On Wed, May 24, 2017 at 1:47 PM, Shain Miley <SMiley@xxxxxxx> wrote: > Hello, > We just upgraded from Hammer to Jewel, and after the cluster once again > reported a healthy state I set the crush tunables to ‘optimal’ (from > legacy). > 12 hours later and the cluster is almost done with the pg remapping under > the new rules. > > The issue I am having is the server where we mount the krbd images is > showing errors in the kern.log: > > May 24 07:28:14 rbd1 kernel: [5600763.226208] libceph: osd192 > 10.35.1.235:6844 feature set mismatch, my 2b84a042a42 < server's > 40002b84a042a42, missing 400000000000000 > > And I can no longer list any of the mounted filesystems or unmap the rbd > images, etc. > > My options seem to be: > > 1)set the tunables back to legacy and see if the rbd server starts > responding. > > 2)upgrade the kernel on the rbd server to at least version 4.5 (currently > using 3.18 on Ubuntu 14.04). As per [1], I'd recommend upgrading to 4.9.z. > > 3)disable some features on our current images? This is the "cluster" feature bit, not the image feature. Don't enable new image features just because they are there in jewel though ;) > > I would like to try option 2 first…but I am wondering if is safe to reboot > the server with the rbd images still mapped…is there any chance of data loss > from an rbd image getting corrupted? Take a look at /sys/kernel/debug/ceph/*/osdc. If it's empty, there are no in-flight requests and you should be able to cold reboot safely. If there is a lot of pending requests, the safest option is to revert the tunables setting. [1] http://docs.ceph.com/docs/master/start/os-recommendations/ Thanks, Ilya _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |