[qemu-server.git] / PVE / QemuMigrate.pm

package PVE::QemuMigrate;

use strict;
use warnings;
use PVE::AbstractMigrate;
use IO::File;
use IPC::Open2;
use PVE::INotify;
use PVE::Cluster;
use PVE::Storage;
use PVE::QemuServer;
use Time::HiRes qw( usleep );

use base qw(PVE::AbstractMigrate);

sub fork_command_pipe {
    my ($self, $cmd) = @_;

    my $reader = IO::File->new();
    my $writer = IO::File->new();

    my $orig_pid = $$;

    my $cpid;

    eval { $cpid = open2($reader, $writer, @$cmd); };

    my $err = $@;

    # catch exec errors
    if ($orig_pid != $$) {
	$self->log('err', "can't fork command pipe\n");
	POSIX::_exit(1);
	kill('KILL', $$);
    }

    die $err if $err;

    return { writer => $writer, reader => $reader, pid => $cpid };
}

sub finish_command_pipe {
    my ($self, $cmdpipe, $timeout) = @_;

    my $writer = $cmdpipe->{writer};
    my $reader = $cmdpipe->{reader};

    $writer->close();
    $reader->close();

    my $cpid = $cmdpipe->{pid};

    if ($timeout) {
	for (my $i = 0; $i < $timeout; $i++) {
	    return if !PVE::ProcFSTools::check_process_running($cpid);
	    sleep(1);
	}
    }

    $self->log('info', "ssh tunnel still running - terminating now with SIGTERM\n");
    kill(15, $cpid);

    # wait again
    for (my $i = 0; $i < 10; $i++) {
	return if !PVE::ProcFSTools::check_process_running($cpid);
	sleep(1);
    }

    $self->log('info', "ssh tunnel still running - terminating now with SIGKILL\n");
    kill 9, $cpid;
    sleep 1;
}

sub fork_tunnel {
    my ($self, $nodeip, $lport, $rport) = @_;

    my $cmd = [@{$self->{rem_ssh}}, '-L', "$lport:localhost:$rport",
	       'qm', 'mtunnel' ];

    my $tunnel = $self->fork_command_pipe($cmd);

    my $reader = $tunnel->{reader};

    my $helo;
    eval {
	PVE::Tools::run_with_timeout(60, sub { $helo = <$reader>; });
	die "no reply\n" if !$helo;
	die "no quorum on target node\n" if $helo =~ m/^no quorum$/;
	die "got strange reply from mtunnel ('$helo')\n"
	    if $helo !~ m/^tunnel online$/;
    };
    my $err = $@;

    if ($err) {
	$self->finish_command_pipe($tunnel);
	die "can't open migration tunnel - $err";
    }
    return $tunnel;
}

sub finish_tunnel {
    my ($self, $tunnel) = @_;

    my $writer = $tunnel->{writer};

    eval {
	PVE::Tools::run_with_timeout(30, sub {
	    print $writer "quit\n";
	    $writer->flush();
	});
    };
    my $err = $@;

    $self->finish_command_pipe($tunnel, 30);

    die $err if $err;
}

sub lock_vm {
    my ($self, $vmid, $code, @param) = @_;

    return PVE::QemuServer::lock_config($vmid, $code, @param);
}

sub prepare {
    my ($self, $vmid) = @_;

    my $online = $self->{opts}->{online};

    $self->{storecfg} = PVE::Storage::config();

    # test is VM exist
    my $conf = $self->{vmconf} = PVE::QemuServer::load_config($vmid);

    PVE::QemuServer::check_lock($conf);

    my $running = 0;
    if (my $pid = PVE::QemuServer::check_running($vmid)) {
	die "cant migrate running VM without --online\n" if !$online;
	$running = $pid;
    }

    if (my $loc_res = PVE::QemuServer::check_local_resources($conf, 1)) {
	if ($self->{running} || !$self->{opts}->{force}) {
	    die "can't migrate VM which uses local devices\n";
	} else {
	    $self->log('info', "migrating VM which uses local devices");
	}
    }

    # activate volumes
    my $vollist = PVE::QemuServer::get_vm_volumes($conf);
    PVE::Storage::activate_volumes($self->{storecfg}, $vollist);

