Application Monitoring

Introduction

Monitoring provides continuous visibility into application health. While profiling is for diagnosing specific problems, APM tools track performance over time, revealing trends and catching regressions.

Key Concepts

• APM (Application Performance Monitoring): Tools like New Relic or Scout that continuously track response times, error rates, and throughput in production.
• Percentiles: Response time metrics (p50, p95, p99) that show the distribution of request speeds, revealing outliers that averages hide.
• Instrumentation: Rails' built-in event system that fires notifications for SQL queries, controller actions, and view rendering.

Real World Context

Your average response time is 200ms, but monitoring reveals p99 is 12 seconds. 1% of users — potentially hundreds per day — experience terrible performance. Without percentile monitoring, this goes undetected.

Deep Dive

Rails Built-in Instrumentation

ruby
ActiveSupport::Notifications.subscribe('process_action.action_controller') do |*args|
  event = ActiveSupport::Notifications::Event.new(*args)
  Rails.logger.info({
    path: event.payload[:path],
    status: event.payload[:status],
    duration: event.duration,
    db_time: event.payload[:db_runtime]
  }.to_json)
end

New Relic Setup

ruby
# Gemfile
gem 'newrelic_rpm'

yaml
# config/newrelic.yml
production:
  license_key: <%= ENV['NEW_RELIC_LICENSE_KEY'] %>
  app_name: My Rails App

Custom Metrics with StatsD

ruby
class OrdersController < ApplicationController
  def create
    start_time = Time.current
    @order = Order.create!(order_params)
    StatsD.increment('orders.created')
    StatsD.timing('orders.create_time', (Time.current - start_time) * 1000)
  rescue => e
    StatsD.increment('orders.failed')
    raise
  end
end

Common Pitfalls

1. Only tracking averages — Averages hide outliers. Track p95 and p99 to catch the worst user experiences.
2. Not monitoring Solid Queue latency — A healthy web server with a backed-up job queue means emails aren't sending and imports aren't processing.

Best Practices

1. Set alerts on p95 response time — Alert when response times exceed your SLA threshold.
2. Monitor error rates — A sudden spike in 500 errors after a deploy indicates a problem. Set alerts at > 1% error rate.

Summary

• APM tools provide continuous production monitoring.
• Track percentiles (p50, p95, p99), not just averages.
• Use Rails' built-in instrumentation for custom event tracking.
• Monitor both web response times and job queue latency.

# Log slow SQL queries automatically
ActiveSupport::Notifications.subscribe('sql.active_record') do |*args|
  event = ActiveSupport::Notifications::Event.new(*args)
  if event.duration > 100
    Rails.logger.warn "SLOW QUERY (#{event.duration.round}ms): #{event.payload[:sql]}"
  end
end