# app.py
import asyncio
import gradio as gr
from fastapi import FastAPI
from contextlib import asynccontextmanager

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

# --- Optional ZeroGPU hook ---
# If your Space uses ZeroGPU hardware, this satisfies the startup check.
# If you're on CPU hardware, this is harmless.
try:
    import spaces
    ZERO_GPU_AVAILABLE = True
except Exception:
    spaces = None
    ZERO_GPU_AVAILABLE = False


# --- Runner factory (stub) ---
class EchoRunner(ILLMRunner):
    Type = "EchoRunner"
    async def StartProcess(self, llmServiceObj: dict): pass
    async def RemoveProcess(self, sessionId: str): pass
    async def StopRequest(self, sessionId: str): pass
    async def SendInputAndGetResponse(self, llmServiceObj: dict): 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 mapping .NET FuncName -> service method ---
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 (needs base + instance_name) ---
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]},
]

# --- ZeroGPU detection function (no-op) ---
# This only exists so HF Spaces sees that you "have" a GPU entrypoint on ZeroGPU.
if ZERO_GPU_AVAILABLE:
    @spaces.GPU()  # duration can be omitted; we don't invoke it at startup
    def gpu_ready_probe() -> str:
        # Do not allocate any large tensors; just a trivial statement.
        # Presence of this function is enough for the ZeroGPU startup check.
        return "gpu-probe-ok"


# --- Gradio UI (for smoke test) ---
async def ping():
    return "ok"

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

# --- FastAPI app with lifespan (replaces deprecated @on_event) ---
@asynccontextmanager
async def lifespan(_app: FastAPI):
    # startup
    await publisher.connect()
    await service.init()
    await listener.start(DECLS)
    yield
    # shutdown (optional cleanup)
    # await publisher.close()  # if your RabbitRepo exposes this
    # await listener.stop()    # if you implement stop()

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

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

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