    # fixme: check if storage is available on both nodes

    # test ssh connection
    my $cmd = [ @{$self->{rem_ssh}}, '/bin/true' ];
    eval { $self->cmd_quiet($cmd); };
    die "Can't connect to destination address using public key\n" if $@;

    return $running;
}

sub sync_disks {
    my ($self, $vmid) = @_;

    $self->log('info', "copying disk images");

    my $conf = $self->{vmconf};

    $self->{volumes} = [];

    my $res = [];

    eval {

	my $volhash = {};
	my $cdromhash = {};

	my $sharedvm = 1;

	my @sids = PVE::Storage::storage_ids($self->{storecfg});
        foreach my $storeid (@sids) {
	    my $scfg = PVE::Storage::storage_config($self->{storecfg}, $storeid);
            next if $scfg->{shared};
	    next if !PVE::Storage::storage_check_enabled($self->{storecfg}, $storeid, undef, 1);

            # get list from PVE::Storage (for unused volumes)
            my $dl = PVE::Storage::vdisk_list($self->{storecfg}, $storeid, $vmid);
            PVE::Storage::foreach_volid($dl, sub {
                my ($volid, $sid, $volname) = @_;

                # check if storage is available on target node
                PVE::Storage::storage_check_node($self->{storecfg}, $sid, $self->{node});

                $volhash->{$volid} = 1;
		$sharedvm = 0; # there is a non-shared disk
            });
        }

	# and add used, owned/non-shared disks (just to be sure we have all)

	PVE::QemuServer::foreach_volid($conf, sub {
	    my ($volid, $is_cdrom) = @_;

	    return if !$volid;

	    die "cant migrate local file/device '$volid'\n" if $volid =~ m|^/|;

	    if ($is_cdrom) {
		die "cant migrate local cdrom drive\n" if $volid eq 'cdrom';
		return if $volid eq 'none';
		$cdromhash->{$volid} = 1;
	    }

	    my ($sid, $volname) = PVE::Storage::parse_volume_id($volid);

	    # check if storage is available on both nodes
	    my $scfg = PVE::Storage::storage_check_node($self->{storecfg}, $sid);
	    PVE::Storage::storage_check_node($self->{storecfg}, $sid, $self->{node});

	    return if $scfg->{shared};

	    die "can't migrate local cdrom '$volid'\n" if $cdromhash->{$volid};

	    $sharedvm = 0;

	    my ($path, $owner) = PVE::Storage::path($self->{storecfg}, $volid);

	    die "can't migrate volume '$volid' - owned by other VM (owner = VM $owner)\n"
		if !$owner || ($owner != $self->{vmid});

	    $volhash->{$volid} = 1;
	});

	if ($self->{running} && !$sharedvm) {
	    die "can't do online migration - VM uses local disks\n";
	}

	# do some checks first
	foreach my $volid (keys %$volhash) {
	    my ($sid, $volname) = PVE::Storage::parse_volume_id($volid);
	    my $scfg =  PVE::Storage::storage_config($self->{storecfg}, $sid);

	    die "can't migrate '$volid' - storagy type '$scfg->{type}' not supported\n"
		if $scfg->{type} ne 'dir';
	}

	foreach my $volid (keys %$volhash) {
	    my ($sid, $volname) = PVE::Storage::parse_volume_id($volid);
	    push @{$self->{volumes}}, $volid;
	    PVE::Storage::storage_migrate($self->{storecfg}, $volid, $self->{nodeip}, $sid);
	}
    };
    die "Failed to sync data - $@" if $@;
}

sub phase1 {
    my ($self, $vmid) = @_;

    $self->log('info', "starting migration of VM $vmid to node '$self->{node}' ($self->{nodeip})");

    my $conf = $self->{vmconf};

    # set migrate lock in config file
    $conf->{lock} = 'migrate';
    PVE::QemuServer::update_config_nolock($vmid, $conf, 1);

    sync_disks($self, $vmid);

};

sub phase1_cleanup {
    my ($self, $vmid, $err) = @_;

