Thanks a lot Adrien. This worked for us. We might wipe and recreate the OSDs when the fix is backported.
~# lvcreate -L 512g -n manual-db-1 ceph-60bd102a-b11a-44cc-9c28-106e14603b88 ~# ceph orch daemon add osd s3db23:data_devices=/dev/sdk,db_devices=/dev/ceph-60bd102a-b11a-44cc-9c28-106e14603b88/manual-db-1,encrypted=true lvm Am Di., 24. Feb. 2026 um 15:39 Uhr schrieb Adrien Georget < [email protected]>: > Hi Boris, > > The workaround we used to replace disks affected by this bug is to force > devices in the osd creation using the orchestrator without yaml specs. > VG and LV are first created manually and then we used something like this > : > > ceph orch daemon add osd > <host>:data_devices=/dev/data01,db_devices=/dev/fast01/db01 > > > See the advanced OSD creation : > https://docs.ceph.com/en/reef/cephadm/services/osd/#creating-new-osds > > Adrien > Le 24/02/2026 à 14:57, Boris via ceph-users a écrit : > > Thanks Marek, but this did not work here, because the SSDs still have other > block.db LVs on it. > > > Am Di., 24. Feb. 2026 um 14:46 Uhr schrieb Marek Szuba via ceph-users > <[email protected]>: > > > On 2026-02-24 11:15, Robert Sander via ceph-users wrote: > > > The orchestrator in Ceph 19 and 20 has a bug with hybrid OSDs: > https://tracker.ceph.com/issues/72696 > > There is a patch available but it has not been merged yet. > > Meanwhile, a workaround that has worked for us is: > * tell the orchestrator to delete unwanted non-hybrid OSDs _without > zapping the drives_; > * manually wipe both underlying devices, with as little time in > between the two as possible; > * explicitly tell the orchestrator to refresh its device list, so that > it sees both available drives at the same time. > > -- > MS > _______________________________________________ > ceph-users mailing list -- [email protected] > To unsubscribe send an email to [email protected] > > > -- Die Selbsthilfegruppe "UTF-8-Probleme" trifft sich diesmal abweichend im groüen Saal. _______________________________________________ ceph-users mailing list -- [email protected] To unsubscribe send an email to [email protected]
