Lyft iOS tooling infrastructure summary

A short presentation about Lyft's iOS tooling infrastructure presented at an informal meetup.

Transcript

2. Scale Code • 9 apps • 100% Swift • App code: 22k files, 1.3m loc • Test code: 4k files, 300k loc People • 100 committers per month • 320 commits per week
3. Continuous Integration Hardware • 100 Intel Mac minis, many AWS Linux containers • Macs are Bare Metal, managed with ARD Software • Buildkite, generated pipelines for many concurrent jobs • Xcode upgrades performed ASAP, single version supported • macOS upgrades performed as needed for Xcode
4. Build Speed CI (Build + Test) • P50: 8 min • P95: 15 min Local (Build) • P50: 89-121 sec • P95: 205-621 sec
5. Tooling • IDE: Xcode • Build System: Bazel • Dependency Management: Bazel + Artifactory • Caching: Currently Google RBE, evaluating others • Testing: XCTest & XCUITest
6. Recent Updates SwiftUI Previews • Works with our Bazel Xcode integration Snapshot Tests • Checked with CI, a command can accept changes in PR CLI • Migrating away from our Makefile based solution Remote Execution • Evaluated a lot of solutions, finalizing soon • Up to 50% faster incremental builds
7. Pain Points Growing codebase
8. Pain Points Growing codebase • Builds times increasing • More longer running tests (e.g. XCUITest) Bazel Xcode integration • “Not quite the same” (i.e. Run Script Phase) • Can only run a single test target • Flakiness (e.g. Indexing, debugging, profiling, etc.)
9. Future plans Remote Execution • Repurpose about 66% of minis for runners Apple Silicon • Ensure Bazel fully works, try not to have 2x cache usage Improved Bazel Xcode integration • Possibly leveraging XCBBuildServiceProxy Use managed hardware • Possibly AWS EC2 minis