Ceph

03/20/2024

03:20 PM CephFS Bug #64977: mds spinlock due to lock contention leading to memory exaustion

Found a thread in list.ceph.io [1] reporting the problem we see following an upgrade to 6.5 kernel client.
MDS alloc... Enrico Bocchi

11:40 AM CephFS Bug #64977: mds spinlock due to lock contention leading to memory exaustion

There's also a potential cap sequence mismatch issue tracked at https://tracker.ceph.com/issues/61782
The fix for th... Enrico Bocchi

11:36 AM CephFS Bug #64977: mds spinlock due to lock contention leading to memory exaustion

Looking further into this reveled that:
- All the clients triggering the issue are kernel clients running 6.5.5-200.... Enrico Bocchi

03/19/2024

11:53 AM CephFS Bug #64977 (Fix Under Review): mds spinlock due to lock contention leading to memory exaustion

We have started to observe an MDS spinlock issue on two CephFS clusters running 16.2.9.
On both clusters, the offe... Enrico Bocchi

12/19/2023

08:29 AM rgw Bug #63373: multisite: Deadlock in RGWDeleteMultiObj with default rgw_multi_obj_del_max_aio > 1

Hi Mark,
??
In my checks it DOES repro on 18.2.x and current main (74ad7ff7)
with and without https://github.com... Enrico Bocchi

12/14/2023

04:55 PM rgw Bug #63373: multisite: Deadlock in RGWDeleteMultiObj with default rgw_multi_obj_del_max_aio > 1

Hi Mark,
Interestingly, @`warp delete`@ does not seem to suffer from this issue -- I have reproduced what you repo... Enrico Bocchi

11/17/2023

04:24 PM rgw Bug #63373: multisite: Deadlock in RGWDeleteMultiObj with default rgw_multi_obj_del_max_aio > 1

Pattern leading to the deadlock in 18.2.1 and patch has not changed:... Enrico Bocchi

02:01 PM rgw Bug #63373: multisite: Deadlock in RGWDeleteMultiObj with default rgw_multi_obj_del_max_aio > 1

Just built reef 18.2.1 with the patch on top, but does not seem to fix the problem.
Looking further... Enrico Bocchi

11/14/2023

02:06 PM rgw Bug #63528: rgw recursive delete deadlock

Do you see this happening in a multi-site RGW setup or also with simpler single-zone clusters?
We are seeing the s... Enrico Bocchi

02:02 PM rgw Bug #63373: multisite: Deadlock in RGWDeleteMultiObj with default rgw_multi_obj_del_max_aio > 1

We also suffer from the same deletion problem with large buckets reported above; Signature is identical. Enrico Bocchi