Re: [pbs-devel] [PATCH backup v2 2/2] disks: wipe: only zero out the first 1 MiB

[Thomas Lamprecht <t.lamprecht@proxmox.com>]

Am 12.02.25 um 11:24 schrieb Filip Schauer:
> On 11/02/2025 19:42, Thomas Lamprecht wrote:
>> I have some faint remembrance that we had to increase the area we zero
>> out for ceph OSDs and/or for the installer.
> 
> You are right. We zero out the first 200 MiB to prevent a problem with
> OSD creation. [0] [1] Although I am unable to reproduce this on Ceph
> Quincy, making sure the disk is properly wiped does not hurt. So we can
> drop this patch.

Thanks for checking, seems like Ceph nowadays writes out labels on quite a
few more places [0] depending on the block dev size; albeit only once in the
range of 0 to 200 MB. I checked and this is relatively new though (commit
from January 2024 [1]) and only included in Squid 19.2.1 FWICT.

[0]: https://github.com/ceph/ceph/blob/main/src/os/bluestore/BlueStore.cc#L138-L146
[1]: https://github.com/ceph/ceph/commit/4b8197b9931e1597b5aa3b02660a9b77ea4aa124

FWIW, I'm not completely opposed in doing less writes, if the block device
is backed by some thin provisioned storage it might be also thankful if it
has less data to explicitly track (albeit zeros are often special cased
and as such relatively cheap most of the time).
If we ever reuse this code for PVE it would need to become opt-in though
so that we can enable it for the cases like wiping in preparation of
creating a Ceph OSD.

But I saw you send a v3 already with this left out, which is IMO fine too
for now.


_______________________________________________
pbs-devel mailing list
pbs-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pbs-devel