Mobile ci cd overview

Mobile CI/CD Overview

Instructions

A mobile CI/CD pipeline reliably turns a commit into a signed, store-ready artifact. Builds are slow, signing is finicky, and store APIs are rate-limited. A pragmatic pipeline invests in caching, parallelism, and guard rails more than in clever automation.

1. Pipeline Shape

Commit -> PR checks (lint, unit, integration)
       -> Build artifact (debug)
       -> UI smoke on emulator
       -> Merge to main
       -> Build artifact (release, signed)
       -> Upload to internal/TestFlight/Play Internal
       -> Optional: promote to production with staged rollout

Keep PR lane fast (under 10 minutes). Promote heavy work (device cloud, E2E, size analysis) to the main lane.

2. Branching

• Trunk-based with short-lived branches works best for mobile. Long release branches invite merge pain and diverging metadata.
• Release branches only when multiple versions must be supported simultaneously (enterprise distribution).
• Tag every production release on the merge commit (v2.5.0).
• Protect main with required checks, linear history, and signed commits.

3. Build Matrices

• iOS: one macOS runner per job. Match Xcode version to the project's required_xcode_version. Keep SDK up to date.
• Android: Linux runners; use cached Gradle caches and AVD images.
• Flutter: matrix of iOS + Android builds; share analyzer/test job.
• React Native: similar matrix; use Hermes; cache Metro.
• Avoid matrices that fan out to 20 jobs for a PR. Most issues repro on one target.

4. Caching

Caching is where mobile pipelines win or lose.

• Gradle: cache ~/.gradle/caches, ~/.gradle/wrapper, keyed on **/*.gradle*, **/gradle-wrapper.properties, gradle/libs.versions.toml.
• CocoaPods: cache Pods and ~/Library/Caches/CocoaPods, keyed on Podfile.lock.
• SwiftPM: cache ~/Library/Developer/Xcode/DerivedData/*/SourcePackages, keyed on Package.resolved.
• Yarn/npm: cache ~/.yarn/cache or ~/.npm keyed on lockfile. Enable Yarn zero-install if appropriate.
• Dart/pub: cache ~/.pub-cache, keyed on pubspec.lock.
• Xcode DerivedData: cache selectively; often cheaper to rebuild than to invalidate.

5. Signing

• Keep signing assets out of the repo. Use CI secrets (.p12, keystore, provisioning profiles, App Store Connect API key JSON).
• Prefer App Store Connect API keys over Apple ID-based altool. They are session-safe and scriptable.
• Prefer signing by Google Play (upload key in CI, app signing key with Google).
• Use fastlane match or equivalent to centralize provisioning profiles and certs for iOS across engineers and CI.

# GitHub Actions snippet - iOS signing
- name: Import certs
  run: |
    echo "$IOS_DIST_CERT_P12_BASE64" | base64 -d > cert.p12
    security create-keychain -p "" build.keychain
    security import cert.p12 -k build.keychain -P "$IOS_DIST_CERT_PASSWORD" -T /usr/bin/codesign
- name: Download provisioning profile
  run: echo "$IOS_PROVISIONING_PROFILE_BASE64" | base64 -d > app.mobileprovision

6. Artifacts

• Upload debug builds to Firebase App Distribution or TestFlight internal for quick QA.
• Store .ipa, .aab, mapping.txt (Android), dSYM (iOS) for every release with 90-day retention at minimum.
• dSYM and mapping files must be uploaded to the crash reporter on every release; without them stack traces are unreadable.

7. Store Upload

• Fastlane: pilot (TestFlight), supply (Play). Battle-tested.
• Gradle Play Publisher: Play upload via Gradle tasks.
• Xcode Cloud: first-class Apple pipeline; good for iOS-only teams.
• Codemagic / Bitrise / AppCircle: mobile-focused hosted CI with prebuilt store steps.

Automate internal/TestFlight on every main merge. Automate production only when staged rollout and halt criteria are owned by the pipeline.

8. Environments and Flavors

• Use build flavors / schemes / productFlavors to separate dev, staging, prod bundles with distinct bundle IDs.
• Separate Firebase / analytics projects per flavor. Never mix staging and prod analytics.
• Secrets come from the CI environment, not from committed files. Use gradle.properties readers or xcconfig with CI injection.

9. Versioning

• Version code / build number must strictly increase. Compute from commit count (git rev-list --count HEAD) or CI run number.
• Version name from a tag or package.json/pubspec.yaml/Info.plist/build.gradle.
• Fail the build if version code is not greater than the last uploaded artifact.

10. Observability for the Pipeline

• Track build duration, cache hit rate, test duration, and failure reasons.
• Alert when PR lane exceeds a budget (e.g., 15 minutes) consistently.
• Keep a rolling dashboard of flaky jobs; quarantine or fix.

11. Anti-Patterns

• Self-hosted runners without auto-scaling. Either accept queueing or invest in elasticity.
• Signing secrets stored in plain repo files, even in private repos.
• Single giant job that does lint, build, test, and upload. Split so failures are isolated.
• Version number bumped manually. Automate.
• No dSYM/mapping upload step. Crashes will be unreadable and ownership lost.
• Running real device tests on every PR. Sample on PR, full matrix on main.

Checklist

• Trunk-based branching with required PR checks.
• PR lane runs under 10 minutes with caches warm.
• Release lane signs artifacts using secrets from CI storage.
• dSYMs and mapping files uploaded to the crash reporter per release.
• Store uploads automated to internal/TestFlight on main merge.
• Version code/build number auto-computed and monotonic.
• Separate flavors for dev/staging/prod with isolated analytics.
• Pipeline metrics (duration, flake rate) are monitored.
• Cache keys include the lockfiles that govern their invalidation.