2024年8月14日(水) 8:23 Raphaël Ducom <rducom@luccasoftware.com>:
Hi
I'm reaching out to check on the status of the XFS deadlock issue with RBD in hyperconverged environments, as detailed in Ceph tracker issue #43910 ( https://tracker.ceph.com/issues/43910?tab=history). It looks like there hasn’t been much activity on this for a while, and I'm wondering if there have been any updates or if it’s just been lost in the issue volume.
The issue is a deadlock when using XFS, leading Rook to recommend using EXT4 instead. However, kernel 5.6 introduced PR_SET_IO_FLUSHER, which should allow Ceph to handle this scenario better. Versions lacking PR_SET_IO_FLUSHER are either already EOL or soon will be, with the exception of kernel 5.4.
While EXT4 works according to the Ceph documentation, the limitations with XATTR on EXT4 are still a concern. It’s a bit unfortunate that Rook has to move away from XFS where Ceph recommends it. The issue on the Rook side : https://github.com/rook/rook/issues/3132#issuecomment-580508760
Could anyone provide an update on this Ceph issue or suggest how we might push it forward? Any insights would be really appreciated, as this impacts the broader use of XFS in hyperconverged Ceph setups deployed with Rook.
I've also encountered this issue. My team haven't be able to XFS for RBD due to this problem. I hope that someone will fix this problem, in other words, calling `prctl(PR_SET_IO_FLUSHER)` somewhere in the Ceph daemons' initialization code. Or if someone teach me the appropriate place to insert this call, I'll create a patch. Best, Satoru