[PATCH qemu-server 1/6] lvmqcow2: set disk write threshold

[Alexandre Derumier]

on vm start or when we hotplug a disk, we add a write threshold.

The threshold is: size of the lvm - (chunk_usage_percent * chunksize)

qemu will emit an event when the vm write on an offset higher than the threshold,
and the counter is reset to 0.
(So, we'll need to set threshold again when we extend the disk)

Signed-off-by: Alexandre Derumier <alexandre.derumier at groupe-cyllene.com>
---
 PVE/QemuServer.pm | 97 +++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 97 insertions(+)

diff --git a/PVE/QemuServer.pm b/PVE/QemuServer.pm
index 88c274d..3acb32e 100644
--- a/PVE/QemuServer.pm
+++ b/PVE/QemuServer.pm
@@ -4260,6 +4260,7 @@ sub vm_deviceplug {
 	    warn $@ if $@;
 	    die $err;
         }
+        eval { set_disks_write_threshold($storecfg, $conf->{pending}, $vmid, $deviceid) };
     } elsif ($deviceid =~ m/^(virtioscsi|scsihw)(\d+)$/) {
 	my $scsihw = defined($conf->{scsihw}) ? $conf->{scsihw} : "lsi";
 	my $pciaddr = print_pci_addr($deviceid, undef, $arch, $machine_type);
@@ -4289,6 +4290,7 @@ sub vm_deviceplug {
 	    warn $@ if $@;
 	    die $err;
         }
+        eval { PVE::QemuServer::set_disks_write_threshold($storecfg, $conf->{pending}, $vmid, $deviceid); };
     } elsif ($deviceid =~ m/^(net)(\d+)$/) {
 	return if !qemu_netdevadd($vmid, $conf, $arch, $device, $deviceid);
 
@@ -6069,6 +6071,7 @@ sub vm_start_nolock {
 	    qemu_set_link_status($vmid, $opt, 0) if $nicconf->{link_down};
 	}
 	add_nets_bridge_fdb($conf, $vmid);
+	PVE::QemuServer::set_disks_write_threshold($storecfg, $conf, $vmid);
     }
 
    if (!defined($conf->{balloon}) || $conf->{balloon}) {
@@ -8829,4 +8832,98 @@ sub delete_ifaces_ipams_ips {
     }
 }
 
+sub qemu_block_set_write_threshold {
+    my ($vmid, $nodename, $threshold) = @_;
+
+    print"set threshold $nodename $threshold\n";
+
+    PVE::QemuServer::mon_cmd(
+        $vmid,
+        "block-set-write-threshold",
+        'node-name' => $nodename,
+        'write-threshold' => int($threshold),
+    );
+}
+
+sub get_block_info {
+    my ($vmid, $disk, $block_info) = @_;
+
+    my $res = { deviceid => undef, blocknodeid => undef, wr_highest_offset => 0};
+
+    if($disk =~ m/block/) {
+	$res->{blocknodeid} = $disk;
+	for my $id (keys %$block_info) {
+	    if($block_info->{$id}->{parent}->{'node-name'} eq $disk) {
+		    $res->{deviceid} = $id;
+		    $res->{deviceid} =~ s/^drive-//;
+		    $res->{wr_highest_offset} = $block_info->{$id}->{parent}->{stats}->{wr_highest_offset};
+		    last;
+	    } elsif($block_info->{$id}->{parent}->{parent}->{'node-name'} eq $disk) {
+		    $res->{deviceid} = $id;
+		    $res->{deviceid} =~ s/^drive-//;
+		    $res->{wr_highest_offset} = $block_info->{$id}->{parent}->{parent}->{stats}->{wr_highest_offset};
+		    last;
+	    }
+	}
+    } else {
+	$res->{deviceid} = $disk;
+	#when backup is running, the chain of image is different
+	if($block_info->{"drive-$disk"}->{parent}->{parent}->{'node-name'} &&
+	    $block_info->{"drive-$disk"}->{parent}->{parent}->{'driver-specific'}->{driver} eq 'host_device'
+	) {
+	    $res->{blocknodeid} = $block_info->{"drive-$disk"}->{parent}->{parent}->{'node-name'};
+	    $res->{wr_highest_offset} = $block_info->{"drive-$disk"}->{parent}->{parent}->{stats}->{wr_highest_offset};
+	} elsif($block_info->{"drive-$disk"}->{parent}->{'node-name'} &&
+		$block_info->{"drive-$disk"}->{parent}->{'driver-specific'}->{driver} eq 'host_device'
+	) {
+	    $res->{blocknodeid} = $block_info->{"drive-$disk"}->{parent}->{'node-name'};
+	    $res->{wr_highest_offset} = $block_info->{"drive-$disk"}->{parent}->{stats}->{wr_highest_offset};
+	}
+    }
+    die "can't find blocknodeid" if !$res->{blocknodeid};
+    die "can't find devicedeid" if !$res->{deviceid};
+    return $res;
+}
+
+sub compute_write_threshold {
+    my ($size, $scfg) = @_;
+
+    my $chunksize = $scfg->{chunksize} // 1024 * 1024 * 1024;
+    my $chunk_pct_extension = $scfg->{chunk_pct_extension} // 0.5;
+
+    my $threshold = $size - ($chunksize * $chunk_pct_extension);
+
+    return $threshold;
+}
+
+sub set_disks_write_threshold {
+    my ($storecfg, $conf, $vmid, $deviceid) = @_;
+
+    $deviceid =~ s/^drive-// if $deviceid;
+
+    my $blockstats = mon_cmd($vmid, "query-blockstats");
+    $blockstats = { map { $_->{device} => $_ } $blockstats->@* };
+
+    PVE::QemuConfig->foreach_volume($conf, sub {
+	my ($ds, $drive) = @_;
+
+	return if $deviceid && $ds ne $deviceid;
+
+	my $volid = $drive->{file};
+	return if !$volid;
+
+	my ($sid, $volname) = PVE::Storage::parse_volume_id($volid, 1);
+	return if !$sid;
+
+	my $scfg = PVE::Storage::storage_config($storecfg, $sid);
+	return if $scfg->{type} ne 'lvmqcow2';
+
+	my $size = PVE::Storage::volume_size_info($storecfg, $volid, 5);
+	my $threshold = compute_write_threshold($size, $scfg);
+
+	my $blockinfo = get_block_info($vmid, $ds, $blockstats);
+	qemu_block_set_write_threshold($vmid, $blockinfo->{blocknodeid}, $threshold);
+    });
+}
+
 1;
-- 
2.39.2