rest-core

by Lin Jen-Shin (godfat)

Lin Jen-Shin had given a talk about rest-core on RubyConf Taiwan 2011. The slide is in English, but the talk is in Mandarin. There's another talk about The Promise of rest-core

DESCRIPTION:

Various rest-builder middleware for building REST clients.

Checkout rest-more for pre-built clients.

FEATURES:

Modular interface for REST clients similar to WSGI/Rack for servers via rest-builder.
Concurrent requests with synchronous or asynchronous interfaces with threads via promise_pool.

WHY?

Build your own API clients for less dependencies, less codes, less memory, less conflicts, and run faster.

REQUIREMENTS:

Mandatory:

Tested with MRI (official CRuby) and JRuby.
gem rest-builder
gem promise_pool
gem timers
gem httpclient
gem mime-types

Optional:

gem json or yajl-ruby, or multi_json (if JsonResponse or JsonRequest middleware is used)

INSTALLATION:

gem install rest-core

Or if you want development version, put this in Gemfile:

gem 'rest-core', :git => 'git://github.com/godfat/rest-core.git',
                 :submodules => true

If you just want to use Facebook or Twitter clients, please take a look at [rest-more][] which has a lot of clients built with rest-core.

Build Your Own Clients:

You can use RestCore::Builder to build your own dedicated clients. Note that RC is an alias of RestCore

require 'rest-core'
YourClient = RC::Builder.client do
  use RC::DefaultSite , 'https://api.github.com/users/'
  use RC::JsonResponse, true
  use RC::CommonLogger, method(:puts)
  use RC::Cache       , nil, 3600 # :expires_in if cache store supports
end

Basic Usage:

And use it with per-instance basis (clients could have different configuration, e.g. different cache time or timeout time):

client = YourClient.new(:cache => {})
client.get('godfat') # cache miss
client.get('godfat') # cache hit

client.site = 'https://github.com/api/v2/json/user/show/'
client.get('godfat') # cache miss
client.get('godfat') # cache hit

Concurrent Requests with Futures:

You can also make concurrent requests easily: (see "Advanced Concurrent HTTP Requests -- Embrace the Future" for detail)

a = [client.get('godfat'), client.get('cardinalblue')]
puts "It's not blocking... but doing concurrent requests underneath"
p a.map{ |r| r['name'] } # here we want the values, so it blocks here
puts "DONE"

Exception Handling for Futures:

Note that since the API call would only block whenever you're looking at the response, it won't raise any exception at the time the API was called. So if you want to block and handle the exception at the time API was called, you would do something like this:

begin
  response = client.get('bad-user').tap{} # .tap{} is the point
  do_the_work(response)
rescue => e
  puts "Got an exception: #{e}"
end

The trick here is forcing the future immediately give you the exact response, so that rest-core could see the response and raise the exception. You can call whatever methods on the future to force this behaviour, but since tap{} is a method from Kernel (which is included in Object), it's always available and would return the original value, so it is the easiest method to be remembered and used.

If you know the response must be a string, then you can also use to_s. Like this:

begin
  response = client.get('bad-user').to_s
  do_the_work(response)
rescue => e
  puts "Got an exception: #{e}"
end

Or you can also do this:

begin
  response = client.get('bad-user')
  response.class # simply force it to load
  do_the_work(response)
rescue => e
  puts "Got an exception: #{e}"
end

The point is simply making a method call to force it load, whatever method should work.

Concurrent Requests with Callbacks:

On the other hand, callback mode also available:

client.get('godfat'){ |v| p v }
puts "It's not blocking... but doing concurrent requests underneath"
client.wait # we block here to wait for the request done
puts "DONE"

Exception Handling for Callbacks:

What about exception handling in callback mode? You know that we cannot raise any exception in the case of using a callback. So rest-core would pass the exception object into your callback. You can handle the exception like this:

client.get('bad-user') do |response|
  if response.kind_of?(Exception)
    puts "Got an exception: #{response}"
  else
    do_the_work(response)
  end
end
puts "It's not blocking... but doing concurrent requests underneath"
client.wait # we block here to wait for the request done
puts "DONE"

Thread Pool / Connection Pool

Underneath, rest-core would spawn a thread for each request, freeing you from blocking. However, occasionally we would not want this behaviour, giving that we might have limited resource and cannot maximize performance.

