Hi again, On 29.06.2018 17:37, ulembke@xxxxxxxxxxxx wrote: > ... > 24.cc crushmap: [8,111,12,88,128,44,56] > real live: [8,121, X,88,130,44,56] - due the new osd-12 and the > wrong searchlist (osd-121 + osd-130) the PG is undersized! > > /var/lib/ceph/osd/ceph-8/current/24.ccs0_head > /var/lib/ceph/osd/ceph-44/current/24.ccs5_head > /var/lib/ceph/osd/ceph-56/current/24.ccs6_head > /var/lib/ceph/osd/ceph-88/current/24.ccs3_head > /var/lib/ceph/osd/ceph-121/current/24.ccs1_head > /var/lib/ceph/osd/ceph-130/current/24.ccs4_head > > ... unfortunality the PG-slices on the "wrong" ODSs (121,130) are empty - so the data are gone... and now it's happens to more PGs... looks, that deep-scrup remove important data?! We have disabled scrubbing and deep-scrubbing for now. The cluster was installed a long time ago (Cuttlefish/Dumpling) and was updated to 0.94.7 over the time but allways online. This year (Februrary) there was an power outage and the whole cluster was down and fresh restartet. After that, some rbd-devices, which are lying on the EC-Pool, are corrupt and the VM (an archive, with an huge FS on lvm (80TB)) need an repair of the lvm, which ends with an damaged FS without usefull data. Looks that after restarting the cluster, ceph use another calculation for the PGs (EC-Pool) than before?! Is that possible? Best regards Udo _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |