Feature Request: Rule-level configurable stickiness / rollout key #5859
Replies: 7 comments 1 reply
|
Hey @pratik0907 and @parthagarwal4 I haven't seen this feature in other providers. Are there any good real-world examples of it being used? Is there any reason why |
|
Hey @erka As far as I know this is currently supported by LD. This is particularly helpful if I wanna run multiple experiments on different variants of a single flag and use different bucketing strategies per rule. For example, testing |
|
Hm... it sounds like a segment could solve that. Do I miss something? |
|
Segments help with deciding who is eligible for a rule, but this is more about how users are bucketed for percentage rollouts. For example:
In some cases, we need rollout consistency at the organization level/ site level or anyther other entity instead of user level. Example:
|
|
@erka thanks for the promt reply , Ill try to explain my usecase with a example Consider an application with two user types:
We want to run two experiments within the same feature flag. Rule 1: Organization-level rollout
Requirement: all employees in the same organization must see the same version. Problem with user-level stickiness:
This creates inconsistent experiences within the same org. Required behavior:
This requires bucketing by organization_id. Rule 2: Individual user beta rollout
Here, users should be evaluated independently using user_id. Key insight: So different rules in the same flag may require different bucketing keys:
|
|
Sounds good. Will try this. Thanks! |


Uh oh!
There was an error while loading. Please reload this page.
Hi Team,
I wanted to propose support for configurable stickiness keys at the rule level, similar to how popular feature flag services support percentage rollouts over a selectable attribute.
Current Behavior
Currently, all rules within a flag use the same entity_id + flag_key combination for consistent hashing and percentage rollouts.
Proposed Behavior:
Allow each rule to specify its own stickiness field(s) from the evaluation context for hashing / bucketing purposes.
Example attributes:
Why this is useful
Example use cases:
Example configuration idea:
This would make rollout behavior significantly more flexible while preserving deterministic assignments.
Thanks!
All reactions