Puppet at Constant Contact
- 1. Puppet at Constant ContactAdoption, the process
- 2. Constant ContactMarket leader for Small BusinessesEmail, Event, Survey & Social Campaigns
- 4. Many customers pay as little as $15 a monthTechnology87 applications, mix of old/new codebase
- 6. ~2 million database transactions per minuteWhy Puppet?Scaling
- 12. Social apps need to iterate fasterHow to manage all these servers?
- 14. Daemon mode, at last
- 15. Cultural Change, part oneOk, let’s start withIncremental Config& Push
- 16. Time to market better, but…
- 17. Special thanks to:Our DevOps, “we can do this with Puppet” guy, Mark Schena& our puppetmaster, AleksMargan
- 18. Cultural Change, part two
- 19. It works, right?Yes, errr. Well… kinda.
- 20. 12
- 22. The dream of CI
- 23. Fix The Flow
- 24. I Я COW!
- 27. 19
- 28. puppet repositoryapplication repositorycommon puppet modules
- 30. code
- 31. puppet module
- 33. database schemaRepository layoutSource Control
- 34. Two types of puppet modules +=common modulerunning applicationapplication moduleCommon modules provide non-application-specific classes and resources.
- 35. Application modules leverage common modules to configure an environment for a given application.Okay, I lied, there’s three types of puppet modules
- 37. complete set of puppet modulesenvironmentenvironmentenvironmentenvironmentQA-1QA-2PRODDeploymentpuppetmaster
- 38. Thanks!“Dave Connors” <dconnors@constantcontact.com>VP, Operations“Hawk Newton” <hnewton@constantcontact.com> Software Architect
Editor's Notes
- #2: Quick snapshot of CTCTLittle bit on how we adopted PuppetThen over to the cleanup hitter
- #3: Fast growingMicro businessesOffices in Loveland, Boca, SF and NYCLots of apps, at decent scale --- Top 20 email senders worldwide
- #4: Business DriversDiverse: not all uniform web app servers multiple applicationsDistributed: memcached, sso, coherence, cassandraManage it cost-effectively, keep things consistent in order to max uptime and mean time to recovery - don’t slow down the rate of changeAnd oh yeah, improve gross margins
- #5: We hit a bit of a tipping point on managing servers, couple of hundred or soStarted with a small piece of the puzzleSystem stuff resolve.conf, host…etcWork out the basic process, gain trust
- #6: Getting adoption by leaving control in hands of adminsThe declarative model was challenging to many at firstPush via func was a good interim step
- #7: Good news along the way thoughConfidence gained – Automated System Warning!Moved mail subsystem to daemon modeStandard well understood configuration postfix and centosSystems folks were onboard
- #8: Change in control, ownership, processesRetooling without stopping rate of changeGaining trust while changing folx mindset
- #9: Manage social data cost-effectively3 months cycle, cheaperBut still had challenges around branching and testingIntegration in pre-productionIndependent upgradesSo we had some work to do, including orchestration
- #10: Quick Shout Out
- #11: We could only go so far pulling from the Systems sideBut it had to work for Developers, so…Here you go Dev, own the config, trust in Hawk
- #12: The monolith!We had some significant challenges:New appsUgly handoffsEnvironmentsDecreased velocityFUDFrustrated parties on both sides of the house…