How We Scale At Shopify

Shopify started as a single online snowboard shop, and has grown to power 80,000 stores and billions in sales with a single cluster of servers. Customer's flash sales and features like our Turing-complete templating language made this growth challenging. This is the story of how we raced to cache, shard and measure ALL the things in order to be ready for Cyber Monday.

Transcript

1. How We Scale At Shopify CUSEC 2014
11. “A crash course in scaling modern web applications.”
12. The Problem
15. September 2012
18. Measure
20. Response Time
21. Constraints
23. Multitenancy
24. DODOCase Old Faithful
25. Shopify DODOCase
26. Solution
27. • Fetching Resources from Cache • Charging Credit Cards via API • Writing Orders to Database External Calls
28. Cache
29. Page Cache Request HTML Cache Database MISS HIT User
30. Request HTML User Cache Database MISS HIT Object Cache
31. Liquid <ul id="products"> {% for product in products %} <li> <h2>{{ product.name }}</h2> Only {{ product.price | price }} ! {{ product.description | prettyprint | paragraph }} </li> {% endfor %} </ul>
32. Cache
33. Background Processes
35. 6 Requests 1,600 Workers = 9.6K Orders Response Time - Before 60 Seconds 0.1 Requests/Second = 6 Requests x x
36. 600 Requests 1,600 Workers = 960K Orders Response Time - After 60 Seconds 10 Requests/Second = 600 Requests x x
37. 50 Requests 1,600 Workers = 80K Orders Response Time - After 5 Seconds 10 Requests/Second = 50 Requests x x
38. Background Processes
39. Shard
40. Request Master Shard A Shard B Shard ID Product Domain
41. Shard
43. Matt Helm @matthelm