app.py

# app.py
import asyncio
from contextlib import asynccontextmanager

import gradio as gr
from fastapi import FastAPI

from config import settings
from rabbit_base import RabbitBase
from listener import RabbitListenerBase
from rabbit_repo import RabbitRepo
from service import LLMService
from runners.base import ILLMRunner


# =========================
# @spaces.GPU() SECTION
# =========================
# This trivial GPU endpoint keeps ZeroGPU Spaces alive at startup.
try:
    import spaces
    ZERO_GPU_AVAILABLE = True

    @spaces.GPU()  # keep it trivial (no tensor allocations)
    def gpu_ready_probe() -> str:
        """
        Minimal GPU-decorated function so ZeroGPU detects a GPU entrypoint.
        It's also referenced by a Gradio button and a FastAPI route below.
        """
        return "gpu-probe-ok"

except Exception:
    ZERO_GPU_AVAILABLE = False

    # Fallback for local/CPU-only runs (same signature)
    def gpu_ready_probe() -> str:
        return "cpu-only"


# ---------------- Runner factory (stub) ----------------
class EchoRunner(ILLMRunner):
    Type = "EchoRunner"

    async def StartProcess(self, llmServiceObj: dict):  # noqa: N802
        pass

    async def RemoveProcess(self, sessionId: str):  # noqa: N802
        pass

    async def StopRequest(self, sessionId: str):  # noqa: N802
        pass

    async def SendInputAndGetResponse(self, llmServiceObj: dict):  # noqa: N802
        pass


async def runner_factory(llmServiceObj: dict) -> ILLMRunner:
    return EchoRunner()


# ---------------- Publisher and Service ----------------
publisher = RabbitRepo(external_source="https://space.external")
service = LLMService(publisher, runner_factory)

# ---------------- Handlers (.NET FuncName -> service) ----------------
async def h_start(data):  await service.StartProcess(data or {})
async def h_user(data):   await service.UserInput(data or {})
async def h_remove(data): await service.RemoveSession(data or {})
async def h_stop(data):   await service.StopRequest(data or {})
async def h_qir(data):    await service.QueryIndexResult(data or {})
async def h_getreg(_):    await service.GetFunctionRegistry(False)
async def h_getreg_f(_):  await service.GetFunctionRegistry(True)

handlers = {
    "llmStartSession": h_start,
    "llmUserInput": h_user,
    "llmRemoveSession": h_remove,
    "llmStopRequest": h_stop,
    "queryIndexResult": h_qir,
    "getFunctionRegistry": h_getreg,
    "getFunctionRegistryFiltered": h_getreg_f,
}

# ---------------- Listener wiring ----------------
base = RabbitBase()
listener = RabbitListenerBase(
    base,
    instance_name=settings.RABBIT_INSTANCE_NAME,  # queue prefix like your .NET instance
    handlers=handlers,
)

# Declarations mirror your C# InitRabbitMQObjs()
DECLS = [
    {"ExchangeName": f"llmStartSession{settings.SERVICE_ID}", "FuncName": "llmStartSession",
     "MessageTimeout": 600_000, "RoutingKeys": [settings.RABBIT_ROUTING_KEY]},
    {"ExchangeName": f"llmUserInput{settings.SERVICE_ID}", "FuncName": "llmUserInput",
     "MessageTimeout": 600_000, "RoutingKeys": [settings.RABBIT_ROUTING_KEY]},
    {"ExchangeName": f"llmRemoveSession{settings.SERVICE_ID}", "FuncName": "llmRemoveSession",
     "MessageTimeout": 60_000, "RoutingKeys": [settings.RABBIT_ROUTING_KEY]},
    {"ExchangeName": f"llmStopRequest{settings.SERVICE_ID}", "FuncName": "llmStopRequest",
     "MessageTimeout": 60_000, "RoutingKeys": [settings.RABBIT_ROUTING_KEY]},
    {"ExchangeName": f"queryIndexResult{settings.SERVICE_ID}", "FuncName": "queryIndexResult",
     "MessageTimeout": 60_000, "RoutingKeys": [settings.RABBIT_ROUTING_KEY]},
    {"ExchangeName": f"getFunctionRegistry{settings.SERVICE_ID}", "FuncName": "getFunctionRegistry",
     "MessageTimeout": 60_000, "RoutingKeys": [settings.RABBIT_ROUTING_KEY]},
    {"ExchangeName": f"getFunctionRegistryFiltered{settings.SERVICE_ID}", "FuncName": "getFunctionRegistryFiltered",
     "MessageTimeout": 60_000, "RoutingKeys": [settings.RABBIT_ROUTING_KEY]},
]


# ---------------- Gradio UI (smoke test + GPU probe) ----------------
async def ping():
    return "ok"


with gr.Blocks() as demo:
    gr.Markdown("### LLM Runner (Python) — RabbitMQ listener")
    with gr.Row():
        btn = gr.Button("Ping")
        out = gr.Textbox(label="Ping result")
    btn.click(ping, inputs=None, outputs=out)

    # IMPORTANT: reference the decorated function DIRECTLY (no lambda)
    if ZERO_GPU_AVAILABLE:
        probe_btn = gr.Button("GPU Probe")
        probe_out = gr.Textbox(label="GPU Probe Result")
        probe_btn.click(gpu_ready_probe, None, probe_out)


# ---------------- FastAPI + lifespan ----------------
@asynccontextmanager
async def lifespan(_app: FastAPI):
    # startup
    await publisher.connect()
    await service.init()
    await listener.start(DECLS)
    yield
    # shutdown (optional)
    # await publisher.close()
    # await listener.stop()


app = FastAPI(lifespan=lifespan)
app = gr.mount_gradio_app(app, demo, path="/")

@app.get("/health")
async def health():
    return {"status": "ok"}

# Also expose probe via HTTP (belt & braces for ZeroGPU detectors)
@app.get("/gpu-probe")
def gpu_probe_route():
    return {"status": gpu_ready_probe()}


if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)