Lazy provider initialization — RPC server starts before discovery completes

pyvider 2026-03-22

2 files · 123+ · 29-

1 file changed

provide_command.py Refactored _run_provider_server() to fire component discovery as an asyncio background task. Added initialize_and_register_provider() coroutine that awaits discovery then instantiates and registers providers in the hub. Server.serve() now starts before initialization completes. Removed premature synchronous _ensure_components_discovered() call from provide_cmd().

Component discovery is now launched as asyncio.create_task() rather than awaited synchronously. A new initialize_and_register_provider() coroutine awaits the discovery task, instantiates providers, and registers the primary provider in the hub — all while the RPC server is already listening. A single await asyncio.sleep(0) yields control so background tasks start before server.serve() blocks the event loop. decouple behavioral
1 file
- provide_command.py
Removed the blocking _ensure_components_discovered() call and _handle_discovery_errors() from provide_cmd(). Discovery errors are now surfaced asynchronously during background initialization rather than blocking the CLI entry point. deprecate behavioral
1 file
- provide_command.py

1 file changed

handler.py Made _provider field optional (default=None); added _resolved_provider cache; added _ensure_provider_ready() async method that fetches the provider from the hub on first RPC request. _delegate() calls _ensure_provider_ready() before dispatching.

ProviderHandler no longer requires the provider at construction time. _ensure_provider_ready() checks a local cache first, then falls back to hub.get_component(‘singleton’, ‘provider’). If the provider is not yet registered (initialization still in progress), it waits 100ms and retries once before raising RuntimeError. interface behavioral
1 file
- handler.py