Create Microservices in Rails by pretty much just writing regular Rails code.
This gem provides:
This, plus much of what you get from Rails already, means you can create a microservice Rails application by just writing the same Rails code you write today. Instead of rendering web views, you render JSON (which is built into Rails).
Add to your Gemfile
:
gem 'stitches'
Then:
bundle install
Then, set it up:
> bin/rails generate stitches:api
> bin/rails generate stitches:api_migration # only if you're using API key authentication
> bundle exec rake db:migrate # only if you're using API key authentication
If you're not using the API Key authentication feature of the library, configure stitches:
Stitches.configure do |config|
config.disable_api_key_support = true
end
You can enabled it like so
Stitches.configure do |config|
config.max_cache_ttl = 5 # seconds
config.max_cache_size = 100 # how many keys to cache
end
You can also set a leniency for disabled API keys, which will allow old API keys to continue to be used if they have a
disabled_at
field set as long as the leniency is not exceeded. Note that if the disabled_at
field is not populated
the behavior will remain the same as it always was, and the request will be denied when the enabled
field is set to
true
. If Stitches allows a call due to leniency settings, a log message will be generated with a severity depending on
how long ago the API key was disabled.
Stitches.configure do |config|
config.disabled_key_leniency_in_seconds = 3 * 24 * 60 * 60 # Time in seconds, defaults to three days
config.disabled_key_leniency_error_log_threshold_in_seconds = 2 * 24 * 60 * 60 # Time in seconds, defaults to two days
end
If a disabled key is used within the disabled_key_leniency_in_seconds
, it will be allowed.
Anytime a disabled key is used a log will be generated. If it is before the
disabled_key_leniency_error_log_threshold_in_seconds
it will be a warning log message, if it is after that, it will be
an error message. disabled_key_leniency_error_log_threshold_in_seconds
should never be a greater number than
disabled_key_leniency_in_seconds
, as this provides an escallating series of warnings before finally disabling access.
If you are upgrading from a version older than 3.3.0 you need to run three generators, two of which create database
migrations on your api_clients
table:
> bin/rails generate stitches:add_enabled_to_api_clients
> bin/rails generate stitches:add_deprecation
> bin/rails generate stitches:add_disabled_at_to_api_clients
If you are upgrading from a version between 3.3.0 and 3.5.0 you need to run two generators:
> bin/rails generate stitches:add_deprecation
> bin/rails generate stitches:add_disabled_at_to_api_clients
If you are upgrading from a version between 3.6.0 and 4.0.2 you need to run one generator:
> bin/rails generate stitches:add_disabled_at_to_api_clients
Suppose we wish to allow our consumers to create Widgets
class Api::V1::WidgetsController < ApiController
def create
widget = Widget.create(widget_params)
if widget.valid?
head 201
else
render json: {
errors: Stitches::Errors.from_active_record_object(widget)
}, status: 422
end
end
private
def widget_params
params.require(:widget).permit(:name, :type, :sub_type)
end
end
If you think there's nothing special about this—you are correct. This is the vanillaest of vanilla Rails controllers, with a few notable exceptions:
respond_to
and friends.v1
in it, the Accept
header will require a version and get
routed here. If you make a V2, it's just a new controller and this concern is handled at the routing layer.All this means that the Rails skills of you and your team can be directly applied to building microservices. You don't have to make a bunch of boring decisions about auth, versioning, or content-types. It also means you can start deploying and creating microservices with little friction. No need to deal with a complex DSL or new programming language to get yourselves going with Microservices.
See the wiki for how to setup stitches.
Sunset
headerSince version 4.0.0, stitches now has the ability to cache API keys in memory for a configurable amount of time. This may be an improvement for some applications.
You must configure the API Cache for it be used.
Stitches.configure do |config|
config.max_cache_ttl = 5 # seconds
config.max_cache_size = 100 # how many keys to cache
end
Your cache size should be larger then the number of consumer keys your service has.
Although Stitches.configuration
is global, do not depend directly on that in your logic. Instead, allow all classes to receive a configuration object in their constructor. This makes the classes easier to deal with and change, without incurring much of a real cost to development. Global symbols suck, but are convenient. This is how you make the most of it.
Also, the integration test does a lot of "testing the implementation", but since Rails generators are notorious for silently
failing with a successful result, we have to make sure that the various inject_into_file
calls are actually working. Do not do
any fancy refactors here, just keep it up to date.
See the release process for open source gems in the Stitch Fix engineering wiki under technical topics.
Provided with love by your friends at Stitch Fix Engineering