require 'new_relic/agent/instrumentation/metric_frame'
module NewRelic
module Agent
module Instrumentation
# == NewRelic instrumentation for controller actions and tasks
#
# This instrumentation is applied to the action controller to collect
# metrics for every web request.
#
# It can also be used to capture performance information for
# background tasks and other non-web transactions, including
# detailed transaction traces and traced errors.
#
# For details on how to instrument background tasks see
# ClassMethods#add_transaction_tracer and
# #perform_action_with_newrelic_trace
#
module ControllerInstrumentation
def self.included(clazz) # :nodoc:
clazz.extend(ClassMethods)
end
# This module is for importing stubs when the agent is disabled
module ClassMethodsShim # :nodoc:
def newrelic_ignore(*args); end
def newrelic_ignore_apdex(*args); end
end
module Shim # :nodoc:
def self.included(clazz)
clazz.extend(ClassMethodsShim)
end
def newrelic_notice_error(*args); end
def new_relic_trace_controller_action(*args); yield; end
def newrelic_metric_path; end
def perform_action_with_newrelic_trace(*args); yield; end
end
module ClassMethods
# Have NewRelic ignore actions in this controller. Specify the actions as hash options
# using :except and :only. If no actions are specified, all actions are ignored.
def newrelic_ignore(specifiers={})
newrelic_ignore_aspect('do_not_trace', specifiers)
end
# Have NewRelic omit apdex measurements on the given actions. Typically used for
# actions that are not user facing or that skew your overall apdex measurement.
# Accepts :except and :only options, as with #newrelic_ignore.
def newrelic_ignore_apdex(specifiers={})
newrelic_ignore_aspect('ignore_apdex', specifiers)
end
def newrelic_ignore_aspect(property, specifiers={}) # :nodoc:
if specifiers.empty?
self.newrelic_write_attr property, true
elsif ! (Hash === specifiers)
logger.error "newrelic_#{property} takes an optional hash with :only and :except lists of actions (illegal argument type '#{specifiers.class}')"
else
self.newrelic_write_attr property, specifiers
end
end
# Should be monkey patched into the controller class implemented
# with the inheritable attribute mechanism.
def newrelic_write_attr(attr_name, value) # :nodoc:
instance_variable_set "@#{attr_name}", value
end
def newrelic_read_attr(attr_name) # :nodoc:
instance_variable_get "@#{attr_name}"
end
# Add transaction tracing to the given method. This will treat
# the given method as a main entrypoint for instrumentation, just
# like controller actions are treated by default. Useful especially
# for background tasks.
#
# Example for background job:
# class Job
# include NewRelic::Agent::Instrumentation::ControllerInstrumentation
# def run(task)
# ...
# end
# # Instrument run so tasks show up under task.name. Note single
# # quoting to defer eval to runtime.
# add_transaction_tracer :run, :name => '#{args[0].name}'
# end
#
# Here's an example of a controller that uses a dispatcher
# action to invoke operations which you want treated as top
# level actions, so they aren't all lumped into the invoker
# action.
#
# MyController < ActionController::Base
# include NewRelic::Agent::Instrumentation::ControllerInstrumentation
# # dispatch the given op to the method given by the service parameter.
# def invoke_operation
# op = params['operation']
# send op
# end
# # Ignore the invoker to avoid double counting
# newrelic_ignore :only => 'invoke_operation'
# # Instrument the operations:
# add_transaction_tracer :print
# add_transaction_tracer :show
# add_transaction_tracer :forward
# end
#
# Here's an example of how to pass contextual information into the transaction
# so it will appear in transaction traces:
#
# class Job
# include NewRelic::Agent::Instrumentation::ControllerInstrumentation
# def process(account)
# ...
# end
# # Include the account name in the transaction details. Note the single
# # quotes to defer eval until call time.
# add_transaction_tracer :process, :params => '{ :account_name => args[0].name }'
# end
#
# See NewRelic::Agent::Instrumentation::ControllerInstrumentation#perform_action_with_newrelic_trace
# for the full list of available options.
#
def add_transaction_tracer(method, options={})
# The metric path:
options[:name] ||= method.to_s
# create the argument list:
options_arg = []
options.each do |key, value|
valuestr = case
when value.is_a?(Symbol)
value.inspect
when key == :params
value.to_s
else
%Q["#{value.to_s}"]
end
options_arg << %Q[:#{key} => #{valuestr}]
end
class_eval <<-EOC
def #{method.to_s}_with_newrelic_transaction_trace(*args, &block)
perform_action_with_newrelic_trace(#{options_arg.join(',')}) do
#{method.to_s}_without_newrelic_transaction_trace(*args, &block)
end
end
EOC
alias_method "#{method.to_s}_without_newrelic_transaction_trace", method.to_s
alias_method method.to_s, "#{method.to_s}_with_newrelic_transaction_trace"
NewRelic::Control.instance.log.debug("Traced transaction: class = #{self.name}, method = #{method.to_s}, options = #{options.inspect}")
end
end
# Must be implemented in the controller class:
# Determine the path that is used in the metric name for
# the called controller action. Of the form controller_path/action_name
#
def newrelic_metric_path(action_name_override = nil) # :nodoc:
raise "Not implemented!"
