[pve-devel] [PATCH qemu-server 08/11] enable live-restore for PBS

Stefan Reiter s.reiter at proxmox.com
Mon Jan 11 12:14:06 CET 2021


Enables live-restore functionality using the 'alloc-track' QEMU driver.
This allows starting a VM immediately when restoring from a PBS
snapshot. The snapshot is mounted into the VM, so it can boot from that,
while guest reads and a 'block-stream' job handle the restore in the
background.

If an error occurs, the VM is deleted and all data written during the
restore is lost.

The VM remains locked during the restore, which automatically prohibits
any modifications to the config while restoring. Some modifications
might potentially be safe, however, this is experimental enough that I
believe this would cause more bad stuff(tm) than actually satisfy any
use cases.

Pool handling is slightly adjusted so the VM can be added to the pool
before the restore starts.

Signed-off-by: Stefan Reiter <s.reiter at proxmox.com>
---

Looks better with -w

 PVE/API2/Qemu.pm  |  15 ++++-
 PVE/QemuServer.pm | 148 ++++++++++++++++++++++++++++++++++++++--------
 2 files changed, 137 insertions(+), 26 deletions(-)

diff --git a/PVE/API2/Qemu.pm b/PVE/API2/Qemu.pm
index e2d2d67..a02a354 100644
--- a/PVE/API2/Qemu.pm
+++ b/PVE/API2/Qemu.pm
@@ -490,6 +490,12 @@ __PACKAGE__->register_method({
 		    description => "Assign a unique random ethernet address.",
 		    requires => 'archive',
 		},
+		'live-restore' => {
+		    optional => 1,
+		    type => 'boolean',
+		    description => "Start the VM immediately from the backup and restore in background. PBS only.",
+		    requires => 'archive',
+		},
 		pool => {
 		    optional => 1,
 		    type => 'string', format => 'pve-poolid',
@@ -531,6 +537,10 @@ __PACKAGE__->register_method({
 	my $start_after_create = extract_param($param, 'start');
 	my $storage = extract_param($param, 'storage');
 	my $unique = extract_param($param, 'unique');
+	my $live_restore = extract_param($param, 'live-restore');
+
+	raise_param_exc({ 'start' => "cannot specify 'start' with 'live-restore'" })
+	    if $start_after_create && $live_restore;
 
 	if (defined(my $ssh_keys = $param->{sshkeys})) {
 		$ssh_keys = URI::Escape::uri_unescape($ssh_keys);
@@ -613,9 +623,12 @@ __PACKAGE__->register_method({
 		    pool => $pool,
 		    unique => $unique,
 		    bwlimit => $bwlimit,
+		    live => $live_restore,
 		};
 		if ($archive->{type} eq 'file' || $archive->{type} eq 'pipe') {
+		    die "live-restore is only compatible with PBS\n" if $live_restore;
 		    PVE::QemuServer::restore_file_archive($archive->{path} // '-', $vmid, $authuser, $restore_options);
+		    PVE::AccessControl::add_vm_to_pool($vmid, $pool) if $pool;
 		} elsif ($archive->{type} eq 'pbs') {
 		    PVE::QemuServer::restore_proxmox_backup_archive($archive->{volid}, $vmid, $authuser, $restore_options);
 		} else {
@@ -628,8 +641,6 @@ __PACKAGE__->register_method({
 		    eval { PVE::QemuServer::template_create($vmid, $restored_conf) };
 		    warn $@ if $@;
 		}
-
-		PVE::AccessControl::add_vm_to_pool($vmid, $pool) if $pool;
 	    };
 
 	    # ensure no old replication state are exists
diff --git a/PVE/QemuServer.pm b/PVE/QemuServer.pm
index 6de6ff6..4c34595 100644
--- a/PVE/QemuServer.pm
+++ b/PVE/QemuServer.pm
@@ -6103,7 +6103,7 @@ sub restore_proxmox_backup_archive {
 
     my $repo = PVE::PBSClient::get_repository($scfg);
 
-    # This is only used for `pbs-restore`!
+    # This is only used for `pbs-restore` and the QEMU PBS driver (live-restore)
     my $password = PVE::Storage::PBSPlugin::pbs_get_password($scfg, $storeid);
     local $ENV{PBS_PASSWORD} = $password;
     local $ENV{PBS_FINGERPRINT} = $fingerprint if defined($fingerprint);
@@ -6200,34 +6200,35 @@ sub restore_proxmox_backup_archive {
 	# allocate volumes
 	my $map = $restore_allocate_devices->($storecfg, $virtdev_hash, $vmid);
 
-	foreach my $virtdev (sort keys %$virtdev_hash) {
-	    my $d = $virtdev_hash->{$virtdev};
-	    next if $d->{is_cloudinit}; # no need to restore cloudinit
+	if (!$options->{live}) {
+	    foreach my $virtdev (sort keys %$virtdev_hash) {
+		my $d = $virtdev_hash->{$virtdev};
+		next if $d->{is_cloudinit}; # no need to restore cloudinit
 
-	    my $volid = $d->{volid};
+		my $volid = $d->{volid};
 
-	    my $path = PVE::Storage::path($storecfg, $volid);
+		my $path = PVE::Storage::path($storecfg, $volid);
 
-	    # This is the ONLY user of the PBS_ env vars set on top of this function!
-	    my $pbs_restore_cmd = [
-		'/usr/bin/pbs-restore',
-		'--repository', $repo,
-		$pbs_backup_name,
-		"$d->{devname}.img.fidx",
-		$path,
-		'--verbose',
-		];
+		my $pbs_restore_cmd = [
+		    '/usr/bin/pbs-restore',
+		    '--repository', $repo,
+		    $pbs_backup_name,
+		    "$d->{devname}.img.fidx",
+		    $path,
+		    '--verbose',
+		    ];
 
