Hi all Just to let you know that eventually we (myself, a colleague and chatgpt6) were able to recover from the disaster Basically: * we exported the shards of the problematic PG from the crashed OSDs * we prepared 3 new OSDs using a 200 GB file as backend, with the "identities" of the 3 crashed OSDs (the ones expected by the PG in down) * we imported the PG shards on these "recovery" OSDs Now we have started the process of draining and then reinstalling all the OSDs created with squid 19.2.3. It will be a long process (80 OSDs...) Have a good weekend Massimo On Fri, Sep 18, 2026 at 4:28 PM Anthony D'Atri <anthony.datri@gmail.com> wrote:
From what I understand, when you inject a "foreign" PG into the OSD, it
to the mons that it now has this PG and its objects (at some version) and then the mons/mgrs will often decide it is available, but perhaps misplaced and move it to where it now should reside. And if it is not broken, it will start serve data and/or be the source for backfills if it was the only replica for
will report this PG.
Exactly, sometimes in conjunction with temporarily lowering min_size.
Of course I have never had to do this myself, but from what reports here
on
the maillist have said, I have understood it as being a good way to inject a copy of a PG if you can extract it from an otherwise failed OSD which will not start.
I've done it once or twice with success.
-- May the most significant bit of your life be positive. _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io