A Tale of Two Feature Flags

RubyConf 2015

Transcript

1. RubyConf 2016 A Tale of Two Feature Flags November 16 Rebecca Sliter
2. Hi RubyConf! I’m @rebeccasliter November 16 Engineering Manager at Kickstarter
3. Hi RubyConf! I care about helping teams deliver better code more efficiently November 16
4. Hi RubyConf! A Tale of Two Feature Flags November 16
5. It was the best of times, It was the worst of times. It was the age of foolishness, it was the epoch of belief, it was the epoch of incredulity, it was the season of Light, it was the season of Darkness, it was the spring of hope, it was the winter of despair, we had everything before us, we had nothing before us, we were all going direct to Heaven, we were all going direct the other way. November 16
6. It was the best of times, It was the worst of times. It was the age of foolishness, it was the epoch of belief, it was the epoch of incredulity, it was the season of Light, it was the season of Darkness, it was the spring of hope, it was the winter of despair, we had everything before us, we had nothing before us, we were all going direct to Heaven, we were all going direct the other way. November 16
7. It was the best of times. November 16
8. I have an idea! The best of times How are feature flags supposed to work?
9. The best of times I have an idea! How are feature flags supposed to work?
10. The best of times I have an idea! But… • It’s a lot of work. How are feature flags supposed to work?
11. The best of times • It’s a lot of work. • I’m worried about deploying months of work all at once. How are feature flags supposed to work? I have an idea! But…
12. The best of times • It’s a lot of work. • I’m worried about deploying months of work all at once. • We’d really like to see how it performs with a small subset of users. How are feature flags supposed to work? I have an idea! But…
13. The best of times • It’s a lot of work. • I’m worried about deploying months of work all at once. • We’d really like to see how it performs with a small subset of users. • There are lots of external dependencies before we can roll it out to everyone. How are feature flags supposed to work? I have an idea! But…
14. The best of times I have an idea! … so, how can I release this feature while minimizing risk? How are feature flags supposed to work?
15. big_feature: false The best of times Put a feature flag on it! Configuration file How are feature flags supposed to work?
16. The best of times Put a feature flag on it! big_feature: false Configuration file .search-results - if toggled_on? big_feature / new view for new feature / existing view UI implementation How are feature flags supposed to work?
17. big_feature: false The best of times Put a feature flag on it! Configuration file UI + UI What the user sees - if toggled_on? big_feature / new view for new feature / existing view UI implementation How are feature flags supposed to work?
18. The best of times Put a feature flag on it! • But it’s a lot of work. • And I’m worried about deploying months of work all at once. • And we’d really like to see how it performs with a small subset of users. • Beyond that, there are lots of external dependencies before we can roll it out to everyone. How are feature flags supposed to work?
19. • Finer-grained toggling (admin-only, language-specific, location-specific) • A/B testing! • Graceful degradation when a feature fails The best of times And that’s just the beginning. How are feature flags supposed to work?
20. It was the best of times, It was the worst of times. November 16
21. Feature flags in practice (They always start with the best of intentions). One clearly-defined feature behind a single flag.
22. Dependent feature flags Good intentions don’t always align with IRL practices. The project’s changed: let’s release different parts to different users!
23. Dependent feature flags The project’s changed: let’s release different parts to different users! Good intentions don’t always align with IRL practices.
24. Dependent feature flags Good intentions don’t always align with IRL practices. Flags on flags on flags
25. Dependent feature flags Segmented releases lead to exponential testing complexity what the user sees T T Both features T F F T F F Good intentions don’t always align with IRL practices.
26. Dependent feature flags Segmented releases lead to exponential testing complexity what the user sees T T Both features T F only F T F F Good intentions don’t always align with IRL practices.
27. Dependent feature flags Segmented releases lead to exponential testing complexity what the user sees T T Both features T F only F T ¯\_(ツ)_/¯ F F ¯\_(ツ)_/¯ Good intentions don’t always align with IRL practices.
28. Dependent feature flags ¯\_(ツ)_/¯ Segmented releases lead to exponential testing complexity Good intentions don’t always align with IRL practices.
29. Dependent feature flags Stick to one flag per feature (Or get ready for a manual QA headache)
30. def search_results # existing pre-flag logic if # append new results end end Where to flag? Different developers and teams have different strategies for implementing flags. Feature flags on the backend
31. Where to flag? Different developers and teams have different strategies for implementing flags. Lack of strategy leads to a really beautiful mess.
32. Where to flag? Different developers and teams have different strategies for implementing flags. Removing old flags is hard.
33. Where to flag? Different developers and teams have different strategies for implementing flags. Removing old flags is hard.
34. Where to flag? Coupled flags quickly become institutionalized as kill switches Different developers and teams have different strategies for implementing flags.
35. Where to flag? Different developers and teams have different strategies for implementing flags. Stack Overflow’s read-only mode
36. Where to flag? Isolate the feature flag
37. Where to flag? Treat flagged code as new code
38. features: active_feature: true nonactive_feature: false legacy_homepage: false : true more_payment_types: true another_feature: false Long-lived feature flags What happens when flags outlive their utility? Never-ending configuration
39. Long-lived feature flags Unattended flags are tech debt What happens when flags outlive their utility?
40. Long-lived feature flags What happens when flags outlive their utility? Unattended flags are high-risk flags
41. Long-lived feature flags What happens when flags outlive their utility? Unattended flags are high-risk flags
42. Long-lived feature flags Dashboards lack delegation mechanisms Dashboards make the issue more visible but aren’t a forcing function.
43. features: active_feature: true nonactive_feature: false legacy_homepage: false : true Long-lived feature flags Recall a typical config file… Dashboards lack delegation mechanisms
44. features: : enabled: true deadline: 12-10-2015 maintainer: rsliter Long-lived feature flags Delegate deadlines and maintainers through configuration A better way
45. A better way Long-lived feature flags if .past_deadline? # check out a new branch # amend file # cc maintainer in the commit # push new branch Delegate deadlines and maintainers through configuration
46. Long-lived feature flags Define lifespan for each flag Be ruthless about deleting them afterwards
47. Feature Flags introduce technical debt November 16
48. Feature Flags enable teams to release better-quality code faster November 16
49. Feature Flags enable teams to release better-quality code faster *When done right * November 16
50. Thank you. Questions?
51. Credits • CC image courtesy of Jared Tarbell on Flickr • CC image, Tangle, courtesy of loumurphy on Flickr • CC image, Arrows, courtesy of Byron on Flickr • Le Jour ni l’Heure 5941 : Cy Twombly, 1928-2011 • CC image, Volcano • Flip screenshot from Github • Flipper screenshot from Github • Dog gif courtesy Giphy • Error By Knight Capital Rips through Stock Market • Knight’s bizarre trades rattle markets • Knightmare: a Devops Cautionary Tale • Knight Capital Group Thank you!!!