    $self->log('info', "aborting phase 1 - cleanup resources");

    my $conf = $self->{vmconf};
    delete $conf->{lock};
    eval { PVE::QemuServer::update_config_nolock($vmid, $conf, 1) };
    if (my $err = $@) {
	$self->log('err', $err);
    }

    if ($self->{volumes}) {
	foreach my $volid (@{$self->{volumes}}) {
	    $self->log('err', "found stale volume copy '$volid' on node '$self->{node}'");
	    # fixme: try to remove ?
	}
    }
}

sub phase2 {
    my ($self, $vmid) = @_;

    my $conf = $self->{vmconf};

    $self->log('info', "starting VM $vmid on remote node '$self->{node}'");

    my $rport;

    my $nodename = PVE::INotify::nodename();

    ## start on remote node
    my $cmd = [@{$self->{rem_ssh}}, 'qm', 'start',
               $vmid, '--stateuri', 'tcp', '--skiplock', '--migratedfrom', $nodename];

    PVE::Tools::run_command($cmd, outfunc => sub {
	my $line = shift;

	if ($line =~ m/^migration listens on port (\d+)$/) {
	    $rport = $1;
	}
    }, errfunc => sub {
	my $line = shift;
	$self->log('info', $line);
    });

    die "unable to detect remote migration port\n" if !$rport;

    $self->log('info', "starting migration tunnel");

    ## create tunnel to remote port
    my $lport = PVE::QemuServer::next_migrate_port();
    $self->{tunnel} = $self->fork_tunnel($self->{nodeip}, $lport, $rport);

    $self->log('info', "starting online/live migration on port $lport");
    # start migration

    my $start = time();

    # load_defaults
    my $defaults = PVE::QemuServer::load_defaults();

    # always set migrate speed (overwrite kvm default of 32m)
    # we set a very hight default of 8192m which is basically unlimited
    my $migrate_speed = $defaults->{migrate_speed} || 8192;
    $migrate_speed = $conf->{migrate_speed} || $migrate_speed;
    $migrate_speed = $migrate_speed * 1048576;
    $self->log('info', "migrate_set_speed: $migrate_speed");
    eval {
        PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate_set_speed", value => int($migrate_speed));
    };
    $self->log('info', "migrate_set_speed error: $@") if $@;

    my $migrate_downtime = $defaults->{migrate_downtime};
    $migrate_downtime = $conf->{migrate_downtime} if defined($conf->{migrate_downtime});
    if (defined($migrate_downtime)) {
	$self->log('info', "migrate_set_downtime: $migrate_downtime");
	eval {
	    PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate_set_downtime", value => int($migrate_downtime*100)/100);
	};
	$self->log('info', "migrate_set_downtime error: $@") if $@;
    }

    my $capabilities = {};
    $capabilities->{capability} =  "xbzrle";
    $capabilities->{state} = JSON::false;

    eval {
	PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate-set-capabilities", capabilities => [$capabilities]);
    };

    #set cachesize 10% of the total memory
    my $cachesize = int($conf->{memory}*1048576/10);
    eval {
	PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate-set-cache-size", value => $cachesize);
    };

    eval {
        PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate", uri => "tcp:localhost:$lport");
    };
    my $merr = $@;

    my $lstat = 0;
    my $usleep = 2000000;
    my $i = 0;
    my $err_count = 0;
    my $lastrem = undef;
    my $downtimecounter = 0;
    while (1) {
	$i++;
	my $avglstat = $lstat/$i if $lstat;

	usleep($usleep);
	my $stat;
	eval {
	    $stat = PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "query-migrate");
	};
	if (my $err = $@) {
	    $err_count++;
	    warn "query migrate failed: $err\n";
	    if ($err_count <= 5) {
		usleep(1000000);
		next;
	    }
	    die "too many query migrate failures - aborting\n";
	}
	if ($stat->{status} =~ m/^(active|completed|failed|cancelled)$/im) {
	    $merr = undef;
	    $err_count = 0;
	    if ($stat->{status} eq 'completed') {
		my $delay = time() - $start;
		if ($delay > 0) {
		    my $mbps = sprintf "%.2f", $conf->{memory}/$delay;
		    my $downtime = $stat->{downtime} || 0;
		    $self->log('info', "migration speed: $mbps MB/s - downtime $downtime ms");
		}
	    }

	    if ($stat->{status} eq 'failed' || $stat->{status} eq 'cancelled') {
		die "aborting\n"
	    }

	    if ($stat->{status} ne 'active') {
		$self->log('info', "migration status: $stat->{status}");
		last;
	    }