end
# Yield to the given block with NewRelic tracing. Used by
# default instrumentation on controller actions in Rails and Merb.
# But it can also be used in custom instrumentation of controller
# methods and background tasks.
#
# This is the method invoked by instrumentation added by the
# ClassMethods#add_transaction_tracer.
#
# Here's a more verbose version of the example shown in
# ClassMethods#add_transaction_tracer using this method instead of
# #add_transaction_tracer.
#
# Below is a controller with an +invoke_operation+ action which
# dispatches to more specific operation methods based on a
# parameter (very dangerous, btw!). With this instrumentation,
# the +invoke_operation+ action is ignored but the operation
# methods show up in RPM as if they were first class controller
# actions
#
# MyController < ActionController::Base
# include NewRelic::Agent::Instrumentation::ControllerInstrumentation
# # dispatch the given op to the method given by the service parameter.
# def invoke_operation
# op = params['operation']
# perform_action_with_newrelic_trace(:name => op) do
# send op, params['message']
# end
# end
# # Ignore the invoker to avoid double counting
# newrelic_ignore :only => 'invoke_operation'
# end
#
#
# When invoking this method explicitly as in the example above, pass in a
# block to measure with some combination of options:
#
# * :category => :controller indicates that this is a
# controller action and will appear with all the other actions. This
# is the default.
# * :category => :task indicates that this is a
# background task and will show up in RPM with other background
# tasks instead of in the controllers list
# * :category => :rack if you are instrumenting a rack
# middleware call. The :name is optional, useful if you
# have more than one potential transaction in the #call.
# * :category => :uri indicates that this is a
# web transaction whose name is a normalized URI, where 'normalized'
# means the URI does not have any elements with data in them such
# as in many REST URIs.
# * :name => action_name is used to specify the action
# name used as part of the metric name
# * :params => {...} to provide information about the context
# of the call, used in transaction trace display, for example:
# :params => { :account => @account.name, :file => file.name }
# These are treated similarly to request parameters in web transactions.
#
# Seldomly used options:
#
# * :force => true indicates you should capture all
# metrics even if the #newrelic_ignore directive was specified
# * :class_name => aClass.name is used to override the name
# of the class when used inside the metric name. Default is the
# current class.
# * :path => metric_path is *deprecated* in the public API. It
# allows you to set the entire metric after the category part. Overrides
# all the other options.
# * :request => Rack::Request#new(env) is used to pass in a
# request object that may respond to uri and referer.
#
# If a single argument is passed in, it is treated as a metric
# path. This form is deprecated.
def perform_action_with_newrelic_trace(*args, &block)
# Skip instrumentation based on the value of 'do_not_trace' and if
# we aren't calling directly with a block.
if !block_given? && _is_filtered?('do_not_trace')
# Also ignore all instrumentation in the call sequence
NewRelic::Agent.disable_all_tracing do
return perform_action_without_newrelic_trace(*args)
end
end
return perform_action_with_newrelic_profile(args, &block) if NewRelic::Control.instance.profiling?
frame_data = _push_metric_frame(block_given? ? args : [])
begin
NewRelic::Agent.trace_execution_scoped frame_data.recorded_metrics, :force => frame_data.force_flag do
frame_data.start_transaction
begin
NewRelic::Agent::BusyCalculator.dispatcher_start frame_data.start
if block_given?
yield
else
perform_action_without_newrelic_trace(*args)
end
rescue Exception => e
frame_data.notice_error(e)
raise
end
end
ensure
NewRelic::Agent::BusyCalculator.dispatcher_finish
# Look for a metric frame in the thread local and process it.
# Clear the thread local when finished to ensure it only gets called once.
frame_data.record_apdex unless _is_filtered?('ignore_apdex')
frame_data.pop
end
end
protected
# Should be implemented in the dispatcher class
def newrelic_response_code; end
def newrelic_request_headers
self.respond_to?(:request) && self.request.respond_to?(:headers) && self.request.headers
end
private
# Profile the instrumented call. Dev mode only. Experimental.
def perform_action_with_newrelic_profile(args)
frame_data = _push_metric_frame(block_given? ? args : [])
val = nil
NewRelic::Agent.trace_execution_scoped frame_data.metric_name do
MetricFrame.current(true).start_transaction
NewRelic::Agent.disable_all_tracing do
# turn on profiling
profile = RubyProf.profile do
if block_given?
