How do you release a dashboard change to only a few accounts?
Decide who should receive the same version. If people in one workspace must see the same dashboard, give the flag a workspace to evaluate. LaunchDarkly calls the person, account or device supplied to a rule a context.
A rule can match specific accounts or send a percentage of eligible contexts to a version. Changing the percentage can change which contexts receive a version. Check the kind of context used for the rollout and avoid treating a percentage as a handpicked tester list.
What is the difference between off, unmatched and unavailable?
Write down what should appear in all three cases. A dashboard can look fine for a tester matched by a rule and still surprise the next person who visits.
| Situation | What decides the version |
|---|---|
| Targeting is on and no rule matches | The flag’s default rule for that environment. |
| The flag is turned off | The configured off variation. If none is set, the app’s fallback. |
| The library cannot evaluate the flag | The fallback passed in the app code. An initialized library may continue using available stored rules during a disconnect. |
What should you check before flipping a switch in a test app?
Use a disposable app and a dedicated test flag. Before opening a screen that evaluates it, match the library’s key to the non-production environment and confirm that any data, messages and event records go to test destinations.
- Match the test flag and the test accountsIn the dashboard, select the same environment. Verify the practice flag and rule only apply to test accounts before changing or evaluating the flag. If the settings are unclear, inspect them or sketch the cases on paper.
- Try both branches and a missing valueHave one fictional account match the rule and another miss it. Then check the configured off version and the fallback from a fresh test instance without a usable flag state.
When should you use a rollout or an experiment?
A percentage rollout limits who gets the change. A progressive rollout raises that percentage over time. A guarded rollout can monitor selected metrics and react to a regression depending on how you configure it.
An experiment answers a different question, such as whether the dashboard helps more accounts finish setup. Choose the thing assigned to a version, the feature variations and the metric first. Check that the plan you are evaluating includes the release controls and measurement volume you need.
What can change the bill or make a move harder?
LaunchDarkly offers a Developer plan with usage limits. Its billing documentation distinguishes active contexts, service connections and experiment usage. Check which units your plan meters before planning a large rollout.
Keep the flag keys, targeting rules, fallback behavior and metric definitions in a form you can review. A switch to another platform also needs a way to preserve or intentionally change who sees each version. Check export availability and what your plan retains before ending the account.