For example, maybe we could not afford so many threads running concurrently, or the target server cannot accept so many concurrent connections. In those cases, we would want to have limited concurrent threads or connections.

YourClient.pool_size = 10
YourClient.pool_idle_time = 60

This could set the thread pool size to 10, having a maximum of 10 threads running together, growing from requests. Each threads idled more than 60 seconds would be shut down automatically.

Note that pool_size should at least be larger than 4, or it might be very likely to have deadlock if you're using nested callbacks and having a large number of concurrent calls.

Also, setting pool_size to -1 would mean we want to make blocking requests, without spawning any threads. This might be useful for debugging.

Gracefully shutdown

To shutdown gracefully, consider shutdown the thread pool (if we're using it), and wait for all requests for a given client. For example, suppose we're using RC::Universal, we'll do this when we're shutting down:

RC::Universal.shutdown

We could put them in at_exit callback like this:

at_exit do
  RC::Universal.shutdown
end

If you're using unicorn, you probably want to put that in the config.

Random Asynchronous Tasks

Occasionally we might want to do some asynchronous tasks which could take the advantage of the concurrency facilities inside rest-core, for example, using wait and shutdown. You could do this with defer for a particular client. Still take RC::Universal as an example:

RC::Universal.defer do
  sleep(1)
  puts "Slow task done"
end

RC::Universal.wait

Persistent connections (keep-alive connections)

Since we're using httpclient by default now, we would reuse connections, making it much faster for hitting the same host repeatedly.

Streaming Requests

Suppose we want to POST a file, instead of trying to read all the contents in memory and send them, we could stream it from the file system directly.

client.post('path', File.open('README.md'))

Basically, payloads could be any IO object. Check out RC::Payload for more information.

Streaming Responses

This one is much harder then streaming requests, since all built-in middleware actually assume the responses should be blocking and buffered. Say, some JSON parser could not really parse from streams.

We solve this issue similarly to the way Rack solves it. That is, we hijack the socket. This would be how we're doing:

sock = client.get('path', {}, RC::HIJACK => true)
p sock.read(10)
p sock.read(10)
p sock.read(10)

Of course, if we don't want to block in order to get the socket, we could always use the callback form:

client.get('path', {}, RC::HIJACK => true) do |sock|
  p sock.read(10)
  p sock.read(10)
  p sock.read(10)
end

Note that since the socket would be put inside RC::RESPONSE_SOCKET instead of RC::RESPONSE_BODY, not all middleware would handle the socket. In the case of hijacking, RC::RESPONSE_BODY would always be mapped to an empty string, as it does not make sense to store the response in this case.

SSE (Server-Sent Events)

Not only JavaScript could receive server-sent events, any languages could. Doing so would establish a keep-alive connection to the server, and receive data periodically. We'll take Firebase as an example:

If you are using Firebase, please consider rest-firebase instead.

require 'rest-core'

# Streaming over 'users/tom.json'
cl = RC::Universal.new(:site => 'https://SampleChat.firebaseIO-demo.com/')
es = cl.event_source('users/tom.json', {}, # this is query, none here
                     :headers => {'Accept' => 'text/event-stream'})

@reconnect = true

es.onopen   { |sock| p sock } # Called when connected
es.onmessage{ |event, data, sock| p event, data } # Called for each message
es.onerror  { |error, sock| p error } # Called whenever there's an error
# Extra: If we return true in onreconnect callback, it would automatically
#        reconnect the node for us if disconnected.
es.onreconnect{ |error, sock| p error; @reconnect }

# Start making the request
es.start

# Try to close the connection and see it reconnects automatically
es.close

# Update users/tom.json
p cl.put('users/tom.json', RC::Json.encode(:some => 'data'))
p cl.post('users/tom.json', RC::Json.encode(:some => 'other'))
p cl.get('users/tom.json')
p cl.delete('users/tom.json')

# Need to tell onreconnect stops reconnecting, or even if we close
# the connection manually, it would still try to reconnect again.
@reconnect = false