	    if ($stat->{ram}->{transferred} ne $lstat) {
		my $trans = $stat->{ram}->{transferred} || 0;
		my $rem = $stat->{ram}->{remaining} || 0;
		my $total = $stat->{ram}->{total} || 0;
		my $xbzrlecachesize = $stat->{"xbzrle-cache"}->{"cache-size"} || 0;
		my $xbzrlebytes = $stat->{"xbzrle-cache"}->{"bytes"} || 0;
		my $xbzrlepages = $stat->{"xbzrle-cache"}->{"pages"} || 0;
		my $xbzrlecachemiss = $stat->{"xbzrle-cache"}->{"cache-miss"} || 0;
		my $xbzrleoverflow = $stat->{"xbzrle-cache"}->{"overflow"} || 0;
		my $expected_downtime = $stat->{"expected-downtime"} || 0;
		#reduce sleep if remainig memory if lower than the everage transfert 
		$usleep = 300000 if $avglstat && $rem < $avglstat;

		$self->log('info', "migration status: $stat->{status} (transferred ${trans}, " .
			   "remaining ${rem}), total ${total}) , expected downtime ${expected_downtime}");

		#$self->log('info', "migration xbzrle cachesize: ${xbzrlecachesize} transferred ${xbzrlebytes} pages ${xbzrlepages} cachemiss ${xbzrlecachemiss} overflow ${xbzrleoverflow}");
		if (($lastrem  && $rem > $lastrem ) || ($rem == 0)) {
		    $downtimecounter++;
		}
		$lastrem = $rem;

		if ($downtimecounter > 5) {
		    $downtimecounter = 0;
		    $migrate_downtime *= 2;
		    $self->log('info', "migrate_set_downtime: $migrate_downtime");
		    eval {
			PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate_set_downtime", value => int($migrate_downtime*100)/100);
		    };
		    $self->log('info', "migrate_set_downtime error: $@") if $@;
            	}

	    }


	    $lstat = $stat->{ram}->{transferred};
	    
	} else {
	    die $merr if $merr;
	    die "unable to parse migration status '$stat->{status}' - aborting\n";
	}
    }
}

sub phase2_cleanup {
    my ($self, $vmid, $err) = @_;

    return if !$self->{errors};
    $self->{phase2errors} = 1;

    $self->log('info', "aborting phase 2 - cleanup resources");

    my $conf = $self->{vmconf};
    delete $conf->{lock};
    eval { PVE::QemuServer::update_config_nolock($vmid, $conf, 1) };
    if (my $err = $@) {
        $self->log('err', $err);
    }

    # cleanup ressources on target host
    my $nodename = PVE::INotify::nodename();
 
    my $cmd = [@{$self->{rem_ssh}}, 'qm', 'stop', $vmid, '--skiplock', '--migratedfrom', $nodename];
    eval{ PVE::Tools::run_command($cmd, outfunc => sub {}, errfunc => sub {}) };
    if (my $err = $@) {
        $self->log('err', $err);
        $self->{errors} = 1;
    }
}

sub phase3 {
    my ($self, $vmid) = @_;

    my $volids = $self->{volumes};
    return if $self->{phase2errors};

    # destroy local copies
    foreach my $volid (@$volids) {
	eval { PVE::Storage::vdisk_free($self->{storecfg}, $volid); };
	if (my $err = $@) {
	    $self->log('err', "removing local copy of '$volid' failed - $err");
	    $self->{errors} = 1;
	    last if $err =~ /^interrupted by signal$/;
	}
    }
}

sub phase3_cleanup {
    my ($self, $vmid, $err) = @_;

    my $conf = $self->{vmconf};
    return if $self->{phase2errors};

    # move config to remote node
    my $conffile = PVE::QemuServer::config_file($vmid);
    my $newconffile = PVE::QemuServer::config_file($vmid, $self->{node});

    die "Failed to move config to node '$self->{node}' - rename failed: $!\n"
        if !rename($conffile, $newconffile);

    # now that config file is move, we can resume vm on target if livemigrate
    if ($self->{tunnel}) {
	my $cmd = [@{$self->{rem_ssh}}, 'qm', 'resume', $vmid, '--skiplock'];
	eval{ PVE::Tools::run_command($cmd, outfunc => sub {}, errfunc => sub {}) };
	if (my $err = $@) {
	    $self->log('err', $err);
	    $self->{errors} = 1;
	}
    }

