Python trio primer

1. Install the dependencies

Trio plays well with httpx because httpx ships a native trio transport. Pin both so cancellation semantics stay consistent across environments.

pip install "trio>=0.24" "httpx[trio]>=0.27"

2. Open a nursery for parallel calls

Nurseries give you a structured scope. Every task spawned with start_soon is awaited before the block exits, and any exception cancels its siblings. Meridian's gateway tolerates bursts up to your tier ceiling, so a nursery is the right unit.

import trio
import httpx

async def fetch_status(client: httpx.AsyncClient, url: str) -> None:
    response = await client.get(url, headers={"x-meridian-key": "sk-..."})
    print(url, response.status_code)

async def main() -> None:
    urls = [
        "https://llm.getnimbus.net/v1/models",
        "https://llm.getnimbus.net/v1/health",
        "https://llm.getnimbus.net/v1/usage",
    ]
    async with httpx.AsyncClient(timeout=15.0) as client:
        async with trio.open_nursery() as nursery:
            for url in urls:
                nursery.start_soon(fetch_status, client, url)

if __name__ == "__main__":
    trio.run(main)

3. Cancel cleanly on the first failure

Wrap the nursery in move_on_after or fail_after to bound total latency. Because trio uses checkpoints rather than thread interrupts, an in-flight httpx request will close its socket cleanly when the scope cancels — no half-written billing records on the Meridian side.

• Use one AsyncClient per nursery so the connection pool is shared.
• Surface partial results through a trio.MemoryChannel if a caller needs streaming.
• Set timeout=httpx.Timeout(15.0, connect=5.0) to keep slow upstreams from starving the nursery.