1. Continuous Deployment at : A Tale of Two ApproachesRoss Snyderross@etsy.com@beamrider9 March 9, 2013
2. A quick primer on
3. is:The global marketplace we make together.
4. is:The premier destination for handmade goods, vintage items, and craft supplies. simplertimestoys lacklusterco norwesterseaglass
5. quick facts: (as of March 2013)• 22+ million members• 800,000+ active shops• 18+ million items currently for sale• 20 cents to list item, 3.5% transaction fee• 400+ employees (majority in Brooklyn)
6. Since opening its doors in June 2005, Etsy has grown virtually non-stop. $1,000 $800 $600 $400 $200 $0 2005 2006 2007 2008 2009 2010 2011 2012 Gross Merchandise Sales ($MM)
7. A nice problem to have: “Our site is sosuccessful, how can we move fast enough tokeep up with demand?”
9. :The Early Years (2005 - 2008)
10. : The Early Years1. Spend signiﬁcant time writing code
11. : The Early Years1. Weeks writing code 2. Painful source control merge
12. : The Early Years1. Weeks writing code2. Painful merge 3. Hand oﬀ to someone else to deploy
13. : The Early Years1. Weeks writing code2. Painful merge3. Hand oﬀ to deployers 4. Deploy, site goes down
14. : The Early Years1. Weeks writing code 4. Deploy, site down2. Painful merge3. Hand oﬀ to deployers 5. Roll back deploy
15. : The Early Years1. Weeks writing code 4. Deploy, site down2. Painful merge 5. Roll back deploy3. Hand oﬀ to deployers 6. Spend hours (days?) ﬁxing bugs
16. : The Early Years1. Weeks writing code 4. Deploy, site down2. Painful merge 5. Roll back deploy3. Hand oﬀ to deployers 6. Fix bugs 7. Go back to step 2
17. : The Early Years WATERFALL!
18. : The Early YearsPros:Early Etsy engineersused this releasecycle to bootstrap themarketplace fromnothing.Forever grateful.
19. : The Early YearsCons:• Large changesets• Infrequent deploys• Weak conﬁdence in deploy success• Signiﬁcant time spent deploying• Low ability to experiment/iterate/react• Developer stress/unhappiness
20. : The Early YearsBy late 2008, Etsy is still a startup, but has the deploy process of a much bulkier company.Popularity is on the verge of outpacing capacity.
22. : Today1. Small changesets, deployed frequently
23. : Today1. Small changesets 2. Engineers deploy the site
24. : TodayAnd not just engineers, but also:• Designers• Product Folks• Upper Management• Board Members• Dogs
25. : Today1. Small changesets2. Engineers deploy 3. Deploys are fast and near-eﬀortless
26. : Today1. Small changesets2. Engineers deploy3. Deploys are fast 4. Most changes behind conﬁg ﬂags (safer deploys)
27. : Today1. Small changesets 4. Changes behind ﬂags2. Engineers deploy3. Deploys are fast 5. Graphs/metrics to assess deploy
28. : Today1. Small changesets 4. Changes behind ﬂags2. Engineers deploy 5. Copious graphs/metrics3. Deploys are fast6. If issues, ﬁx immediately & roll forward
29. : Today This isn’t license to break stuﬀ, quickly.Engineer-driven QA and solid unit testing are integral parts of the process.
30. : Today1. Small changesets 4. Changes behind ﬂags2. Engineers deploy 5. Copious graphs/metrics3. Deploys are fast 6. Fix fast & roll forward 7. Repeat 25+ times per day, every day
31. Then: 1. Weeks writing code 2. Painful merge 3. Hand oﬀ to deployers 4. Deploy, site down 5. Roll back deploy 6. Fix bugs, go to step 2Now: 1. Small changesets 2. Engineers deploy 3. Deploys are fast 4. Changes behind ﬂags 5. Copious graphs/metrics 6. Fix fast & roll forward
32. Etsy Deploy Stats: 2012• Deployed to production 6,419 times• On average, 535/month, 25/day• Additional 3,851 conﬁg-only deploys• 196 diﬀerent people deployed to prod• Nov/Dec 2012: deployed 752 times
33. Why does it work?
34. Continuous Deployment Math • N = # of deploys • P = probability of site degradation • S = average severity of degradation • T = time to detect/resolve Expected = N*P*S*T Downtime
35. Continuous Deployment MathN = # of deploys S = avg. severity of degradationP = prob. of degradation T = time to detect/resolve Before: Now: •N=1 • N = 250 ↑↑↑↑ • P = 0.5 • P = 0.1 ↓ • S = 0.7 • S = 0.05 ↓↓ • T = 100 • T = 5 ↓↓↓ E.D. = 35 E.D. = 6.25 (all numbers completely arbitrary)
36. Big Takeaway Etsy circa 2013 (400+ employees)acts, in some ways, more like a startup than Etsy circa 2008 (40+ employees).
37. Continuous Deployment makes possible: “Continuous Experimentation” http://etsy.me/continuous-experimentation
38. Continuous Experimentation 1. Small changes 2. Run experiment (A/B test) 3. Analyze data 4. Re-examine assumptionsRepeat continuously in pursuit of larger goals.
39. Heard since 2010:“Neat experiment, but this will never scale.” As of 2013, Etsy has 100+ engineers - still going strong.
40. Some Etsy CustomizationsDeploying is a ﬁrst-class feature. Inability todeploy is a P1 incident (same as site down).
41. Some Etsy CustomizationsWe continuously deploy not just the main Etsy website, but as much as possible: • Internal admin site • API • Big data • Search • Blog • Deployinator itself
42. Some Etsy Customizations In the rare case we can’t continuously deploy, we create alternative tools: • Database schema changes • PCI-DSS environment (credit cards) We do continuously deploy as much of ourpayment processing as is safe & legal (98%).
43. Some Etsy CustomizationsKeeping deploys fast is paramount and worth the investment in manpower & hardware.
44. Some Etsy Customizations Continuous deployment is all about movingforward, sometimes at the expense of the past.Our solution: engineering-wide bug rotation,one day a month, every engineer participates.
45. Fun Fact: Continuous Deployment is a fantasticrecruitment tool for attracting engineerswho like to move fast and get stuﬀ done.
46. Learn more:http://codeascraft.etsy.com/Etsy open source (Deployinator, StatsD)http://etsy.github.com/Join the fun:http://www.etsy.com/careers