Error Reporting in Rails Applications

This guide introduces ways to manage exceptions that occur in Ruby on Rails applications.

After reading this guide, you will know:

• How to use Rails’ error reporter to capture and report errors.
• How to create custom subscribers for your error-reporting service.

The Rails error reporter provides a standard way to collect exceptions that occur in your application and report them to your preferred service or location.

The error reporter aims to replace boilerplate error-handling code like this:

begin
  do_something
rescue SomethingIsBroken => error
  MyErrorReportingService.notify(error)
end

with a consistent interface:

Rails.error.handle(SomethingIsBroken) do
  do_something
end

Rails wraps all executions (such as HTTP requests, jobs, and rails runner invocations) in the error reporter, so any unhandled errors raised in your app will automatically be reported to your error-reporting service via their subscribers.

This means that third-party error-reporting libraries no longer need to insert a Rack middleware or do any monkey-patching to capture unhandled exceptions. Libraries that use ActiveSupport can also use this to non-intrusively report warnings that would previously have been lost in logs.

Using the Rails’ error reporter is not required. All other means of capturing errors still work.

To use the error reporter, you need a subscriber. A subscriber is any object with a report method. When an error occurs in your application or is manually reported, the Rails error reporter will call this method with the error object and some options.

Some error-reporting libraries, such as Sentry’s and Honeybadger’s, automatically register a subscriber for you. Consult your provider’s documentation for more details.

You may also create a custom subscriber. For example:

# config/initializers/error_subscriber.rb
class ErrorSubscriber
  def report(error, handled:, severity:, context:, source: nil)
    MyErrorReportingService.report_error(error, context: context, handled: handled, level: severity)
  end
end

After defining the subscriber class, register it by calling Rails.error.subscribe method:

Rails.error.subscribe(ErrorSubscriber.new)

You can register as many subscribers as you wish. Rails will call them in turn, in the order in which they were registered.

Note: The Rails error-reporter will always call registered subscribers, regardless of your environment. However, many error-reporting services only report errors in production by default. You should configure and test your setup across environments as needed.

There are three ways you can use the error reporter:

Rails.error.handle will report any error raised within the block. It will then swallow the error, and the rest of your code outside the block will continue as normal.

result = Rails.error.handle do
  1 + '1' # raises TypeError
end
result # => nil
1 + 1 # This will be executed

If no error is raised in the block, Rails.error.handle will return the result of the block, otherwise it will return nil. You can override this by providing a fallback:

user = Rails.error.handle(fallback: -> { User.anonymous }) do
  User.find_by(params[:id])
end

Rails.error.record will report errors to all registered subscribers and then re-raise the error, meaning that the rest of your code won’t execute.

Rails.error.record do
  1 + '1' # raises TypeError
end
1 + 1 # This won't be executed

If no error is raised in the block, Rails.error.record will return the result of the block.

You can also manually report errors by calling Rails.error.report:

begin
  # code
rescue StandardError => e
  Rails.error.report(e)
end

Any options you pass will be passed on the error subscribers.

All 3 reporting APIs (#handle, #record, and #report) support the following options, which are then passed along to all registered subscribers:

• handled: a Boolean to indicate if the error was handled. This is set to true by default. #record sets this to false.
• severity: a Symbol describing the severity of the error. Expected values are: :error, :warning, and :info. #handle sets this to :warning, while #record sets it to :error.
• context: a Hash to provide more context about the error, like request or user details
• source: a String about the source of the error. The default source is "application". Errors reported by internal libraries may set other sources; the Redis cache library may use "redis_cache_store.active_support", for instance. Your subscriber can use the source to ignore errors you aren’t interested in.

Rails.error.handle(context: {user_id: user.id}, severity: :info) do
  # ...
end

With Rails.error.handle and Rails.error.record, you can also choose to only report errors of certain classes. For example:

Rails.error.handle(IOError) do
  1 + '1' # raises TypeError
end
1 + 1 # TypeErrors are not IOErrors, so this will *not* be executed

Here, the TypeError will not be captured by the Rails error reporter. Only instances of IOError and its descendants will be reported. Any other errors will be raised as normal.

In addition to setting context through the context option, you can use the #set_context API. For example:

Rails.error.set_context(section: "checkout", user_id: @user.id)

Any context set this way will be merged with the context option

Rails.error.set_context(a: 1)
Rails.error.handle(context: { b: 2 }) { raise }
# The reported context will be: {:a=>1, :b=>2}
Rails.error.handle(context: { b: 3 }) { raise }
# The reported context will be: {:a=>1, :b=>3}

Error-reporting libraries can register their subscribers in a Railtie:

module MySdk
  class Railtie < ::Rails::Railtie
    initializer "error_subscribe.my_sdk" do
      Rails.error.subscribe(MyErrorSubscriber.new)
    end
  end
end

If you register an error subscriber, but still have other error mechanisms like a Rack middleware, you may end up with errors reported multiple times. You should either remove your other mechanisms or adjust your report functionality so it skips reporting an exception it has seen before.