# Close the connection to gracefully shut it down.
es.close

Those callbacks would be called in a separate background thread, so we don't have to worry about blocking it. If we want to wait for the connection to be closed, we could call wait:

es.wait # This would block until the connection is closed

More Control with `request_full`:

You can also use request_full to retrieve everything including response status, response headers, and also other rest-core options. But since using this interface is like using Rack directly, you have to build the env manually. To help you build the env manually, everything has a default, including the path.

client.request_full({})[RC::RESPONSE_BODY] # {"message"=>"Not Found"}
# This would print something like this:
# RestCore: spent 1.135713 Requested GET https://api.github.com/users/

client.request_full(RC::REQUEST_PATH => 'godfat')[RC::RESPONSE_STATUS]
client.request_full(RC::REQUEST_PATH => 'godfat')[RC::RESPONSE_HEADERS]
# Headers are normalized with all upper cases and
# dashes are replaced by underscores.

# To make POST (or any other request methods) request:
client.request_full(RC::REQUEST_PATH   => 'godfat',
                    RC::REQUEST_METHOD => :post)[RC::RESPONSE_STATUS] # 404

Examples:

Runnable example is at: example/simple.rb. Please see [rest-more][] for more complex examples to build clients, and slides from rubyconf.tw/2011 for concepts.

Playing Around:

You can also play around with RC::Universal client, which has installed all reasonable middleware built-in rest-core. So the above example could also be achieved by:

require 'rest-core'
client = RC::Universal.new(:site          => 'https://api.github.com/users/',
                           :json_response => true,
                           :log_method    => method(:puts))
client.get('godfat')

RC::Universal is defined as:

module RestCore
  Universal = Builder.client do
    use Timeout       , 0

    use DefaultSite   , nil
    use DefaultHeaders, {}
    use DefaultQuery  , {}
    use DefaultPayload, {}
    use JsonRequest   , false
    use AuthBasic     , nil, nil
    use CommonLogger  , method(:puts)
    use ErrorHandler  , nil
    use ErrorDetectorHttp

    use SmashResponse , false
    use ClashResponse , false
    use  JsonResponse , false
    use QueryResponse , false

    use Cache         , {}, 600 # default :expires_in 600 but the default
                                # cache {} didn't support it

    use FollowRedirect, 10
  end
end

If you have both rib and [rest-more][] installed, you can also play around with an interactive shell, like this:

rib rest-core

And you will be entering a rib shell, which self is an instance of RC::Universal you can play:

rest-core>> get 'https://api.github.com/users/godfat'

will print out the response from Github. You can also do this to make calling Github easier:

rest-core>> self.site = 'https://api.github.com/users/'
rest-core>> self.json_response = true

Then it would do exactly like the original example:

rest-core>> get 'godfat' # you get a nice parsed hash

This is mostly for fun and experimenting, so it's only included in [rest-more][] and rib. Please make sure you have both of them installed before trying this.

List of built-in Middleware:

RC::AuthBasic

RC::Bypass

RC::Cache

use RC::Cache, cache, expires_in

where cache is the cache store which the cache data would be storing to. expires_in would be passed to cache.store(key, value :expires_in => expires_in) if store method is available and its arity should be at least 3. The interface to the cache could be referenced from moneta, namely:

(required) [](key)
(required) []=(key, value)
(optional, required if :expires_in is needed) store(key, value, options)

Note that {:expires_in => seconds} would be passed as the options in store(key, value, options), and a plain old Ruby hash {} satisfies the mandatory requirements: [](key) and []=(key, value), but not the last one for :expires_in because the store method for Hash did not take the third argument. That means we could use {} as the cache but it would simply ignore :expires_in.

RC::CommonLogger

RC::DefaultHeaders

RC::DefaultPayload

RC::DefaultQuery

RC::DefaultSite

RC::Defaults

RC::ErrorDetector

RC::ErrorDetectorHttp

RC::ErrorHandler

RC::FollowRedirect

RC::JsonRequest

RC::JsonResponse

RC::Oauth1Header

RC::Oauth2Header

RC::Oauth2Query

RC::Timeout

Advanced Concurrent HTTP Requests -- Embrace the Future

The Interface

There are a number of different ways to make concurrent requests in rest-core. They could be roughly categorized to two different forms. One is using the well known callbacks, while the other one is using through a technique called future. Basically, it means it would return you a promise, which would eventually become the real value (response here) you were asking for whenever you really want it. Otherwise, the program keeps running until the value is evaluated, and blocks there if the computation (response) hasn't been done yet. If the computation is already done, then it would simply return you the result.