    # always stop local VM
    eval { PVE::QemuServer::vm_stop($self->{storecfg}, $vmid, 1, 1); };
    if (my $err = $@) {
	$self->log('err', "stopping vm failed - $err");
	$self->{errors} = 1;
    }

    if ($self->{tunnel}) {
	eval { finish_tunnel($self, $self->{tunnel});  };
	if (my $err = $@) {
	    $self->log('err', $err);
	    $self->{errors} = 1;
	}
    }

    # always deactivate volumes - avoid lvm LVs to be active on several nodes
    eval {
	my $vollist = PVE::QemuServer::get_vm_volumes($conf);
	PVE::Storage::deactivate_volumes($self->{storecfg}, $vollist);
    };
    if (my $err = $@) {
	$self->log('err', $err);
	$self->{errors} = 1;
    }

    # clear migrate lock
    my $cmd = [ @{$self->{rem_ssh}}, 'qm', 'unlock', $vmid ];
    $self->cmd_logerr($cmd, errmsg => "failed to clear migrate lock");
}

sub final_cleanup {
    my ($self, $vmid) = @_;

    # nothing to do
}

1;
Commit	Line	Data
3ea94c60	1	package PVE::QemuMigrate;
1ef75254	2
1e3baf05	3	use strict;
3ea94c60	4	use warnings;
16e903f2	5	use PVE::AbstractMigrate;
3ea94c60	6	use IO::File;
1e3baf05	7	use IPC::Open2;
3ea94c60 DM	8	use PVE::INotify;
3ea94c60 DM	9	use PVE::Cluster;
1e3baf05	10	use PVE::Storage;
3ea94c60	11	use PVE::QemuServer;
e52bd94c	12	use Time::HiRes qw( usleep );
1e3baf05	13
16e903f2	14	use base qw(PVE::AbstractMigrate);
1e3baf05	15
1ef75254	16	sub fork_command_pipe {
46a84fd4	17	my ($self, $cmd) = @_;
19672434	18
1ef75254 DM	19	my $reader = IO::File->new();
	20	my $writer = IO::File->new();
	21
	22	my $orig_pid = $$;
	23
	24	my $cpid;
	25
	26	eval { $cpid = open2($reader, $writer, @$cmd); };
	27
	28	my $err = $@;
	29
	30	# catch exec errors
	31	if ($orig_pid != $$) {
46a84fd4	32	$self->log('err', "can't fork command pipe\n");
19672434 DM	33	POSIX::_exit(1);
19672434 DM	34	kill('KILL', $$);
1ef75254 DM	35	}
	36
	37	die $err if $err;
	38
	39	return { writer => $writer, reader => $reader, pid => $cpid };
	40	}
	41
19672434	42	sub finish_command_pipe {
97439670	43	my ($self, $cmdpipe, $timeout) = @_;
1ef75254 DM	44
	45	my $writer = $cmdpipe->{writer};
	46	my $reader = $cmdpipe->{reader};
	47
	48	$writer->close();
	49	$reader->close();
	50
	51	my $cpid = $cmdpipe->{pid};
	52
97439670 DM	53	if ($timeout) {
	54	for (my $i = 0; $i < $timeout; $i++) {
	55	return if !PVE::ProcFSTools::check_process_running($cpid);
	56	sleep(1);
	57	}
	58	}
	59
	60	$self->log('info', "ssh tunnel still running - terminating now with SIGTERM\n");
	61	kill(15, $cpid);
1ef75254	62
97439670 DM	63	# wait again
	64	for (my $i = 0; $i < 10; $i++) {
	65	return if !PVE::ProcFSTools::check_process_running($cpid);
	66	sleep(1);
	67	}
	68
	69	$self->log('info', "ssh tunnel still running - terminating now with SIGKILL\n");
	70	kill 9, $cpid;
	71	sleep 1;
1ef75254 DM	72	}
1ef75254 DM	73
1e3baf05	74	sub fork_tunnel {
16e903f2	75	my ($self, $nodeip, $lport, $rport) = @_;
1e3baf05	76
16e903f2	77	my $cmd = [@{$self->{rem_ssh}}, '-L', "$lport:localhost:$rport",
1e3baf05	78	'qm', 'mtunnel' ];
19672434	79
46a84fd4	80	my $tunnel = $self->fork_command_pipe($cmd);
