Struct WorkerSpec 

pub struct WorkerSpec {

    pub url: String,
    pub models: WorkerModels,
    pub worker_type: WorkerType,
    pub connection_mode: ConnectionMode,
    pub runtime_type: RuntimeType,
    pub provider: Option<ProviderType>,
    pub labels: HashMap<String, String>,
    pub priority: u32,
    pub cost: f32,
    pub api_key: Option<String>,
    pub bootstrap_port: Option<u16>,
    pub bootstrap_host: String,
    pub dp_base_url: Option<String>,
    pub dp_rank: Option<usize>,
    pub dp_size: Option<usize>,
    pub kv_connector: Option<String>,
    pub kv_role: Option<String>,
    pub kv_engine_id: Option<String>,
    pub kv_block_size: Option<usize>,
    pub health: HealthCheckUpdate,
    pub http_pool: HttpPoolConfig,
    pub resilience: ResilienceUpdate,
    pub max_connection_attempts: u32,
    pub load_monitor_interval_secs: Option<u64>,
    pub multimodal_tensor_transport: Option<TransportMode>,
    pub multimodal_shm_min_bytes: Option<usize>,
    pub zmq_handshake_address: Option<String>,
    pub overload: OverloadUpdate,
}

Core worker identity and configuration.

The single canonical representation of “what is a worker”. Used as the shared sub-struct across API requests, API responses, and internal runtime state via #[serde(flatten)].

Fields use #[serde(default)] so the same struct works for both input (partial config from user) and output (fully resolved state).

Fields

url: String

Worker URL.

models: WorkerModels

Models this worker can serve.

worker_type: WorkerType

Worker type: regular, prefill, or decode.

connection_mode: ConnectionMode

Connection mode: http or grpc.

runtime_type: RuntimeType

Runtime type: sglang, vllm, trtllm, or external.

provider: Option<ProviderType>

External provider for API transformations. None means native/passthrough.

labels: HashMap<String, String>

Additional labels/tags.

priority: u32

Worker priority (higher = preferred).

cost: f32

Worker cost factor (baseline = 1.0).

api_key: Option<String>

Worker API key. Accepted on input, never included in responses.

bootstrap_port: Option<u16>

Bootstrap port for prefill workers in PD disaggregated mode.

bootstrap_host: String

Bootstrap hostname (derived from URL at construction time).

dp_base_url: Option<String>

Base URL without DP rank suffix (for DP-aware workers). When set, url contains the rank-suffixed form ({base}@{rank}).

dp_rank: Option<usize>

Data-parallel rank (None = not DP-aware).

dp_size: Option<usize>

Total data-parallel group size (None = not DP-aware).

kv_connector: Option<String>

KV connector type (e.g. “MooncakeConnector”, “NixlConnector”).

kv_role: Option<String>

KV role (e.g. “kv_producer”, “kv_consumer”, “kv_both”).

kv_engine_id: Option<String>

KV transfer engine id (vLLM kv_transfer_config.engine_id; Mooncake PD).

kv_block_size: Option<usize>

KV cache block size (tokens per block) for event-driven routing. When set, overrides the router-level default for this worker’s model. Typically matches the backend engine’s page size (e.g. 16 for SGLang).

health: HealthCheckUpdate

Per-worker health check overrides (partial — only Some fields override router defaults).

http_pool: HttpPoolConfig

Per-worker HTTP connection pool overrides.

resilience: ResilienceUpdate

Per-worker resilience overrides (retry + circuit breaker).

max_connection_attempts: u32

Maximum connection attempts during worker registration (default: 20).

load_monitor_interval_secs: Option<u64>

Per-worker load monitor interval override (seconds). When set, workers in the same group use this interval for load polling. Falls back to the global load_monitor_interval_secs from router config.

multimodal_tensor_transport: Option<TransportMode>

Per-worker multimodal tensor transport override (inline | shm | auto). Overrides the router-level multimodal_tensor_transport for this worker (e.g. force shm for a co-located worker, inline for a remote one).

multimodal_shm_min_bytes: Option<usize>

Per-worker minimum multimodal tensor size (bytes) before the SHM transport is used. Overrides the router-level multimodal_shm_min_bytes.

zmq_handshake_address: Option<String>

Handshake address the frontend BINDS for this worker’s engine handshake. Only meaningful for connection_mode: zmq; must be a tcp:// address. When unset, the frontend derives a deterministic port from the worker’s ipc:// path instead — this override exists for engines that dial a fixed, pre-agreed address rather than the derived one.

overload: OverloadUpdate

Per-worker absolute overload threshold overrides (partial — only Some fields override gateway defaults). Either field set enables overload protection for this worker even when the gateway leaves it off.

Implementations

impl WorkerSpec

pub fn new(url: impl Into<String>) -> Self

Create a new WorkerSpec with the given URL and sensible defaults.

Trait Implementations

impl Clone for WorkerSpec

fn clone(&self) -> WorkerSpec

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for WorkerSpec

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for WorkerSpec

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl JsonSchema for WorkerSpec

fn schema_name() -> Cow<'static, str>

The name of the generated JSON Schema.

fn schema_id() -> Cow<'static, str>

Returns a string that uniquely identifies the schema produced by this type.

fn json_schema(generator: &mut SchemaGenerator) -> Schema

Generates a JSON Schema for this type.

fn inline_schema() -> bool

Whether JSON Schemas generated for this type should be included directly in parent schemas, rather than being re-used where possible using the $ref keyword.

impl Serialize for WorkerSpec

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

where __S: Serializer,

Serialize this value into the given Serde serializer.