element-web/docs/feature-flags.md
2019-11-08 15:57:46 +00:00

4.6 KiB

Feature flags

When developing new features for Riot, we use feature flags to give us more flexibility and control over when and where those features are enabled.

For example, flags make the following things possible:

  • Extended testing of a feature via labs on develop
  • Enabling features when ready instead of the first moment the code is released
  • Testing a feature with a specific set of users (by enabling only on a specific Riot instance)

The size of the feature controlled by a feature flag may vary widely: it could be a large project like reactions or a smaller change to an existing algorithm. A large project might use several feature flags if it's useful to control the deployment of different portions independently.

Everyone involved in a feature (engineering, design, product, reviewers) should think about its deployment plan up front as best as possible so we can have the right feature flags in place from the start.

Interaction with spec process

Historically, we have often used feature flags to guard client features that depend on unstable spec features. Unfortunately, there was never clear agreement about how long such a flag should live for, when it should be removed, etc.

Under the new spec process, server-side unstable features can be used by clients and enabled by default as long as clients commit to doing the associated clean up work once a feature stabilises.

Starting work on a feature

When starting work on a feature, we should create a matching feature flag:

  1. Add a new setting of the form:
    "feature_cats": {
        isFeature: true,
        displayName: _td("Adds cats everywhere"),
        supportedLevels: LEVELS_FEATURE,
        default: false,
    },
  1. Check whether the feature is enabled as appropriate:
    SettingsStore.isFeatureEnabled("feature_cats")
  1. Add the feature to the set of labs on develop:
    "features": {
        "feature_cats": "labs"
    },
  1. Document the feature in the labs documentation

With these steps completed, the feature is disabled by default, but can be enabled on develop by interested users for testing.

Different features may have different deployment plans for when to enable where. The following lists a few common options.

Enabling by default on develop

Set the feature to enable in the develop config:

    "features": {
        "feature_cats": "enable"
    },

Enabling by default on staging and app

Set the feature to enable in the app config.

Feature deployed successfully

Once we're confident that a feature is working well, we should remove the flag:

  1. Remove the setting
  2. Remove all isFeatureEnabled lines that test for the feature's setting
  3. Remove the feature from the labs documentation
  4. Remove feature state from develop and app configs
  5. Celebrate! 🥳

Convert to a regular setting (optional)

Sometimes we decide a feature should always be user-controllable as a setting even after it has been fully deployed. In that case, we would craft a new, regular setting:

  1. Remove the feature flag from settings and add a regular setting with the appropriate levels for your feature
  2. Replace the isFeatureEnabled lines with getValue or similar calls according to the settings docs (checking carefully, as we may want a different mix of code paths when the feature is always present but gated by a setting)
  3. Remove the feature from the labs documentation
  4. Remove feature state from develop and app configs