-	    push @$pbs_restore_cmd, '--format', $d->{format} if $d->{format};
-	    push @$pbs_restore_cmd, '--keyfile', $keyfile if -e $keyfile;
+		push @$pbs_restore_cmd, '--format', $d->{format} if $d->{format};
+		push @$pbs_restore_cmd, '--keyfile', $keyfile if -e $keyfile;
 
-	    if (PVE::Storage::volume_has_feature($storecfg, 'sparseinit', $volid)) {
-		push @$pbs_restore_cmd, '--skip-zero';
+		if (PVE::Storage::volume_has_feature($storecfg, 'sparseinit', $volid)) {
+		    push @$pbs_restore_cmd, '--skip-zero';
+		}
+
+		my $dbg_cmdstring = PVE::Tools::cmd2string($pbs_restore_cmd);
+		print "restore proxmox backup image: $dbg_cmdstring\n";
+		run_command($pbs_restore_cmd);
 	    }
-
-	    my $dbg_cmdstring = PVE::Tools::cmd2string($pbs_restore_cmd);
-	    print "restore proxmox backup image: $dbg_cmdstring\n";
-	    run_command($pbs_restore_cmd);
 	}
 
 	$fh->seek(0, 0) || die "seek failed - $!\n";
@@ -6244,7 +6245,9 @@ sub restore_proxmox_backup_archive {
     };
     my $err = $@;
 
-    $restore_deactivate_volumes->($storecfg, $devinfo);
+    if ($err || !$options->{live}) {
+	$restore_deactivate_volumes->($storecfg, $devinfo);
+    }
 
     rmtree $tmpdir;
 
@@ -6261,6 +6264,103 @@ sub restore_proxmox_backup_archive {
 
     eval { rescan($vmid, 1); };
     warn $@ if $@;
+
+    PVE::AccessControl::add_vm_to_pool($vmid, $options->{pool}) if $options->{pool};
+
+    if ($options->{live}) {
+	eval {
+	    # enable interrupts
+	    local $SIG{INT} =
+		local $SIG{TERM} =
+		local $SIG{QUIT} =
+		local $SIG{HUP} =
+		local $SIG{PIPE} = sub { die "interrupted by signal\n"; };
+
+	    my $conf = PVE::QemuConfig->load_config($vmid);
+	    die "cannot do live-restore for template\n"
+		if PVE::QemuConfig->is_template($conf);
+
+	    pbs_live_restore($vmid, $conf, $storecfg, $devinfo, $repo, $keyfile, $pbs_backup_name);
+	};
+
+	$err = $@;
+	if ($err) {
+	    warn "Detroying live-restore VM, all temporary data will be lost!\n";
+	    $restore_deactivate_volumes->($storecfg, $devinfo);
+	    $restore_destroy_volumes->($storecfg, $devinfo);
+	    unlink $conffile;
+	    die $err;
+	}
+    }
+}
+
+sub pbs_live_restore {
+    my ($vmid, $conf, $storecfg, $restored_disks, $repo, $keyfile, $snap) = @_;
+
+    print "Starting VM for live-restore\n";
+
+    my $pbs_backing = {};
+    foreach my $ds (keys %$restored_disks) {
+	$ds =~ m/^drive-(.*)$/;
+	$pbs_backing->{$1} = {
+	    repository => $repo,
+	    snapshot => $snap,
+	    archive => "$ds.img.fidx",
+	};
+	$pbs_backing->{$1}->{keyfile} = $keyfile if -e $keyfile;
+    }
+
+    eval {
+	# make sure HA doesn't interrupt our restore by stopping the VM
+	if (PVE::HA::Config::vm_is_ha_managed($vmid)) {
+	    my $cmd = ['ha-manager', 'set',  "vm:$vmid", '--state', 'started'];
+	    PVE::Tools::run_command($cmd);
+	}
+
+	# start VM with backing chain pointing to PBS backup, environment vars
+	# for PBS driver in QEMU (PBS_PASSWORD and PBS_FINGERPRINT) are already
+	# set by our caller
+	PVE::QemuServer::vm_start_nolock(
+	    $storecfg,
+	    $vmid,
+	    $conf,
+	    {
+		paused => 1,
+		'pbs-backing' => $pbs_backing,
+	    },
+	    {},
+	);
+
+	# begin streaming, i.e. data copy from PBS to target disk for every vol,
+	# this will effectively collapse the backing image chain consisting of
+	# [target <- alloc-track -> PBS snapshot] to just [target] (alloc-track
+	# removes itself once all backing images vanish with 'auto-remove=on')
+	my $jobs = {};
+	foreach my $ds (keys %$restored_disks) {
+	    my $job_id = "restore-$ds";
+	    mon_cmd($vmid, 'block-stream',
+		'job-id' => $job_id,
+		device => "$ds",
+	    );
+	    $jobs->{$job_id} = {};
+	}
+
+	mon_cmd($vmid, 'cont');
+	qemu_drive_job_monitor('stream', $vmid, undef, $jobs, 'auto');
+
+	# all jobs finished, remove blockdevs now to disconnect from PBS
+	foreach my $ds (keys %$restored_disks) {
+	    mon_cmd($vmid, 'blockdev-del', 'node-name' => "$ds-pbs");
+	}
+    };
+
+    my $err = $@;
+
+    if ($err) {
+	warn "An error occured during live-restore: $err\n";
+	_do_vm_stop($storecfg, $vmid, 1, 1, 10, 0, 1);
+	die "live-restore failed\n";
+    }
 }
 
 sub restore_vma_archive {
-- 
2.20.1





More information about the pve-devel mailing list