Effective, Stress-free Varnish Development (at Scale)

Here's some tips about how to leverage infrastructure-as-code to safely and effectively manage changes to a large-scale Varnish caching infrastructure.

Transcript

1. Effective, Stress-free Varnish Development
2. Once upon a time…. 2
3. Pantheon.io Egads! Customer reports cookie issue <3 Product M anagers
4. Pantheon.io Confident engineer tweaks regex WordPress cookie regex in VCL Accompanying test
5. Pantheon.io Pass! Merge! PASS
6. Pantheon.io 4 days, no sweat 4-day cycle time
7. Pantheon.io Total Domination (and fewer heart-attacks)
8. Isn’t testing for code, not config? 8
9. Pantheon.io Varnish is great, serving 360k Websites Breezing through: • Exclusive interviews with President Obama • Drudge Report’d spikes • Various DDoS • Duck Commandering, • Year-end non-profit fundraising • $348 Million Dollar Jackpots • And many more real-world successes!
10. Pantheon.io One VCL mistake away from 15,000 /second
11. Pantheon.io • Version Controlled • Peer Reviewed • Syntax checked • Functionally Tested • Automated Deployment Infra as code .___ _____ | | _____/ ____\___________ | |/ \ __\\_ __ \__ \ | | | \ | | | \// __ \_ |___|___| /__| |__| (____ / \/ \/ \__ \ / ___/ / __ \_\___ \ (____ /____ > \/ \/ .___ ____ ____ __ | _/____ _/ ___\/ _ \ / __ |/ __ \ \ \__( <_> ) /_/ \ ___/ \___ >____/\____ |\___ > \/ \/ \/
12. Pantheon.io THIS IS WHAT PASSES FOR PROFESSIONAL-GRADE VARNISH ADMINISTRATION*: Current State of the Art ;) *Agoddard is awesome and definitely knows what the plots mean.
13. Pantheon.io VCL is…. 13 Easy to understand!
14. Pantheon.io R - E - S - P - E - C -T 14
15. Beyond testing: checking user behaviors 15
16. Pantheon.io Translating to user req’s to VCL hard • Translating User behavior to technical functionality is hard • “Transliterating” technical functionality from code/conf to tests can yield tautologically validating tests 16
17. Pantheon.io (User) Context Matters • Context from years of Drupal, WordPress, 4Kitchens, GoDaddy, Economist.com, Chapter 3, etc • We all rely on context, but how can we nuture it? 17
18. Pantheon.io Develop and nuture user-based context by talking about the users. A lot. Having the (User Behavior) Conversation 18 We talk about user behaviors a lot. Too much?
19. Pantheon.io Varnishtest Cucumber https://github.com/nstielau/varnishtest_cucumber Check this out if you want to get started testing your VCL ASAP.
20. Pantheon.io Same same, but different 20 Adam noted Acquia infra’s similar approach...
21. How we test Varnish 21 (Writing this down, I realized there is a lot that helps make this happen)
22. Pantheon.io CI Workflow Github CircleCI Docker Docker Pull Request Kicks off CircleCI CircleCI Kicks off Docker
23. Pantheon.io Docker Workflow Docker Container chef-solo
24. Pantheon.io Test multiple versions of Fedora, or Varnish Varnish Port 80 Mock Server Port 80 Varnish Port 80 Mock Server Port 80 22 20
25. Pantheon.io Testing Architecture Mock Server Port 80 1. Test process spawns mock server 2. Varnish configured using Chef, pointed at mock server 3. Test process issues use-case specific HTTP requests to Varnish 4. Test process verifies correct response headers, correct backend-server accesses
26. Pantheon.io #4: Assert response. http-Age != 0 Example: Assert images are cached Mock Server Port 80 #1: GET /image.jpg #2: Backend GET /image.jpg #3: Second GET /image.jpg
27. Pantheon.io Snippit: Don’t cache big files For each extension For directory location For each varying cookie Assert 2nd request is miss
28. Pantheon.io Why not varnishtest? 28 ✓ Better looping ✓ Easier to write ✓ Varnish agnostic ✓ Includes Chef templating, spawning ✓ Control request, response, backend request, backend response ✓ (I just found out about it)
29. Other Stuff 29
30. Pantheon.io Delivering JUSTICE with Varnish Fending off (relatively polite) DDoS with Varnish:
31. Pantheon.io 70-90% hit rate Telemetry with Sensu, Graphite, Grafana