Multiple write access on same file issue

Latest response

Hi,
working on a ceph cluster :
CentOS 7.8
Ceph containerized 15.2.9.
10 nodes cluster
each node : 22 HDD, 238 GB RAM, 16 cores, 20 GB access Network, 50 GB replication Network,
3 ways replication
Using a multi-MDS CephFS.
Clients are usgin Kernel module, not fuse.

Using a 8 nodes injector benchmark tool.
When using a dd (or fio ) on 8 different files, everything is ok
When doing the same benchmark a few times later ( *) on the same 8 files, every client hang forever until hard reboot. Nothing appears on server logs, and the cluster does not encounter any issue.
During the benchmark, nor CPUs, OSD,Network seems to be capped.

We do not have a single issue testing with fuse mode..

I stay available for any more information.
Thanks for all
Best regards

SM

Responses