Re: [HELP] FUSE writeback performance bottleneck

From: Miklos Szeredi
Date: Tue Jun 04 2024 - 03:20:54 EST


On Tue, 4 Jun 2024 at 00:10, Dave Chinner <david@xxxxxxxxxxxxx> wrote:

> I thought we had PR_SET_IO_FLUSHER for that. Requires
> CAP_SYS_RESOURCES but no other privileges, then the userspace
> server will then always operate in PF_MEMALLOC_NOIO |
> PF_LOCAL_THROTTLE memory allocation context.

There could be any number of services that are being used while
serving a fuse request. There's no well defined "fuse server
process", as many people seem to think. Any approach depending on
somehow marking the fuse server as a special entity will fail.

Thanks,
Miklos