Hi Yan, thanks for the advice. I will umount the cephfs on osd servers and keep an eye on it.:)
Yan, Zheng <ukernel@xxxxxxxxx> 于2018年8月7日周二 下午7:51写道：
On Tue, Aug 7, 2018 at 7:15 PM Zhenshi Zhou <deaderzzs@xxxxxxxxx> wrote:
>
> Yes, some osd servers mount cephfs
>
this can cause memory deadlock. you should avoid doing this
> Yan, Zheng <ukernel@xxxxxxxxx>于2018年8月7日 周二19:12写道：
>>
>> did you mount cephfs on the same machines that run ceph-osd?
>>
>> On Tue, Aug 7, 2018 at 5:14 PM Zhenshi Zhou <deaderzzs@xxxxxxxxx> wrote:
>> >
>> > Hi Burkhard,
>> > Files located in /sys/kernel/debug/ceph/<some id> are all new files generated after I reboot the server.
>> > The clients were in blacklist and I manully remove them from the blacklist. But the clients hung still.
>> >
>> > Thanks
>> >
>> > Burkhard Linke <Burkhard.Linke@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx> 于2018年8月7日周二 下午4:54写道：
>> >>
>> >> Hi,
>> >>
>> >>
>> >> you are using the kernel implementation of CephFS. In this case some
>> >> information can be retrieved from the /sys/kernel/debug/ceph/<some id>
>> >> directory. Especially the mdsc, monc and osdc files are important, since
>> >> they contain pending operations on mds, mon and osds.
>> >>
>> >>
>> >> We have a similar problem in our setup. If a MDS failover happens, some
>> >> clients (like 1 in 100 clients) do not react to the MDS map change and
>> >> keep on trying to contact the old MDS. In this case the msdc file in the
>> >> above mentioned directory lists from outstanding requests to the old MDS.
>> >>
>> >>
>> >> I would propose to start by having a look at these files and also at the
>> >> ceph client blacklist ('ceph osd blacklist ls').
>> >>
>> >>
>> >> Regards,
>> >>
>> >> Burkhard
>> >>
>> >>
>> >> _______________________________________________
>> >> ceph-users mailing list
>> >> ceph-users@xxxxxxxxxxxxxx
>> >> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
>> >
>> > _______________________________________________
>> > ceph-users mailing list
>> > ceph-users@xxxxxxxxxxxxxx
>> > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
_______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
