Struct SpawnConfig 

pub struct SpawnConfig {
    pub tenant_id: TenantId,
    pub deadline: Option<Duration>,
    pub args: Vec<WasmArg>,
    pub streaming: Option<StreamingContext>,
    pub input: Vec<u8>,
}

Configuration passed to TensorWasmExecutor::spawn_instance.

Fields

tenant_id: TenantId

Owning tenant.

deadline: Option<Duration>

Optional per-call deadline.

args: Vec<WasmArg>

Arguments forwarded to the first TensorWasmExecutor::call_export_with_args invocation against this instance.

Callers that drive a single spawn-then-call flow (CLI run, API /invoke) populate this field so the caller’s argument list survives the trip across crate boundaries without a parallel CallConfig. Multi-call flows should ignore this field and pass arguments directly to each call_export_with_args invocation.

streaming: Option<StreamingContext>

Optional streaming context (roadmap feature #2). When Some, TensorWasmExecutor::spawn_instance builds a wasmtime wasmtime::Linker that wires the wasi:tensor/host host functions (emit-chunk, flush) against this context — guest emits land on the matching mpsc::Receiver<Vec<u8>> the gateway is draining into the SSE / chunked HTTP response.

None means streaming is disabled: the spawn uses the historical empty-imports Instance::new_async path, and a guest that imports wasi:tensor/host will fail to link. /invoke (the synchronous route) takes the None path; only /invoke-stream opts in.

input: Vec<u8>

Bytes staged for the guest to pull via the wasi:tensor/host pull-model input channel (input-len / read-input).

Empty by default (the historical behaviour: the guest has no input channel). When non-empty, TensorWasmExecutor::spawn_instance installs an InputContext on the per-instance state and registers the input-len / read-input host functions on the spawn linker — so a guest can copy these bytes into its own linear memory at the start of the invocation. The OpenAI completions shim sets this to the assembled prompt bytes.

Implementations

impl SpawnConfig

pub fn for_tenant(tenant_id: TenantId) -> Self

Construct with just a tenant and no deadline.

pub fn with_deadline(self, deadline: Duration) -> Self

Add a deadline relative to “now at spawn time”.

pub fn with_args(self, args: Vec<WasmArg>) -> Self

Attach an argument list for the upcoming call. See SpawnConfig::args.

pub fn with_streaming(self, ctx: StreamingContext) -> Self

Attach a streaming context (roadmap feature #2). See SpawnConfig::streaming.

Builder method; pairs with Self::for_tenant / Self::with_deadline. The API gateway’s /invoke-stream route constructs an mpsc::channel, wraps the sender in a StreamingContext via StreamingContext::with_channel, and passes it here so the guest’s wasi:tensor/host.emit-chunk calls land on the matching receiver — which the gateway concurrently drains into the SSE / chunked response body.

pub fn with_input(self, input: Vec<u8>) -> Self

Stage input bytes for the guest to pull via the wasi:tensor/host input channel (input-len / read-input). See SpawnConfig::input.

Builder method; pairs with Self::for_tenant / Self::with_deadline / Self::with_streaming. An empty slice is a no-op (the default) — the guest then observes input-len() == 0. The OpenAI completions shim passes the assembled prompt bytes here so the guest receives the prompt.

Trait Implementations

impl Clone for SpawnConfig

fn clone(&self) -> SpawnConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SpawnConfig

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

Formats the value using the given formatter.