I actually managed to reboot everything today and it ran smoothly for the last few minutes. MDS failover also worked without problems. If anything bad happens in the next days, I will let you know. Markus On Tue, May 19, 2015 at 1:12 PM, Markus Blank-Burian <burian@xxxxxxxxxxx> wrote: > Thanks for the patch! Testing might take up to a week, since I have to > reboot all the client nodes in the computing cluster. > > On Tue, May 19, 2015 at 12:27 PM, Yan, Zheng <ukernel@xxxxxxxxx> wrote: >> could you try the attached patch >> >> On Tue, May 19, 2015 at 5:10 PM, Markus Blank-Burian <burian@xxxxxxxxxxx> wrote: >>> Forgot the attachments. Besides, is there any way to get the cluster >>> running again without restarting all client nodes? >>> >>> On Tue, May 19, 2015 at 10:45 AM, Yan, Zheng <ukernel@xxxxxxxxx> wrote: >>>> On Tue, May 19, 2015 at 4:31 PM, Markus Blank-Burian <burian@xxxxxxxxxxx> wrote: >>>>> I am afraid, I hit the same bug. Giant worked fine, but after upgrading to >>>>> hammer (0.94.1) and putting some load on it, the MDSs eventually crashed and >>>>> now I am stuck in clientreplay most of the time. I am also using the cephfs >>>>> kernel client (3.18.y). As I didn't find a corresponding tracker entry .. is >>>>> there already a patch available? >>>>> >>>> >>>> Please send mds log and /sys/kernel/debug/ceph/*/mdsc on client >>>> machine to us. Besides, Is there warnings like "cluster [WRN] slow >>>> request [several thousands or more ] seconds old, received at ...: >>>> client_request(client.734537:23 ...) " in your ceph cluster log. >>>> >>>> Regards >>>> Yan, Zheng _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |