Rollout Ramps
A rollout ramp raises a percentage rollout for you, one step at a time. You give it a ladder, say 1% for an hour, 5% for four hours, 25% and 50% for a day each, then 100%. Featureflip applies the first step right away and the next one each time a hold runs out. Nobody has to come back to the dashboard at 2am to move a slider.
A ramp only ever changes the rollout percentage. The salt and the Bucket by attribute stay fixed, so every step adds users to the cohort and nobody who already has the feature loses it. See Rollout Strategies for how the bucketing works.
Ramps are set per rule, per environment. A ramp on Production leaves Staging alone.
What a ramp can drive
Section titled “What a ramp can drive”- A targeting rule with a percentage rollout below 100%. The rule keeps its conditions and its variations. The ramp moves its percentage.
- The default rule. This is the rollout on the fallthrough, served to everyone no rule matched. If the default rule has no rollout yet, the ramp sets one up: you pick the variation to roll out and the attribute to bucket by.
Each rule and the default rule can have one active ramp at a time. A rule ramp and a default-rule ramp on the same flag run independently.
Starting a ramp
Section titled “Starting a ramp”- Open the flag and pick the environment.
- On a rollout rule’s card, click the Ramp rollout icon. For the default rule, click Ramp rollout in the Default rule rollout card.
- Add your steps. Each step is a percentage and a hold, the time to wait before the next step. The last step has no hold.
- Click Start ramp.
The dialog offers a preset with the ladder from the top of this page. It only loads the rungs above where the rollout is now, so a rule at 10% gets 25 → 50 → 100.
The step rules:
- 1 to 20 steps
- Percentages are whole numbers that go up from one step to the next, from 1 to 100
- The first step has to be at or above the current percentage
- Each hold runs from 5 minutes to 30 days, in whole minutes
The flag has to be on in that environment to start a ramp, and an archived flag can’t be ramped at all.
Finishing at 100% on the default rule
Section titled “Finishing at 100% on the default rule”When a default-rule ramp reaches a final step of 100%, the rollout is folded away. The ramped variation becomes the default variation, exactly as if you had set it by hand. Contexts that don’t carry the Bucket by attribute were getting the old default all along. At this point they switch to the new variation too, and the dialog says so before you start.
With the rollout gone, the flag serves one fixed variation again. That is the Fully rolled out signal stale flag detection watches for, so after 30 days the flag shows up as Stale, ready to be removed from your code.
Watching a ramp
Section titled “Watching a ramp”While a ramp runs, the rule card shows where it is and what comes next:
Step 2 of 5 · 5% · next step to 25% in 3h 12mEach step is a normal rule update. It lands in the audit log with a reason like Rollout ramp step 3 of 5: 50%, sends a targeting_rule.updated event to your webhooks (flag.environment_config.updated for a default-rule ramp), and reaches connected SDKs over the stream within seconds. Starting, pausing, resuming and aborting the ramp are audited as Rollout ramp entries on the flag’s History tab.
The ramp sends webhook events of its own too: rollout_ramp.started, rollout_ramp.step_applied, rollout_ramp.paused, rollout_ramp.resumed, rollout_ramp.completed and rollout_ramp.aborted. A paused event carries the reason in changes, so an alert can tell a person pausing the ramp from Featureflip pausing it because the rule was edited or a step failed. See Rollout ramp events for the payload.
Pausing, resuming, advancing and aborting
Section titled “Pausing, resuming, advancing and aborting”The ramp’s card has the controls:
- Pause stops the clock. The rule stays at the current step and the rest of the hold is kept for later.
- Resume restarts the clock where it stopped. It continues from the rule as it stands, so if someone changed the percentage while the ramp was paused, the ramp picks up from there.
- Advance now applies the next step without waiting for the hold.
- Abort ends the ramp and sets the rule to a percentage you choose. You can keep the current percentage or lower it, all the way to 0%. Abort never raises the rollout.
In a critical environment that requires a change reason, each of these asks for one, the same as any other change there.
Why a ramp pauses itself
Section titled “Why a ramp pauses itself”A ramp stops as soon as a person or the system steps in, and the card tells you why.
| Shown on the card | What happened | What to do |
|---|---|---|
| Paused: the rule was edited | Someone saved the ramped rule by hand. The rule editor warns you before you save. On the default rule, changing the environment’s prerequisites, a scheduled configuration change and copying configuration in from another environment count as edits too. | Check the edit, then Resume from the rule as it now stands. |
| Paused: the flag was turned off | The flag went off in this environment, from the dashboard, the API, a flag trigger or a scheduled change. | Turn the flag back on, then Resume. The rest of the hold is still there. |
| Paused: a message from the server | The next step failed to apply. The ramp stops rather than retrying on its own, and the message says what went wrong. | Fix the cause it names, then Resume. |
A change freeze pauses a ramp too. When a step falls due inside a freeze, the ramp waits and shows that the freeze is holding it, then continues by itself when the freeze ends. If someone also paused the ramp, it stays paused until they resume it. While a freeze covers the environment, starting, resuming and advancing a ramp are refused. Pausing and aborting still work.
The kill-switch case is deliberate. A ramp never counts down its hold while the flag is off, so turning the flag back on doesn’t jump straight to the next step.
Things that end a ramp
Section titled “Things that end a ramp”Some changes remove the targeting rule a ramp drives, and the ramp ends with it. Each time, the flag’s History tab records that the rollout ramp ended, at which step, and why.
- Deleting the rule
- Replacing the flag’s whole set of rules, which a scheduled targeting change and copying configuration from another environment both do
- Deleting the flag or the environment
A default-rule ramp has no rule to lose, so only the last item ends it.
You can’t archive a flag while it has a running or paused ramp. Finish or abort the ramp first.
From the API or an agent
Section titled “From the API or an agent”The same ramps are on the Management API: list and get ramps under a flag’s environment, start one with POST …/rules/{ruleId}/ramp or POST …/default-rule/ramp, and pause, resume, advance or abort it with POST …/ramps/{rampId}/{action}. Each ramp comes with an ETag, and the four actions accept If-Match, so a change fails with 412 if the ramp moved on since you read it.
The MCP server wraps these as list_rollout_ramps and manage_rollout_ramp, so you can ask an agent to ramp a rule from 1% to 100% over a week, or to abort a ramp that is hurting your error rate.
Plans and downgrades
Section titled “Plans and downgrades”Ramps need the Team plan or above. Below Team, the Ramp rollout button doesn’t appear.
If your organization becomes read-only, for example after a downgrade, ramps that are already running keep going. You can still pause or abort them. Starting, resuming and advancing need the plan back.
Related
Section titled “Related”- Rollout Strategies: how percentage rollouts bucket users
- Targeting Rules: setting up the rollout a ramp drives
- Change Freezes: block changes to an environment for a window
- Scheduled Changes: turn a flag on or off at a set time
- Progressive rollouts: the step ladder and what to watch at each step
- Automated progressive rollouts: choosing steps and hold times for your traffic