1e3baf05 DM	81
	82	my $reader = $tunnel->{reader};
	83
	84	my $helo;
19672434	85	eval {
17eed025	86	PVE::Tools::run_with_timeout(60, sub { $helo = <$reader>; });
1e3baf05	87	die "no reply\n" if !$helo;
1ef75254	88	die "no quorum on target node\n" if $helo =~ m/^no quorum$/;
19672434	89	die "got strange reply from mtunnel ('$helo')\n"
1e3baf05 DM	90	if $helo !~ m/^tunnel online$/;
	91	};
	92	my $err = $@;
	93
	94	if ($err) {
46a84fd4	95	$self->finish_command_pipe($tunnel);
1e3baf05 DM	96	die "can't open migration tunnel - $err";
	97	}
	98	return $tunnel;
	99	}
	100
19672434	101	sub finish_tunnel {
16e903f2	102	my ($self, $tunnel) = @_;
1e3baf05 DM	103
	104	my $writer = $tunnel->{writer};
	105
19672434	106	eval {
17eed025	107	PVE::Tools::run_with_timeout(30, sub {
1e3baf05 DM	108	print $writer "quit\n";
1e3baf05 DM	109	$writer->flush();
19672434	110	});
1e3baf05 DM	111	};
1e3baf05 DM	112	my $err = $@;
19672434	113
97439670	114	$self->finish_command_pipe($tunnel, 30);
19672434	115
1e3baf05 DM	116	die $err if $err;
	117	}
	118
16e903f2 DM	119	sub lock_vm {
16e903f2 DM	120	my ($self, $vmid, $code, @param) = @_;
f5eb281a	121
16e903f2 DM	122	return PVE::QemuServer::lock_config($vmid, $code, @param);
16e903f2 DM	123	}
ff1a2432	124
16e903f2 DM	125	sub prepare {
16e903f2 DM	126	my ($self, $vmid) = @_;
ff1a2432	127
16e903f2	128	my $online = $self->{opts}->{online};
3ea94c60	129
16e903f2	130	$self->{storecfg} = PVE::Storage::config();
3ea94c60	131
16e903f2 DM	132	# test is VM exist
16e903f2 DM	133	my $conf = $self->{vmconf} = PVE::QemuServer::load_config($vmid);
3ea94c60	134
16e903f2	135	PVE::QemuServer::check_lock($conf);
3ea94c60	136
16e903f2 DM	137	my $running = 0;
	138	if (my $pid = PVE::QemuServer::check_running($vmid)) {
	139	die "cant migrate running VM without --online\n" if !$online;
	140	$running = $pid;
3ea94c60 DM	141	}
3ea94c60 DM	142
16e903f2 DM	143	if (my $loc_res = PVE::QemuServer::check_local_resources($conf, 1)) {
	144	if ($self->{running} \|\| !$self->{opts}->{force}) {
	145	die "can't migrate VM which uses local devices\n";
	146	} else {
	147	$self->log('info', "migrating VM which uses local devices");
	148	}
3ea94c60 DM	149	}
3ea94c60 DM	150
ff1a2432 DM	151	# activate volumes
ff1a2432 DM	152	my $vollist = PVE::QemuServer::get_vm_volumes($conf);
16e903f2 DM	153	PVE::Storage::activate_volumes($self->{storecfg}, $vollist);
	154
	155	# fixme: check if storage is available on both nodes
3ea94c60 DM	156
3ea94c60 DM	157	# test ssh connection
16e903f2 DM	158	my $cmd = [ @{$self->{rem_ssh}}, '/bin/true' ];
16e903f2 DM	159	eval { $self->cmd_quiet($cmd); };
3ea94c60	160	die "Can't connect to destination address using public key\n" if $@;
ff1a2432	161
16e903f2	162	return $running;
3ea94c60 DM	163	}
	164
	165	sub sync_disks {
16e903f2 DM	166	my ($self, $vmid) = @_;
	167
	168	$self->log('info', "copying disk images");
3ea94c60	169
16e903f2 DM	170	my $conf = $self->{vmconf};
	171
	172	$self->{volumes} = [];
3ea94c60 DM	173
	174	my $res = [];
	175
	176	eval {
