Web Application Monitoring with Cucumber

Presented at RailsConf 2010 (http://en.oreilly.com/rails2010/public/schedule/detail/14415)

Transcript

1. Feature: Ruby on Rails Application Monitoring with Cucumber In order to ensure continuous application availability A developer should be able to assert the behavior of production apps From the outside in Without using antiquated monitoring tools To protect revenue
2. VP of Research & Development railsmachine.com jesse@railsmachine.com @jnewland github.com/jnewland About me: I get to hack on Ruby tools to manage large Rails deployments all day long. Not a bad job, eh?
3. Before we get into monitoring or cucumber, let’s talk about testing. In my career as a dev, my testing habits have evolved over time, largely inspired by available tools. I’m sure some of you have shared a similar journey - let’s take a quick look back.
4. No more clicking around Save in your editor / refresh in your browser / lather / rinse repeat. Occasional human preformed quality assurance Broken by design
5. I then made the jump to unit testing using Ruby’s Test::Unit - specifically the generated Model and Controller tests Rails generated. This was nice, but it was often devalued by stakeholders due to poor communication of the business value of this work on my part.
6. R Enter Rspec and the BDD movement. Rspec helped me, and I’m sure a lot of others, associate the business value with writing tests / specs. Stakeholder-digestable code if you’re really good, stakeholder-digestable output if you’re doing things right.
7. C U C U M B E R Basically, BDD nirvana. Stakeholder-*writable* if you’re crazy.
8. Cucumber lets software development teams describe how software should behave in plain text. The text is written in a business-readable domain- specific language and serves as documentation, automated tests and development-aid - all rolled into one format. For those of you that aren’t familiar with Cuke
9. TATFT! The most important part of the evolution of these tools is that they make it easy and - legitimately - fun to test first and test all of the time as you’re developing your application.
10. Production Monitoring But what about production? We’re testing all the time in development, while we’re developing the that’s going to create revenue. But in production...
11. Revenue Preservation ...there’s actually revenue being earned. Why not test with the same veracity in production?
12. Current Monitoring Landscape Quiz: * Raise your hand if you are at least partially responsible for the continuous operation of a business critical production rails app * If you have ZERO monitoring of the site’s uptime - meaning your customers or boss would be the one to tell you that the homepage was down - put your hand down * If your monitoring solution runs on your server itself - monit or god, for example - put your hand down * If your external monitoring solution only hits one URL on the site, put your hand down Some sites are monitored very closely, but I’ve found that in most cases, the monitoring of many production apps is rather slim. I generally evaluate monitoring solutions on two axes:
13. What’s being monitored - what URLs, metrics, system statistics, etc are being watched
14. How closely are you looking at it? We’ll call this one the crazy monkey test - How frequently these URLs / metrics are being queried, what values are acceptable, etc.
15. It seems that in many situations, the home page of an application is the only thing checked closely
16. The crazy monkey has laser focus
17. but if the crazy monkey is that focused
19. Bad things can happen when he’s not looking. For example, in Rails apps, I see this happen all the time with...
20. Search is a part of many applications that I’ve seen go unmonitored. I’m not singling out sphinx here - this is just a sweet picture - the same thing happens to Solr, etc
21. Search can fail when the rest of a site works fine due to many reasons: * search daemon may go down * the indicies may be corrupt * or things may fail in a more interesting kind of way...
22. 0 results for “beer” Wherein no results are returned when they obviously should be.
23. TATFT? So are we really testing all the time?
24. TATFT* *except in production It doesn’t seem so
25. _why? But why? Why are we testing so....ferociously in development, but so weakly in production?
26. Old Broken Tools I’m largely convinced it’s because the tools that are presented to us for use in the monitoring space are largely old and broken.
27. How many of your recognize this? Oh, nagios.
28. It’s the industry standard tool for infra monitoring. I haven’t met a single person that’s used nagios that’s been an honest fan. The most widely despised part of nagios
29. is the noise. Unless masterly configured, Nagios is a noisy beast. This leads to “boy cries wolf” type scenarios, wherein alerts are improperly categorized as noise and discarded.
30. EVIL Because of the noise, and the piece of crap interface, esoteric configuration language, and for years and years of waking me up for false positives, I’m going to paint this all in black and white and just call nagios evil.
31. Pingdom’s a relatively new tool that’s gained a good bit of traction. It’s a hosted monitoring service, that can test HTTP and many other types of services from a network of computers around the world.
