Hi Ilya, thanks for your answer - really helpful! We were so desparate today due to this bug that we downgraded to -23. But it's very good to know that -31 doesnt contain this bug and we could safely update back to this release. If a new version (say -33 is released): How/Where can I find out if it contains the fix? Except of trying and having it crash, of course - which I'm obviously very reluctant to do... And, one more question if I may: Would that problem also show up on Eoan (Ubuntu 19.10) which has been released a few days ago if we used the most recent Kernel there? I think it's 5.3.0-something if I'm not mistaken... Thank you / BR Ranjan Am 21.10.19 um 17:43 schrieb Ilya Dryomov:
On Mon, Oct 21, 2019 at 5:09 PM Ranjan Ghosh <ghosh@pw6.de> wrote:
Hi all,
it seems Ceph on Ubuntu Disco (19.04) with the most recent kernel 5.0.0-32 is instable. It crashes sometimes after a few hours, sometimes even after a few minutes. I found this bug here on CoreOS:
https://github.com/coreos/bugs/issues/2616
Which is exactly also the error message I get ("cache_from_obj: Wrong slab cache. inode_cache but object is from ceph_inode_info") and the problem seems fairly recent.
The problem vanished immediately after downgrading the kernel again. Hi Ranjan,
Which kernel did you downgrade to?
Looking at the changelog, disco's 5.0.0-32 kernel has the botched backport mentioned in the coreos issue you linked to. Downgrading to 5.0.0-31 should help, until disco picks up the fix (either just the revert or the revert plus the new backport).
Revert "ceph: use ceph_evict_inode to cleanup inode's resource" ceph: use ceph_evict_inode to cleanup inode's resource
Thanks,
Ilya