Thank you very much for your reply!! I will follow your suggestions.
Regards,
Simone Cariani
Da: Dave Scott [mailto:Dave.Scott@xxxxxxxxxxxxx]
Inviato: lunedì 15 novembre 2010 11.55
A: 'Simone Cariani'; xen-api@xxxxxxxxxxxxxxxxxxx; xen-users@xxxxxxxxxxxxxxxxxxx
Oggetto: RE: [Xen-API] problem with snapshot unallocation
Hi Simone,
Whenever you create a snapshot disk it will increase the length of the “snapshot chain” by 1 and there is a hard length limit of 30. However whenever you delete a snapshot disk the background “coalesce” process should be able to reduce the “snapshot chain” length again. So your procedure (create snapshot; export snapshot; delete snapshot) ought to work fine. This is in fact the same process followed by the “VM protection and recovery” feature which will be part of XCP 1.0.
The problem you linked to is related but not quite the same: it refers to a limitation within the “coalesce” process where it cannot reduce a chain of length 2 into a chain of length 1. The only reason people worry about this is because a chain of length 2 can take up a lot more space on LVM where there is no “thin provisioning” – this isn’t a problem for you on NFS.
Given your observation that this only happens to VMs with more than 1 attached virtual disk, I think you’ve found a bug in the tool which you used to add the second disk to the VM. When you delete a VM snapshot (or uninstall a VM or template generally), only those disks which have a special flag set are actually deleted (specifically those with VBD.other-config:owner IIRC). If you use the “xe vm-disk-add” CLI command it should set this flag. If this flag isn’t set then it would cause the additional disks not to be deleted… this would cause your “snapshot chain” to get ever bigger, hit 30 and then fail.
I think a workaround is to identify the disk (VDI not VM) snapshots which haven’t been deleted and delete them manually, possibly via the CLI (“xe vdi-destroy”). After a short while the background “coalesce” process should kick in and reduce your chain lengths again.
Cheers,
Dave