32. ✔ Nagios and pingdom pass the crazy monkey intense focus test
33. ✘ but in their default configuration generally only monitor a snapshot of what’s neccessary.
34. A recent entry into the space that’s doesn’t get a quick EVIL stamp from me is watchmouse
35. Twitter uses watchmouse to provide a public API status page, hitting many different API endpoints and watching for outages and service problems
36. ✔ Twitter’s use of Watchmouse passes the “what are you looking at test”
37. Business Value Disconnect However, one thing that all of these tools are missing is a clear link between the business value of the things they’re checking and the alerts they’re sending out
38. Hey, I know something that does that well!
39. Cucumber lets software development teams describe how software should behave in plain text. The text is written in a business-readable domain- specific language and serves as documentation, automated tests and development-aid - all rolled into one format. Cucumber’s served well for me in my experience in bringing stakeholders and developers together.
40. Cucumber lets software development teams describe how software should behave in plain text. The text is written in a business-readable domain- specific language and serves as documentation, automated tests and development-aid - all rolled into one format. But with a couple quick edits
41. Cucumber also lets operations teams describe how infrastructure should behave in plain text. The text is written in a business-readable domain- specific language and serves as documentation, monitoring and deployment-aid - all rolled into one format. We have a tool that can help us bring together developers, operations, *and* stakeholders
42. #devops Some of you following the twitterz may have noticed some people in the ops and development space talking about the ‘devops movement’
43. devs ops working together While calling this a movement is pretty wild - a hashtag does not a movement make - the ideas surrounding this ‘movement’ are things that I believe in personally, and things we’re working on everyday at Rails Machine - blurring the line between development and ops, and the line between the infrastructure and the application.
44. Cucumber also lets #devops teams describe how applications should behave in plain text. The text is written in a business-readable domain- specific language and serves as documentation, monitoring and deployment-aid - all rolled into one format. Using cucumber in production embodies everything that is devops, and can blur those lines even more
45. kumbaya And thus result in a big happy #devops family
46. Example Production Cucumber Features
47. Benchmarking
48. Feature: slashdot.com To keep the geek masses satisfied Slashdot must be responsive Scenario: Cached pages are super quick Given I am benchmarking When I go to http://slashdot.org/ Then the elapsed time should be less than 500 milliseconds When I follow "Login" Then the elapsed time should be less than 500 milliseconds When I follow "Contact" Then the elapsed time should be less than 500 milliseconds
49. Email Deliverability
50. Feature: Signup Emails In order to prevent bots from taking over the site A new user should receive a verification email upon signup Scenario: New User signup Given I visit "http://example.com" And I follow "Signup!" When I signup with a random email address and password And I press "Go" And I wait 10 seconds # an unfortunate reality Then I should have one email in my inbox And the email subject should match "^Welcome" And the email body should match "http:\/\/example.com\/v\/\w+" https://github.com/technicalpickles/mailinator-spec
51. Existing Metrics
52. Feature: Response Time As a impatient user Our web server should be in tip-top shape So our app can be super fast Background: Given my Scout account name is 'railsmachine' And my Scout email and password are 'jesse@railsmachine.com' and 'sekret' Scenario: Passenger Queue When I get the metrics from the 'Passenger' plugin on 'example.com' Then the 'passenger_queue_depth' should be 0 Scenatiro: CPU usage is low When I get the metrics from the 'Server Overview' plugin on 'example.com' Then 'cpu_last_minute' should be less than 1 http://github.com/jnewland/cucumber-scout/
