Enum LoadBalancingAlgorithm 

pub enum LoadBalancingAlgorithm {
    RoundRobin,
    LeastConnections,
    Random,
    IpHash,
    Weighted,
    ConsistentHash,
    PowerOfTwoChoices,
    Adaptive,
    LeastTokensQueued,
    Maglev,
    LocalityAware,
    PeakEwma,
    DeterministicSubset,
    WeightedLeastConnections,
    Sticky,
}

Load balancing algorithm

Variants

RoundRobin

LeastConnections

Random

IpHash

Weighted

ConsistentHash

PowerOfTwoChoices

Adaptive

LeastTokensQueued

Least tokens queued - for inference/LLM workloads

Selects the upstream with the fewest estimated tokens currently being processed. Useful for LLM inference backends where token throughput varies significantly between requests.

Maglev

Maglev consistent hashing - Google’s load balancing algorithm

Provides minimal disruption when backend servers are added/removed, with better load distribution than traditional consistent hashing. Uses a permutation-based lookup table for O(1) selection.

LocalityAware

Locality-aware load balancing

Prefers targets in the same zone/region as the proxy, falling back to other zones when local targets are unhealthy or overloaded. Useful for multi-region deployments to minimize latency.

PeakEwma

Peak EWMA (Exponentially Weighted Moving Average)

Twitter Finagle’s algorithm that tracks latency using EWMA and selects the backend with the lowest predicted completion time. Reacts quickly to latency spikes by using the peak of EWMA and recent latency.

DeterministicSubset

Deterministic Subsetting

For very large clusters (1000+ backends), limits each proxy instance to a deterministic subset of backends. Reduces connection overhead while ensuring even distribution across all proxies.

WeightedLeastConnections

Weighted Least Connections

Combines weight with connection counting. Selects the backend with the lowest ratio of active connections to weight. Useful when backends have different capacities.

Sticky

Cookie-based sticky sessions

Routes requests to the same backend based on an affinity cookie. Falls back to a configurable algorithm when no cookie is present or the target is unavailable. Useful for stateful applications that require session affinity.

Trait Implementations

impl Clone for LoadBalancingAlgorithm

fn clone(&self) -> LoadBalancingAlgorithm

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for LoadBalancingAlgorithm

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for LoadBalancingAlgorithm

fn deserialize<__D>( __deserializer: __D, ) -> Result<LoadBalancingAlgorithm, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for LoadBalancingAlgorithm

fn eq(&self, other: &LoadBalancingAlgorithm) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Serialize for LoadBalancingAlgorithm

fn serialize<__S>( &self, __serializer: __S, ) -> Result<<__S as Serializer>::Ok, <__S as Serializer>::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl Copy for LoadBalancingAlgorithm

impl Eq for LoadBalancingAlgorithm

impl StructuralPartialEq for LoadBalancingAlgorithm