Use Murmurhash algorithm for the phased rollout

[rainerschoe]

https://ibm-quantumcomputing.slack.com/archives/C07LYBYEFCH/p1732591598349079

let hashValue;
hashValue = murmurhashV3(entityID + ':' + featureID); // outputs a integer from 1 to 100

if (hashValue < rolloutPercentage) {
  // serve featureflag's enabled value
} else {
  // serve featureflag's disabled value
}

Example implementation:

• Node: https://github.com/IBM/appconfiguration-node-sdk/blob/master/lib/configurations/ConfigurationHandler.js#L576
• Go: https://github.com/IBM/appconfiguration-go-sdk/blob/master/lib/internal/models/Feature.go#L191

Some info:

• murmurhash produces the same input for a given string. When a entityID gets evaluated n-number of times, all the time it gets the same featureflag’s value
• we would need a large sample to test the partial rollout. Atleast 100 samples(100 unique entityIDs). Assume, featureflag rollout percentage is set to 30%. When we pass 100 different entityIDs, and check the evaluated value. The total count which received enabled value will be 30 & disabled value will be 70.

[jgsogo]

Is this the one we need? https://docs.rs/fasthash/latest/fasthash/murmur3/index.html

[jgsogo]

Maybe we are already using the right algorithm, although we cannot be 100% sure until we have some testing strategy that validates the outcome for all the clients in all the languages (see #11 )

[rainerschoe]

Some input from Sai on this:

Algorithm:

let hashValue;
hashValue = murmurhashV3(entityID + ':' + featureID); // outputs a integer from 1 to 100

if (hashValue < rolloutPercentage) {
  // serve featureflag's enabled value
} else {
  // serve featureflag's disabled value
}

Example implementation:
Node: https://github.com/IBM/appconfiguration-node-sdk/blob/master/lib/configurations/ConfigurationHandler.js#L576
Go: https://github.com/IBM/appconfiguration-go-sdk/blob/master/lib/internal/models/Feature.go#L191