On 11/8/18 12:29 AM, Hayashida, Mami wrote: > Yes, that was indeed a copy-and-paste mistake. I am trying to use > /dev/sdh (hdd) for data and a part of /dev/sda (ssd) for the journal. > That's how the Filestore is set-up. So, for the Bluestore, data on > /dev/sdh, wal and db on /dev/sda.
/dev/sda is the SSD you use for all OSDs on each node, right? Keep in mind that what you're doing here is wiping that SSD entirely and converting it to LVM. If any FileStore OSDs are using that SSD as journal then this will kill them. If you're doing one node at a time that's fine, but then you need to out and stop all the FileStore OSDs on that node first. You should throw a "systemctl daemon-reload" in there after tweaking fstab and the systemd configs, to make sure systemd is aware of the changes, e.g. after the `ln -s /dev/null ...`. FWIW I don't think that symlink is necessary, but it won't hurt. Also, `ceph-volume lvm zap` doesn't seem to trigger a re-read of the partition table, from a quick look. Since you used partitions before, it might be prudent to do that. Try `hdparm -z /dev/sdh` after the zap (same for sda). That should get rid of any /dev/sdh1 etc partition devices and leave only /dev/sdh. Do the same for sda and anything else you zap. This also shouldn't strictly be needed as those will disappear after a reboot anyway, and it's possible some other tool implicitly does this for you, but it's good to be safe, and might avoid trouble if some FileStore remnant tries to mount phantom partitions. -- Hector Martin (hec...@marcansoft.com) Public Key: https://mrcn.st/pub _______________________________________________ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
