(user_id) do
    Backend.delete_user(user_id) == :ok
  end
end

To start a server with UserServiceHandler as the callback module:

{:ok, server_pid} = Thrift.Test.UserService.Binary.Framed.Server.start_link(UserServiceHandler, 2345, [])

...and your server is up and running. RPC calls to the server are delegated to UserServiceHandler.

Like the client, the server takes several options. They are:

Name	Type	Description
worker_count	positive integer	The number of acceptor workers available to take requests
name	atom	(Optional) The name of the server. The server's pid becomes registered to this name. If not specified, the handler module's name is used.
max_restarts	non negative integer	The number of times to restart (see the next option)
max_seconds	non negative integer	The number of seconds. This is used by the supervisor to determine when to crash. If a server restarts max_restarts times in max_seconds then the supervisor crashes.

The server defines a Supervisor, which can be added to your application's supervision tree. When adding the server to your applications supervision tree, use the supervisor function rather than the worker function.

Using the binary protocol directly

Each thrift struct, union and exception also has a BinaryProtocol module generated for it. This module lets you serialize and deserialize its own type easily.

For example:

iex(1)> serialized = %User{username: "stinky" id: 1234, first_name: "Stinky", last_name: "Stinkman"}
|> User.BinaryProtocol.serialize
|> IO.iodata_to_binary
iex(2)> User.BinaryProtocol.deserialize(serialized)
{%User{username: "stinky" id: 1234, first_name: "Stinky", last_name: "Stinkman"}, ""}

The return value of the serialize function is an iodata. You can pass it through IO.iodata_to_binary to convert it to a binary. You also can write the iodata directly to a file or socket without converting it.

Other Features

Thrift IDL Parsing

This package also contains support for parsing Thrift IDL files. It is built on a low-level Erlang lexer and parser:

{:ok, tokens, _} = :thrift_lexer.string('enum Colors { RED, GREEN, BLUE }')
{:ok,
 [{:enum, 1}, {:ident, 1, 'Colors'}, {:symbol, 1, '{'}, {:ident, 1, 'RED'},
  {:symbol, 1, ','}, {:ident, 1, 'GREEN'}, {:symbol, 1, ','},
  {:ident, 1, 'BLUE'}, {:symbol, 1, '}'}], 1}

{:ok, schema} = :thrift_parser.parse(tokens)
{:ok,
 %Thrift.AST.Schema{constants: %{},
  enums: %{Colors: %Thrift.AST.TEnum{name: :Colors,
     values: [RED: 1, GREEN: 2, BLUE: 3]}}, exceptions: %{}, includes: [],
  namespaces: %{}, services: %{}, structs: %{}, thrift_namespace: nil,
  typedefs: %{}, unions: %{}}}

But also provides a high-level Elixir parsing interface:

Thrift.Parser.parse("enum Colors { RED, GREEN, BLUE }")
%Thrift.AST.Schema{constants: %{},
 enums: %{Colors: %Thrift.AST.TEnum{name: :Colors,
    values: [RED: 1, GREEN: 2, BLUE: 3]}}, exceptions: %{}, includes: [],
 namespaces: %{}, services: %{}, structs: %{}, thrift_namespace: nil,
 typedefs: %{}, unions: %{}}

You can use these features to support additional languages, protocols, and servers.

Quick and dirty FAQ

Why is it faster than the Apache implementation?

The Apache Thrift implementation uses C++ to write Erlang modules that describe Thrift data structures, then uses that description to turn your Thrift data into bytes. It consults this description every time Thrift data is serialized/deserialized. This on-the-fly conversion costs CPU time.

Additionally, the separation of concerns in Thrift prevent the Erlang VM from doing the best job that it can do during serialization.

On the other hand, this implementation uses Elixir to write Elixir code that's specific to your thrift data. This serialization logic is then compiled, and that compiled code is what converts your data to/from bytes. We've spent a lot of time making sure that the generated serialization code takes advantage of several of the optimizations that the Erlang VM provides.

What tradeoffs have you made to get this performance?

Thrift has the following concepts:

1. Protocols Define a conversion of data into bytes.
2. Transports Define how bytes move; across a network or in and out of a file.
3. Processors Encapsulate reading from streams and doing something with the data. Processors are generated by the Thrift compiler.

In Apache Thrift, Protocols and Transports can be mixed and matched. However, our implementation does the mixing and matching for you and generates a combination of (Protocol + Transport + Processor). This means that if you need to create a new Protocol or Transport, you need to integrate it into this project.

Presently, we implement:

• Binary Protocol, Framed Client
• Binary Protocol, Framed Server

We are more than willing to take PRs to add more. We intend to also add TMux and Finagle servers in the future.