Crate feature_flag 

feature-flag

Server-side feature flag evaluation for async Rust services.

Design rules

1. No external RNG. Sticky bucketing is a SHA-256 of (flag_id ‖ subject_id) mod 100 — deterministic, no entropy source, no rand dep.
2. Tokio is assumed. Hot reload uses tokio::fs + an async watcher.
3. Cheap clones. The evaluator is Arc-shaped under the hood, so handing it to many tasks is free.
4. Snapshot for tests. Build a FlagSet in code, drop it into a FlagEvaluator, evaluate. No I/O required.

Primitives

• FlagSet — declarative collection of flags, deserialised from JSON.
• Flag — one named flag with default + ordered targeting Rules.
• Rule — a match_when Predicate plus an outcome (Variant or percentage rollout).
• Predicate — small DSL of typed comparisons + all_of / any_of / not.
• Subject — the entity being evaluated (id + arbitrary attributes).
• FlagEvaluator — combines a FlagSet with a sticky-bucketing function.
• HotReloader — watches a JSON file for changes and atomically swaps the evaluator’s FlagSet.

Composes with

• reliability-toolkit-rs — wrap a rollout-controlled call in a circuit breaker + retry.
• slo-budget-tracker — if your SLO is burning, flip the flag to default_off from a remote config push.

Re-exports

pub use error::FeatureFlagError;

pub use evaluator::Evaluation;

pub use evaluator::FlagEvaluator;

pub use hot_reload::HotReloader;

pub use model::Flag;

pub use model::FlagSet;

pub use model::Rule;

pub use model::Variant;

pub use predicate::Comparison;

pub use predicate::Predicate;

pub use subject::Subject;

Modules

error

Crate-wide error type.

evaluator

The hot path. FlagEvaluator wraps a FlagSet (behind an ArcSwap-style lock) and resolves (flag_id, subject) to a variant.

hot_reload

File-backed hot reload for FlagEvaluator.

model

Flag / Rule / FlagSet types.

predicate

Targeting predicate DSL — typed comparisons + boolean combinators.

subject

The thing flags are evaluated against.