BinaryLibs.Wasm 0.163.0

Wasm worker hosting

WasmSupervisor manages one or more WasmWorker instances. A worker can load the existing in-process WasmRuntime, or it can own an injected IWasmWorkerProcess/ProcessStartInfo for a separate worker executable. IWasmProtocolTransport is used for the phase-4 control handshake and clean shutdown. WasmWorker reports lifecycle state, failures, heartbeat timeouts, unexpected process exits, deadline-based shutdown, and explicit restart or termination.

The library remains netstandard2.0, so it does not provide a universal OS-specific worker executable or a platform-neutral process IPC implementation. Applications supply the companion worker entry point and transport; the provided WasmProcessWorkerProcess is the concrete BCL process launcher where the host platform supports System.Diagnostics.Process. Invocation of a separate process must be implemented by that companion using the existing generic protocol. This package deliberately does not define domain-specific commands or payloads.

Phase 6 and 7: local callback queue and hardening model

The runtime and worker model now separates fast local callback traffic from the out-of-process supervisor transport. A worker may raise callback events that are not immediately forwarded over IPC; instead they are stored in an in-memory WasmLocalCallbackQueue as opaque WasmLocalOperationRecord entries. Each record carries a correlation identifier and raw payload bytes, so callback records can be exchanged later in a single flush without one IPC round trip per invocation.

The queue is intentionally bounded. WasmLocalCallbackOptions exposes a MaxQueueDepth and a WasmQueueOverflowPolicy so applications can choose the behavior that best fits the worker's latency and safety profile. The default policy is Throw, but callers may select DropOldest, DropNewest, or Terminate to fail closed under pressure. The queue does not implement a blocking producer/consumer model; its explicit contract is that a callback can be queued, inspected, and flushed as an exchange batch to the supervisor or another coordinator. Peek() and Flush() provide the boundary between worker-local activity and transport-level exchange; Flush() drains and returns the queued records as a list for later dispatch or persistence.

The hardening layer follows the same fail-closed philosophy. WasmHardeningLimits controls the maximum payload size, maximum callback nesting depth, and maximum callback duration. WasmHardeningPolicy validates each queued record and raises a WasmHardeningException when those limits are exceeded. The failure taxonomy is captured in WasmHardeningFailure (PayloadLimitExceeded, CallbackDepthExceeded, QueueDepthExceeded, and TimeoutExceeded), and the termination behavior is selected by WasmProcessTerminationPolicy.

This model is meant to keep the library deterministic under resource pressure. A worker cannot silently accept unbounded callback payloads or continue forever when resource ceilings are exceeded. Instead, the runtime treats those conditions as explicit faults that are surfaced to the caller and can trigger process termination according to policy. This is the documented Phase 6/7 behavior that forms part of the Phase 8 documentation update for the package.