Yes, reweight a lot before the upgrade. But from the output of ceph osd df, the reweight values were kept instead of 1 after the upgrade. Of course, the old reweight values are meaningless after the mass data migration. br, Xu Yun > 2020年1月20日 下午10:19，Wido den Hollander <wido@xxxxxxxx> 写道： > > > > On 1/20/20 1:07 AM, 徐蕴 wrote: >> Hi, >> >> We upgraded our cluster from Jewel to Luminous, and it turned out that more than 80% object misplaced. Since our cluster has 130T data, backfilling seems take forever. We didn’t modify any crushmap. Any thoughts about this issue? > > Did you reweight your OSDs prior to the upgrade? > > Otherwise it is probably the case that the reweights of the OSDs were > set back to 1 causing this data migration. > > Or did somebody run: > > $ ceph osd crush tunables optimal > > Wido > >> >> br, >> Xu Yun >> _______________________________________________ >> ceph-users mailing list -- ceph-users@xxxxxxx >> To unsubscribe send an email to ceph-users-leave@xxxxxxx >> _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |