Backups verifies failing

tuxis

Famous Member
Jan 3, 2014
248
221
108
Ede, NL
www.tuxis.nl
I have a lot of backups failing on the beta PBS. The verify complains that the CRC checksum is incorrect.

Is there a way I can check this myself?
 
the crc32 is computed over the actual data stored inside the chunks/blobs. each chunk/blob starts with a magic value (first 8 bytes) which determines whether the blob is compressed and/or encrypted (https://git.proxmox.com/?p=proxmox-...aa43050da74a688e8dade71e50039d6915519;hb=HEAD), followed by the CRC (4 bytes), followed by IV and TAG (16 bytes each) if it is encrypted. so depending on the magic value, you'd need to skip the first 12 or 34 bytes, calculate crc32 for the rrst and compare with bytes 9-12
 
  • Like
Reactions: Moayad
but calculating/comparing yourself won't help much - if the code says the CRC is incorrect, your calculation will also show it's incorrect.. the CRC is calculated server side, so it's also not really possible that the client sent you a wrong checksum. is this for the new LVM datastore, or the ZFS one? if ZFS, does 'zfs scrub' show any errors?

if the blobs are uncompressed, you could also check their size compared to the index that references them.
 
This is a datastore that has not existed on ZFS. So a simple ext4 on LVM.

If you want, I can grant you access to the machine so you can investigate. I'm pretty sure there's nothing special, but maybe I'm missing something completely.
 
You moved the chunks via rsync from your ZFS based datastores to LVM without stopping client access? This can explain the problem.

Better: use sync jobs