Fix part of #6106: Add Release Playbook wiki page #6352
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,153 @@ | ||
| # Release Playbook | ||
|
|
||
| This page is the step-by-step coordinator guide for Oppia Android releases using the automated | ||
| release pipeline. For conceptual background on any step, see the | ||
| [App and Feature Release Process](app-and-feature-release-process.md) wiki page. For manual | ||
| fallback procedures when automation fails, see the | ||
| [In-Depth Release Reference](In-Depth-Release-Reference.md). | ||
|
|
||
| --- | ||
|
|
||
| ## Table of Contents | ||
|
|
||
| 1. [Weekly alpha release](#1-weekly-alpha-release) | ||
| 2. [Changelog review](#2-changelog-review) | ||
| 3. [Production release](#3-production-release) | ||
| 4. [Staged rollout](#4-staged-rollout) | ||
| 5. [Weekly coordinator checklist](#5-weekly-coordinator-checklist) | ||
|
|
||
| --- | ||
|
|
||
| ## 1. Weekly Alpha Release | ||
|
|
||
| The **Auto Release Alpha** workflow, defined by | ||
| [`auto_release_alpha.yml`](../.github/workflows/auto_release_alpha.yml), fires automatically every | ||
| **Tuesday at 03:30 UTC**. It finds the latest passing commit on `develop`, tags it as | ||
| `latest-alpha`, and dispatches **Build and Sign**. The coordinator's job starts at the approval | ||
| gate. | ||
|
|
||
| ### Steps | ||
|
|
||
| - [ ] **Approve the build** — Navigate to Actions → **Build and Sign** run → approve in the | ||
| `oppia-android-release-env` gate. | ||
| - [ ] **Copy the GCS path** — After the build completes, copy the signed AAB path from the job | ||
| summary (format: `gs://oppia-android-alpha-releases/…/*.aab`), which will be used for the | ||
| deployment steps. | ||
| - [ ] **Distribute to QA** — Trigger **Deploy to Firebase** (`deploy_to_firebase.yml`) with the | ||
| GCS path above and approve the run in the `oppia-android-release-env` gate. | ||
| - [ ] **Await QA sign-off** — Notify the alpha tester group and wait for their confirmation that | ||
| the alpha build is stable and ready for Play Store deployment. | ||
| - [ ] **Deploy to Play Console** — After QA sign-off, trigger **Deploy to Play Console** | ||
| (`deploy_to_play_console.yml`) with `track=alpha` and `rollout_fraction=1000` (100% of the | ||
| alpha track). The Firebase and Play Console deployments are sequential: QA on Firebase is | ||
| completed first, then the approved build is pushed to the alpha track: | ||
|
|
||
|  | ||
|
|
||
| > If **Auto Release Alpha** exits with an error or finds no passing commit, see the | ||
| > [In-Depth Release Reference](In-Depth-Release-Reference.md) for manual fallback steps. | ||
|
|
||
| --- | ||
|
|
||
| ## 2. Changelog Review | ||
|
|
||
| The **Generate Changelogs** workflow | ||
|
Member
There was a problem hiding this comment. I understand that this is supposed to trigger when version.bzl is updated, but when does that latter event happen? Can you explain? Also, is the release coordinator ever supposed to run this manually? If so, in what circumstances? Please explain.
Collaborator
Author
There was a problem hiding this comment.
|
||
| ([`generate_changelog.yml`](../.github/workflows/generate_changelog.yml)) opens a PR to | ||
| `develop` automatically when `version.bzl` is updated with a new `MINOR_VERSION`. This update | ||
| happens at the start of each release cycle when the RC bumps `MINOR_VERSION` as part of the | ||
| pre-release step in §3. The workflow can also be triggered manually via `workflow_dispatch` if | ||
| the auto-trigger did not fire (e.g. the version bump happened without triggering CI) or if the | ||
| changelog needs to be regenerated after corrections. | ||
|
|
||
| ### Steps | ||
|
|
||
| - [ ] Review the AI-generated release notes in `config/changelogs/{version}.md` (and any | ||
| flavor-specific overrides such as `{version}_alpha.md` or `{version}_beta.md`). | ||
| - [ ] Edit the release notes if the AI summary needs adjustment before merging. | ||
| - [ ] Merge the changelog PR into `develop`. Merging automatically triggers **Deploy Updated | ||
| Changelog** (`deploy_updated_changelog.yml`), which uploads the release notes to Play Console. | ||
|
|
||
| --- | ||
|
|
||
| ## 3. Production Release | ||
|
|
||
| Follow this checklist for a full beta or production (GA) release. | ||
|
|
||
| ### Pre-release | ||
|
|
||
| - [ ] Bump `MINOR_VERSION` in `version.bzl`. This push triggers **Generate Changelogs** | ||
|
Member
There was a problem hiding this comment. Link "Generate Changelogs" to the section above. Ditto below. |
||
| automatically. | ||
| - [ ] Review and merge the changelog PR opened by **Generate Changelogs** (§2). **Do not cut | ||
| the release branch until this PR is merged** — the release branch must include the correct | ||
| version number and changelog. | ||
| - [ ] Cut the release branch from `develop` HEAD: | ||
| ``` | ||
| git checkout -b release-0.X upstream/develop | ||
| git push upstream release-0.X | ||
| ``` | ||
|
|
||
| ### Build & sign | ||
|
|
||