Here's a very simple example for using futures:

require 'rest-core'
YourClient = RC::Builder.client do
  use RC::DefaultSite , 'https://api.github.com/users/'
  use RC::JsonResponse, true
  use RC::CommonLogger, method(:puts)
end

client = YourClient.new
puts "httpclient with threads doing concurrent requests"
a = [client.get('godfat'), client.get('cardinalblue')]
puts "It's not blocking... but doing concurrent requests underneath"
p a.map{ |r| r['name'] } # here we want the values, so it blocks here
puts "DONE"

And here's a corresponded version for using callbacks:

require 'rest-core'
YourClient = RC::Builder.client do
  use RC::DefaultSite , 'https://api.github.com/users/'
  use RC::JsonResponse, true
  use RC::CommonLogger, method(:puts)
end

client = YourClient.new
puts "httpclient with threads doing concurrent requests"
client.get('godfat'){ |v|
         p v['name']
       }.
       get('cardinalblue'){ |v|
         p v['name']
       }
puts "It's not blocking... but doing concurrent requests underneath"
client.wait # until all requests are done
puts "DONE"

You can pick whatever works for you.

A full runnable example is at: example/simple.rb. If you want to know all the possible use cases, you can also see: example/use-cases.rb. It's also served as a test for each possible combinations, so it's quite complex and complete.

Configure the underlying HTTP engine

Occasionally we might want to configure the underlying HTTP engine, which for now is httpclient. For example, we might not want to decompress gzip automatically, (rest-core configures httpclient to request and decompress gzip automatically). or we might want to skip verifying SSL in some situation. (e.g. making requests against a self-signed testing server)

In such cases, we could use config_engine option to configure the underlying engine. This could be set with request based, client instance based, or client class based. Please refer to: How We Pick the Default Value, except that there's no middleware for config_engine.

Here are some examples:

require 'rest-core'
YourClient = RC::Builder.client

# class based:
def YourClient.default_config_engine
  lambda do |engine|
    # disable auto-gzip:
    engine.transparent_gzip_decompression = false

    # disable verifying SSL
    engine.ssl_config.verify_mode = OpenSSL::SSL::VERIFY_NONE
  end
end

# instance based:
client = YourClient.new(:config_engine => lambda do |engine|
  # disable auto-gzip:
  engine.transparent_gzip_decompression = false

  # disable verifying SSL
  engine.ssl_config.verify_mode = OpenSSL::SSL::VERIFY_NONE
end)

# request based:
client.get('http://example.com/', {}, :config_engine => lambda do |engine|
  # disable auto-gzip:
  engine.transparent_gzip_decompression = false

  # disable verifying SSL
  engine.ssl_config.verify_mode = OpenSSL::SSL::VERIFY_NONE
end)

As we stated in How We Pick the Default Value, the priority here is:

request based
instance based
class based

rest-core users:

Powered sites:

GW2 Account Viewer

CHANGES:

CHANGES

CONTRIBUTORS:

Andrew Liu (@eggegg)
andy (@coopsite)
Barnabas Debreczeni (@keo)
Bruce Chu (@bruchu)
Ethan Czahor (@ethanz5)
Florent Vaucelle (@florent)
Jaime Cham (@jcham)
Joe Chen (@joe1chen)
John Fan (@johnfan)
khoa nguyen (@khoan)
Lin Jen-Shin (@godfat)
lulalala (@lulalala)
Man Vuong (@kidlab)
Mariusz Pruszynski (@snicky)
Mr. Big Cat (@miaout17)
Nicolas Fouché (@nfo)
Robert Balousek (@volksport)
Szu-Kai Hsu (@brucehsu)

LICENSE:

Apache License 2.0 (Apache-2.0)

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

https://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

Runtime