On 13-11-15 10:56, Jens Rosenboom wrote: > 2015-10-20 16:00 GMT+02:00 Wido den Hollander <wido@xxxxxxxx>: > ... >> The system consists out of 39 hosts: >> >> 2U SuperMicro chassis: >> * 80GB Intel SSD for OS >> * 240GB Intel S3700 SSD for Journaling + Bcache >> * 6x 3TB disk > > I'm currently testing a similar setup, but it turns out that setup and > operations are a bit clumsy: > > 1. ceph-disk-prepare when given the bcache device always wants to > create a partition on it, which isn't possible, so as a workaround I > have to manually put a FS onto the device, mount it and give ceph-disk > the mount-point. > 2. Since there is no partition, there is also no type UUID, so the > usual udev based autostart mechanisms do not work. Running "ceph-disk > activate /dev/bcache0" works fine though. > > Did you come up with some more clever solutions for this? > We are not using ceph-disk in those situations. These systems use sysvinit with this in their configuration: [osd.253] host = ceph39 devs = /dev/disk/by-location/encl01-slot03 osd uuid = 16585dc3-1f7e-4810-9d9e-4bf6651f54eb We wrote some custom scripting to set up bcache and mkfs them afterwards. > Also, do you have some experience with broken disks? I'm wondering if > one failing disk will affect performance for the others running in the > same cache set. > We experienced multiple disk failures. They don't seem to affect other processes. Wido _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com