2024-07-26T17:33:21.832Z

<david.casier> Hello,
I don't see an implementation of async (asio) for the libcephfs client.
Am I wrong?
Not proposed yet?

Thanks to those who have some information to bring me

2024-07-26T18:00:28.047Z

<Æmerson> There is, to the best of my knowledge, no integration of Asio with CephFS. There **is** some work to integrate it with NFS-Ganesha's asynchronous operation support, but that's it.

2024-07-26T18:21:50.138Z

<david.casier> I have a hard time assessing the workload this could require. I have the impression that asio is already well implemented on the message part (AsyncConnection) and that it is common to other libraries, am I wrong?

2024-07-26T18:35:51.925Z

<Æmerson> Asio is currently NOT used by the Messenger (unless someone has managed to do a lot of work without me noticing :)

2024-07-26T18:37:27.669Z

<Æmerson> The Asio integreation starts with the Objecter, and a fully asiofied RADOS client is in `include/neorados/RADOS.hpp`. And RBD uses Asio via neorados for some things.

2024-07-26T18:38:22.070Z

<Æmerson> We're currently trying to use Asio more widely through RGW, currently we use it for the networking code, we're switching RADOS operations to use it via neorados.

2024-07-26T18:39:24.696Z

<Æmerson> But, basically, you could certainly wrap the existing CephFS client in an interface that uses Asio if your main goal was to call it from programs that use Asio.

2024-07-26T18:39:46.193Z

<Æmerson> Taking advantage of the performance/memory/nonblocking aspects would be Much More Work.

2024-07-26T19:27:19.297Z

<Casey Bodley> i still like the AsioMessenger idea. especially if we could get it to use uring

2024-07-26T20:03:16.422Z

<Æmerson> Oh yes. I think AsioMessenber's definitely worth pursuing, for multiple reasons

2024-07-26T20:39:44.742Z

<Æmerson> Oh yes. I think AsioMessenger's definitely worth pursuing, for multiple reasons