Re: [pve-devel] [PATCH V3 qemu-server 4/6] add ignore-storage-errors for removing VM with missing storage

[Fiona Ebner <f.ebner@proxmox.com>]

Am 15.11.22 um 11:55 schrieb Stefan Hrdlicka:
> @@ -2341,10 +2346,10 @@ sub destroy_vm {
>  
>  		my $volid = $drive->{file};
>  		return if !$volid || $volid =~ m|^/|;
> -
> -		die "base volume '$volid' is still in use by linked cloned\n"
> -		    if PVE::Storage::volume_is_base_and_used($storecfg, $volid);
> -
> +		my $result;
> +		eval { $result = PVE::Storage::volume_is_base_and_used($storecfg, $volid) };
> +		die "Couldn't remove one or more disks: $@\n" if $@ && !$ignore_storage_errors;

This error message is wrong. The check failed, not the removal. The
check should be repeated in vdisk_free anyways and you should get the
appropriate error then below :)

AFAIU base volumes should still survive if they are still referenced by
linked clones, even when ignore-storage-errors is used (IMHO good). Is
that correct?

Nothing new and not directly related:
I noticed that for containers, we don't have this heads-up check. Maybe
worth adding there too? Arguably minor issue is that I can have a
container template with a disk on lvm-thin and a second disk on
non-lvm-thin. Even if there is a linked clone, removing the template
might remove the lvm-thin disk, and then fail, because the second disk
is referenced.

> +		die "base volume '$volid' is still in use by linked cloned\n" if $result;
>  	});
>      }
>  
> @@ -2370,7 +2375,8 @@ sub destroy_vm {
>  	include_unused => 1,
>  	extra_keys => ['vmstate'],
>      };
> -    PVE::QemuConfig->foreach_volume_full($conf, $include_opts, $remove_owned_drive);
> +    eval { PVE::QemuConfig->foreach_volume_full($conf, $include_opts, $remove_owned_drive); };
> +    die "Couldn't remove one or more disks: $@\n" if $@ && !$ignore_storage_errors;

So, $removed_owned_drive already ignores all storage errors beside if
PVE::Storage::path() fails right? Can't we just add an eval around that
and be done? No need for a new ignore-storage-errors parameter. Most
storage errors are already ignored even without that parameter right
now! I don't think it's a big issue to start ignoring the few missing
ones as well?

>  
>      for my $snap (values %{$conf->{snapshots}}) {
>  	next if !defined($snap->{vmstate});