53. Feature: Response Time As a impatient user Our app should be super fast Background: Given my NewRelic license key is 'omgwtfbbq' Scenario: Average Response time Given that my application is being monitored by New Relic Then my application's 'response time' should be less than 500 milliseconds Scenario: Apdex Given that my application is being monitored by New Relic Then my application's 'apdex' should be 1 http://github.com/jnewland/cucumber-newrelic
54. SEO
55. Feature: Cucumber wiki discoverability In order to learn more about Cucumber As an uninformed developer I should be able easily find the GitHub wiki Scenario: Searching for Cucumber on Google When I go to http://www.google.com/ And I fill in "q" with "cucumber" And I press "Google Search" Then I should see "BDD that talks to domain experts first and code second"
56. Security
57. Feature: example.org ssh logins As a user of example.org I need to login remotely Scenario: Login with a key Given I have the following public keys: | keyfile | | /home/jnewland/.ssh/id_dsa | Then I can ssh to the following hosts with these credentials: | hostname | username | | example.org | jnewland | | mail.example.org | jnewland | Scenario: Checking /etc/passwd When I ssh to "example.org" with the following credentials: | username | password | keyfile | | jnewland | | /home/jnewland/.ssh/id_dsa | And I run "cat /etc/passwd" Then I should see "jnewland" in the output And I should not see "that_dude_we_just_fired" in the output http://github.com/auxesis/cucumber-nagios
58. Infrastructure
59. Feature: RAID To ensure optimal server operation And guarantee data is stored redundantly The RAID array should be in a good state Scenario: RAID Array status When I check the raid array status Then controller "1" should have a status of "optimal" And controller "2" should have a status of "optimal" And controller "1" should have "1" logical device with a status of "optimal" And controller "1" should have "4" drives in "online" state And controller "2" should have "1" logical device with a status of "optimal" And controller "2" should have "4" drives in "online" state http://github.com/auxesis/cucumber-nagios
60. DNS
61. Feature: rubygems.org As a member of the Ruby community I should be able to easily install Ruby gems Scenario: DNS When I lookup "rubygems.org" Then the name should resolve an IP http://github.com/auxesis/cucumber-nagios
62. Possibilities
63. Credit Card Transactions
64. SSL
65. Exception Rate
66. Running features in production
67. $ gem install cucumber-json cucumber-newrelic \ cucumber-scout cucumber-nagios $ cd RAILS_ROOT $ mkdir -p production_features/step_definitions $ mkdir -p production_features/support $ vi config/cucumber.yml production: production_features -f Cucumber::Formatter::JSON --out tmp/cuke.json $ vi production_features/support/env.rb require 'cucumber/nagios/steps' require 'cucumber/newrelic' require 'cucumber/scout' # etc $ # hack on features $ cucumber -p production # doesn’t load the Rails env, just the defined steps $ # profit! Sorta Quick Setup Generator coming soon!
69. Cucumber Scout Plugin
70. http://github.com/jnewland/scout-plugins/raw/cucumber_ci/ cucumber_ci/cucumber_ci.rb Cucumber Scout Plugin
71. Cucumber Scout Plugin
72. Cucumber Scout Plugin ZOMG!
73. Cucumber Scout Plugin Sent via email
74. Power to monitor anything and everything
75. Clearly Defined Business Value
76. I N R U B Y
77. 24/7/365
78. Know before your customers do
79. Thanks! Any questions?
80. Get in touch: Jesse Newland jesse@railsmachine.com @jnewland github.com/jnewland
81. Flickr Creative Commons Photos http://flic.kr/p/4EjsDJ http://flic.kr/p/c1UTf http://flic.kr/p/5BGJMu http://flic.kr/p/5iTuua http://flic.kr/p/77oAy2 http://flic.kr/p/CdYo8 http://flic.kr/p/k2LCm http://flic.kr/p/71vxY6 http://flic.kr/p/5aZYkP http://flic.kr/p/79ikH2 http://flic.kr/p/6rhqad http://flic.kr/p/4MdrW8 http://flic.kr/p/5WuXzM http://flic.kr/p/3jzrJ http://flic.kr/p/5B4TaF http://flic.kr/p/4FAf2R http://flic.kr/p/39poLP http://flic.kr/p/6nq52E http://flic.kr/p/yQuMG http://flic.kr/p/jZ5Ae http://flic.kr/p/4yhzz http://flic.kr/p/nxAqt http://flic.kr/p/4VWY5F http://flic.kr/p/EKbuF http://flic.kr/p/5xitHh http://flic.kr/p/4uE9Wz http://flic.kr/p/65KZaJ http://flic.kr/p/7JKj5H http://flic.kr/p/79HLb5 http://flic.kr/p/xiYny http://flic.kr/p/68vjKV http://flic.kr/p/rvc1 http://flic.kr/p/6y7EfX http://flic.kr/p/2Mxkhp http://flic.kr/p/5t7h5 http://flic.kr/p/29qc7 http://flic.kr/p/JBaj http://flic.kr/p/smfab http://flic.kr/p/4t5Qf9 http://flic.kr/p/M8kdv http://flic.kr/p/z3eWm http://flic.kr/p/4XAQs7