val = yield
else
val = perform_action_without_newrelic_trace(*args)
end
end
NewRelic::Agent.instance.transaction_sampler.notice_profile profile
end
end
return val
ensure
frame_data.pop
end
# Write a metric frame onto a thread local if there isn't already one there.
# If there is one, just update it.
def _push_metric_frame(args) # :nodoc:
frame_data = NewRelic::Agent::Instrumentation::MetricFrame.current(true)
frame_data.apdex_start ||= _detect_upstream_wait(frame_data.start)
_record_queue_length
# If a block was passed in, then the arguments represent options for the instrumentation,
# not app method arguments.
if args.any?
if args.last.is_a?(Hash)
options = args.last
frame_data.force_flag = options[:force]
frame_data.request = options[:request] if options[:request]
end
category, path, available_params = _convert_args_to_path(args)
else
category = 'Controller'
path = newrelic_metric_path
available_params = self.respond_to?(:params) ? self.params : {}
end
frame_data.request ||= self.request if self.respond_to? :request
frame_data.push(category + '/'+ path)
frame_data.filtered_params = (respond_to? :filter_parameters) ? filter_parameters(available_params) : available_params
frame_data
end
def _convert_args_to_path(args)
options = args.last.is_a?(Hash) ? args.pop : {}
params = options[:params] || {}
category = case options[:category]
when :controller, nil then 'Controller'
when :task then 'OtherTransaction/Background' # 'Task'
when :rack then 'Controller/Rack' #'WebTransaction/Rack'
when :uri then 'Controller' #'WebTransaction/Uri'
when :sinatra then 'Controller/Sinatra' #'WebTransaction/Uri'
# for internal use only
else options[:category].to_s
end
unless path = options[:path]
action = options[:name] || args.first
metric_class = options[:class_name] || (self.is_a?(Class) ? self.name : self.class.name)
path = metric_class
path += ('/' + action) if action
end
[category, path, params]
end
# Filter out
def _is_filtered?(key)
ignore_actions = self.class.newrelic_read_attr(key) if self.class.respond_to? :newrelic_read_attr
case ignore_actions
when nil; false
when Hash
only_actions = Array(ignore_actions[:only])
except_actions = Array(ignore_actions[:except])
only_actions.include?(action_name.to_sym) || (except_actions.any? && !except_actions.include?(action_name.to_sym))
else
true
end
end
# Take a guess at a measure representing the number of requests waiting in mongrel
# or heroku.
def _record_queue_length
if newrelic_request_headers
if queue_depth = newrelic_request_headers['HTTP_X_HEROKU_QUEUE_DEPTH']
queue_depth = queue_depth.to_i rescue nil
elsif mongrel = NewRelic::Control.instance.local_env.mongrel
# Always subtrace 1 for the active mongrel
queue_depth = [mongrel.workers.list.length.to_i - 1, 0].max rescue nil
end
NewRelic::Agent.agent.stats_engine.get_stats_no_scope('Mongrel/Queue Length').trace_call(queue_depth) if queue_depth
end
end
# Return a Time instance representing the upstream start time.
# now is a Time instance to fall back on if no other candidate
# for the start time is found.
def _detect_upstream_wait(now)
if newrelic_request_headers
if entry_time = newrelic_request_headers['HTTP_X_REQUEST_START']
if newrelic_request_headers['HTTP_X_HEROKU_QUEUE_DEPTH'] # this is a heroku measure
http_entry_time = entry_time.to_f / 1e3
else # apache / nginx
apache_parsed_time = entry_time[/t=(\d+)/, 1]
http_entry_time = apache_parsed_time.to_f/1e6 if apache_parsed_time
end
end
end
# If we didn't find the custom header, look for the mongrel timestamp
http_entry_time ||= Thread.current[:started_on] and http_entry_time = http_entry_time.to_f
if http_entry_time
queue_stat = NewRelic::Agent.agent.stats_engine.get_stats_no_scope 'WebFrontend/Mongrel/Average Queue Time'
total_time = (now - http_entry_time)
queue_stat.trace_call(total_time.to_f) unless total_time.to_f <= 0.0 # using remote timestamps could lead to negative queue time
end
return http_entry_time ? Time.at(http_entry_time) : now
end
def _dispatch_stat
NewRelic::Agent.agent.stats_engine.get_stats_no_scope 'HttpDispatcher'
end
end
end
end
end