async-xenapi

Async bindings for XenAPI — available in both Python and JavaScript/TypeScript.

Structure

async-xenapi/
├── python/          # Python async library (PyPI: async-xenapi)
│   ├── src/
│   ├── tests/
│   └── README.md
├── javascript/      # TypeScript async library (npm: async-xenapi)
│   ├── src/
│   ├── tests/
│   └── README.md
└── README.md

Quick Start

Python — see python/README.md

from async_xenapi import AsyncXenAPISession
import asyncio

async def main():
    session = AsyncXenAPISession("https://xen-host")
    await session.login_with_password("root", "password")
    hosts = await session.xenapi.host.get_all()
    print(hosts)
    await session.logout()

asyncio.run(main())

JavaScript / TypeScript — see javascript/README.md

import { AsyncXenAPISession } from "async-xenapi";

async function main() {
  const session = new AsyncXenAPISession(process.env.HOST_URL);
  await session.login_with_password(process.env.USERNAME, process.env.PASSWORD);
  const hosts = await session.xenapi.host.get_all();
  console.log(hosts);
  await session.logout();
}

main();

Both implementations follow the same API conventions as the official XenAPI SDK.

Best Practices

For detailed examples with code, see the language-specific READMEs:

• Python best practices
• JavaScript best practices

Key Takeaways

1. Prefer get_all_records() over get_all() + N×get_field() — one round-trip instead of N+1.
2. Use asyncio.gather() / Promise.all() to fire independent calls concurrently — ~2x faster than sequential.
3. Always call logout() in a finally block to release the server-side session.

Benchmark

A benchmark suite compares the official sync XenAPI SDK (XML-RPC) against async-xenapi (JSON-RPC over aiohttp):

make py-bench

Tested against a single XenServer host (33 VMs, 6 SRs, 4 networks):

Step	Sync XenAPI	Async XenAPI	Speedup
login	1.5s	1.3s	1.1x
xapi_version	1.3s	1.7s	0.7x
hosts	1.9s	1.5s	1.2x
vms	3.8s	3.9s	1.0x
storage	628ms	617ms	1.0x
networks	315ms	278ms	1.1x
all_concurrent	3.2s	1.7s	1.9x
logout	313ms	310ms	1.0x
TOTAL	12.8s	11.3s	1.1x

For individual sequential calls, sync and async perform similarly — the bottleneck is network latency. The async advantage appears when firing multiple independent requests concurrently via asyncio.gather(), where all_concurrent (fetching hosts, VMs, SRs, and networks in parallel) runs ~2x faster than the equivalent sequential sync calls.

License

LGPL-2.1-only