| - [ ] Trigger **Build and Sign** (`build_and_sign.yml`) with `flavor=beta` or `flavor=ga` and | ||
| `source_ref=release-0.X`: | ||
|
|
||
|  | ||
|
|
||
| - [ ] Approve the run in the `oppia-android-release-env` gate. | ||
| - [ ] Copy the signed AAB GCS path from the job summary. | ||
|
|
||
| ### QA | ||
|
|
||
| - [ ] Trigger **Deploy to Firebase** (`deploy_to_firebase.yml`) with the GCS path and approve | ||
| the run in the `oppia-android-release-env` gate. | ||
| - [ ] Notify QA testers and await sign-off. | ||
| - [ ] If critical issues are found during QA: land the fix on `develop`, cherry-pick to | ||
| `release-0.X`, then rebuild from Build & sign above. | ||
|
|
||
| ### Deploy | ||
|
|
||
| - [ ] Trigger **Deploy to Play Console** (`deploy_to_play_console.yml`) with `track=beta` or | ||
| `track=production` and `rollout_fraction=100` (10% initial rollout — see §4): | ||
|
|
||
|  | ||
|
|
||
| --- | ||
|
|
||
| ## 4. Staged Rollout | ||
|
|
||
| After the initial deployment, progressively increase the rollout using `update_rollout.yml`. | ||
| Monitor Firebase Crashlytics between each step; halt and investigate if crash rates spike. | ||
|
|
||
| | Day | `rollout_fraction` | Actual rollout | Action | | ||
|
Member
There was a problem hiding this comment. These look like integers rather than "fractions"?
Collaborator
Author
There was a problem hiding this comment. They are in permille (thousandths out of 1000), so 100 = 10%, 1000 = 100%. Added a note below the table to clarify the unit.
Member
There was a problem hiding this comment. But then shouldn't the name be something like "rollout_permille", not "rollout_fraction"? The integer here is still not a fraction.
Collaborator
Author
There was a problem hiding this comment. rollout_fraction is already the established parameter name in update_rollout.yml (the workflow input, job name, and Bazel target are all named accordingly). Keeping the same name in the playbook table makes it easy for release coordinators to directly map the doc to the actual workflow input when triggering a run. PS: The note directly below the table already clarifies the permille semantics: "rollout_fraction is specified in permille (thousandths out of 1000) as required by the Play Console API — 100 means 10%, 1000 means 100%."
Member
There was a problem hiding this comment. Yes, I am actually suggesting that that name is misleading throughout and should be changed, not just here, but in the workflows. This should probably have been caught during earlier review. If a variable does not actually represent a fraction, it should not be called a fraction. Could you please look into this? Is there an external constraint that forces us to keep the incorrect naming?
Collaborator
Author
There was a problem hiding this comment. 
I understand. I had already added a sentence with an example explaining what the number represents. I think this makes values like 1000 or 100 less ambiguous, since readers can refer to the example to understand what the value means.
Member
There was a problem hiding this comment. @Sandesh282 We are talking about two different things here. The message for the users of the feature is good, but the problem I am discussing is the name of the variable in the code, which I still think should be fixed to reflect the type of value that it stores.
Contributor
There was a problem hiding this comment. @Sandesh282, after further consideration, I recommend:
This is mostly a mechanical refactor. My search found 20 files containing the affected naming: Implementation Tests Build definitions Documentation This is about 20 files total. You can change those in this PR. |
||
| |---|---|---|---| | ||
| | 0 | `100` | 10% | Initial deploy via **Deploy to Play Console** | | ||
| | 1 | `250` | 25% | Trigger **Update Rollout** after monitoring | | ||
| | 3 | `500` | 50% | Trigger **Update Rollout** | | ||
| | 7+ | `1000` | 100% | Trigger **Update Rollout** for full rollout | | ||
|
|
||
| > `rollout_fraction` is specified in **permille** (thousandths out of 1000) as required by the | ||
| > Play Console API — `100` means 10%, `1000` means 100%. | ||
|
|
||
| **Inputs for `update_rollout.yml`:** | ||
|
|
||
| | Input | Value | | ||
| |---|---| | ||
| | `track` | `beta` or `production` | | ||
| | `version` | e.g. `0.18` | | ||
| | `rollout_fraction` | New fraction from the table above | | ||
|
|
||
| --- | ||
|
|
||
| ## 5. Weekly Coordinator Checklist | ||
|
|
||
| Run through this checklist each week: | ||
|
|
||
| - [ ] Check whether **Auto Release Alpha** completed successfully on Tuesday. If not, | ||
| consult the [In-Depth Release Reference](In-Depth-Release-Reference.md). | ||
| - [ ] Check whether **Pull Latest Lesson Versions** opened a PR on Monday. If so, review | ||
| and merge the lesson-versions PR. | ||
| - [ ] Check for any pending changelog PRs opened by **Generate Changelogs** and merge them. | ||
| - [ ] Review crash rates in Firebase Crashlytics and advance the rollout fraction if stable | ||
| (see §4). | ||
|
|
||
| --- | ||
Uh oh!
There was an error while loading. Please reload this page.