From: "Fabian Grünbichler" <f.gruenbichler@proxmox.com>
To: Proxmox VE development discussion <pve-devel@lists.proxmox.com>
Subject: Re: [pve-devel] [PATCH qemu-server 6/6] fix #6543: use qcow2 'discard-no-unref' option when using snapshot-as-volume-chain
Date: Fri, 25 Jul 2025 09:38:54 +0200 [thread overview]
Message-ID: <1753428793.n5phbfd5ll.astroid@yuna.none> (raw)
In-Reply-To: <20250724135956.112138-7-f.ebner@proxmox.com>
On July 24, 2025 3:59 pm, Fiona Ebner wrote:
> Without the 'discard-no-unref', a qcow2 file can grow beyond what
> 'qemu-img measure' reports, because of fragmentation. This can lead to
> IO errors with qcow2 on top of LVM storages, where the containing LV
> is allocated with that size. Guard enabling the option with
> having 'snapshot-as-volume-chain' in the storage configuration for
> now. Enabling it always should be evaluated a bit more and tested on
> different storages. It is a runtime-only option just affecting how
> referencing clusters is handled during discard in qcow2 and nothing
> else, so it is also fine for existing images and migration streams.
>
> While 'snapshot-as-volume-chain' is not the perfect proxy, as that's
> not only for LVM, it's an experimental feature that covers the LVM
> case and it seems like a nice fit to try out the new option on
> file-based storages too.
>
> Suggested-by: Alexandre Derumier <alexandre.derumier@groupe-cyllene.com>
> Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
> ---
> src/PVE/QemuServer/Blockdev.pm | 7 ++++++
> src/PVE/QemuServer/QemuImage.pm | 19 +++++++++++++++
> src/test/cfg2cmd/simple-backingchain.conf.cmd | 2 +-
> src/test/run_qemu_img_convert_tests.pl | 24 ++++++++++++++-----
> 4 files changed, 45 insertions(+), 7 deletions(-)
>
> diff --git a/src/PVE/QemuServer/Blockdev.pm b/src/PVE/QemuServer/Blockdev.pm
> index 8528a587..1487bc99 100644
> --- a/src/PVE/QemuServer/Blockdev.pm
> +++ b/src/PVE/QemuServer/Blockdev.pm
> @@ -372,6 +372,13 @@ my sub generate_format_blockdev {
> $blockdev->{size} = int($options->{size});
> }
>
> + # see bug #6543: without this option, fragmentation can lead to the qcow2 file growing larger
> + # than what qemu-img measure reports, which is problematic for qcow2-on-top-of-LVM
> + # TODO test and consider enabling this in general
> + if ($scfg && $scfg->{'snapshot-as-volume-chain'}) {
> + $blockdev->{'discard-no-unref'} = JSON::true if $format eq 'qcow2';
> + }
> +
> return $blockdev;
> }
>
> diff --git a/src/PVE/QemuServer/QemuImage.pm b/src/PVE/QemuServer/QemuImage.pm
> index 026c24e9..7f6d5f01 100644
> --- a/src/PVE/QemuServer/QemuImage.pm
> +++ b/src/PVE/QemuServer/QemuImage.pm
> @@ -3,6 +3,9 @@ package PVE::QemuServer::QemuImage;
> use strict;
> use warnings;
>
> +use Fcntl qw(S_ISBLK);
> +use File::stat;
> +
> use PVE::Format qw(render_bytes);
> use PVE::Storage;
> use PVE::Tools;
> @@ -27,6 +30,18 @@ sub convert_iscsi_path {
> die "cannot convert iscsi path '$path', unknown format\n";
> }
>
> +my sub qcow2_target_image_opts {
> + my ($path, @qcow2_opts) = @_;
> +
> + my $st = File::stat::stat($path) or die "stat for '$path' failed - $!\n";
right now this is only called for PVE-managed volumes.. so we could
actually call qemu_blockdev_options instead in `convert` below, and use
the driver (and possibly other things?) from there?
> +
> + my $driver = S_ISBLK($st->mode) ? 'host_device' : 'file';
> +
> + my $qcow2_opts_str = ',' . join(',', @qcow2_opts);
> +
> + return "driver=qcow2$qcow2_opts_str,file.driver=$driver,file.filename=$path";
> +}
> +
> # The possible options are:
> # bwlimit - The bandwidth limit in KiB/s.
> # is-zero-initialized - If the destination image is zero-initialized.
> @@ -71,6 +86,7 @@ sub convert {
> my $dst_format = checked_volume_format($storecfg, $dst_volid);
> my $dst_path = PVE::Storage::path($storecfg, $dst_volid);
> my $dst_is_iscsi = ($dst_path =~ m|^iscsi://|);
> + my $dst_needs_discard_no_unref = $dst_scfg->{'snapshot-as-volume-chain'};
&& $dst_format eq 'qcow2'
as above in Blockdev.pm?
> my $support_qemu_snapshots = PVE::Storage::volume_qemu_snapshot_method($storecfg, $src_volid);
>
> my $cmd = [];
> @@ -94,6 +110,9 @@ sub convert {
> if ($dst_is_iscsi) {
> push @$cmd, '--target-image-opts';
> $dst_path = convert_iscsi_path($dst_path);
> + } elsif ($dst_needs_discard_no_unref) {
> + push @$cmd, '--target-image-opts';
> + $dst_path = qcow2_target_image_opts($dst_path, 'discard-no-unref=true');
> } else {
> push @$cmd, '-O', $dst_format;
> }
_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
next prev parent reply other threads:[~2025-07-25 7:37 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-07-24 13:59 [pve-devel] [PATCH-SERIES qemu-server 0/6] blockdev and snapshot-as-volume-chain on LVM fixes Fiona Ebner
2025-07-24 13:59 ` [pve-devel] [PATCH qemu-server 1/6] blockdev: helper to add common options Fiona Ebner
2025-07-24 13:59 ` [pve-devel] [PATCH qemu-server 2/6] blockdev: fix discard Fiona Ebner
2025-07-24 13:59 ` [pve-devel] [PATCH qemu-server 3/6] tests: image convert: avoid hard-coded VM ID in result Fiona Ebner
2025-07-24 13:59 ` [pve-devel] [PATCH qemu-server 4/6] tests: image convert: properly set snapshot-as-volume-chain option Fiona Ebner
2025-07-24 13:59 ` [pve-devel] [PATCH qemu-server 5/6] tests: image convert: add tests where storages with 'snapshot-as-volume-chain' are the target Fiona Ebner
2025-07-24 13:59 ` [pve-devel] [PATCH qemu-server 6/6] fix #6543: use qcow2 'discard-no-unref' option when using snapshot-as-volume-chain Fiona Ebner
2025-07-24 18:01 ` DERUMIER, Alexandre via pve-devel
2025-07-25 7:38 ` Fabian Grünbichler [this message]
2025-07-25 8:24 ` Fiona Ebner
2025-07-25 7:40 ` [pve-devel] [PATCH-SERIES qemu-server 0/6] blockdev and snapshot-as-volume-chain on LVM fixes Fabian Grünbichler
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1753428793.n5phbfd5ll.astroid@yuna.none \
--to=f.gruenbichler@proxmox.com \
--cc=pve-devel@lists.proxmox.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.