	177
	178	my $volhash = {};
	179	my $cdromhash = {};
	180
a06c7f7e DM	181	my $sharedvm = 1;
a06c7f7e DM	182
522c8f97 DM	183	my @sids = PVE::Storage::storage_ids($self->{storecfg});
	184	foreach my $storeid (@sids) {
	185	my $scfg = PVE::Storage::storage_config($self->{storecfg}, $storeid);
	186	next if $scfg->{shared};
373ea579 DM	187	next if !PVE::Storage::storage_check_enabled($self->{storecfg}, $storeid, undef, 1);
373ea579 DM	188
80b2cbd1 AD	189	# get list from PVE::Storage (for unused volumes)
	190	my $dl = PVE::Storage::vdisk_list($self->{storecfg}, $storeid, $vmid);
	191	PVE::Storage::foreach_volid($dl, sub {
	192	my ($volid, $sid, $volname) = @_;
	193
373ea579	194	# check if storage is available on target node
80b2cbd1 AD	195	PVE::Storage::storage_check_node($self->{storecfg}, $sid, $self->{node});
	196
	197	$volhash->{$volid} = 1;
a06c7f7e	198	$sharedvm = 0; # there is a non-shared disk
80b2cbd1 AD	199	});
80b2cbd1 AD	200	}
3ea94c60	201
d5769dc2	202	# and add used, owned/non-shared disks (just to be sure we have all)
3ea94c60	203
d5769dc2 DM	204	PVE::QemuServer::foreach_volid($conf, sub {
d5769dc2 DM	205	my ($volid, $is_cdrom) = @_;
3ea94c60	206
3ea94c60 DM	207	return if !$volid;
	208
	209	die "cant migrate local file/device '$volid'\n" if $volid =~ m\|^/\|;
	210
d5769dc2	211	if ($is_cdrom) {
3ea94c60 DM	212	die "cant migrate local cdrom drive\n" if $volid eq 'cdrom';
	213	return if $volid eq 'none';
	214	$cdromhash->{$volid} = 1;
	215	}
	216
	217	my ($sid, $volname) = PVE::Storage::parse_volume_id($volid);
	218
16e903f2 DM	219	# check if storage is available on both nodes
	220	my $scfg = PVE::Storage::storage_check_node($self->{storecfg}, $sid);
	221	PVE::Storage::storage_check_node($self->{storecfg}, $sid, $self->{node});
3ea94c60 DM	222
	223	return if $scfg->{shared};
	224
	225	die "can't migrate local cdrom '$volid'\n" if $cdromhash->{$volid};
	226
	227	$sharedvm = 0;
	228
16e903f2	229	my ($path, $owner) = PVE::Storage::path($self->{storecfg}, $volid);
3ea94c60 DM	230
3ea94c60 DM	231	die "can't migrate volume '$volid' - owned by other VM (owner = VM $owner)\n"
16e903f2	232	if !$owner \|\| ($owner != $self->{vmid});
3ea94c60 DM	233
	234	$volhash->{$volid} = 1;
	235	});
	236
16e903f2	237	if ($self->{running} && !$sharedvm) {
3ea94c60 DM	238	die "can't do online migration - VM uses local disks\n";
	239	}
	240
	241	# do some checks first
	242	foreach my $volid (keys %$volhash) {
	243	my ($sid, $volname) = PVE::Storage::parse_volume_id($volid);
16e903f2	244	my $scfg = PVE::Storage::storage_config($self->{storecfg}, $sid);
3ea94c60 DM	245
	246	die "can't migrate '$volid' - storagy type '$scfg->{type}' not supported\n"
	247	if $scfg->{type} ne 'dir';
	248	}
	249
	250	foreach my $volid (keys %$volhash) {
	251	my ($sid, $volname) = PVE::Storage::parse_volume_id($volid);
16e903f2 DM	252	push @{$self->{volumes}}, $volid;
16e903f2 DM	253	PVE::Storage::storage_migrate($self->{storecfg}, $volid, $self->{nodeip}, $sid);
3ea94c60 DM	254	}
	255	};
	256	die "Failed to sync data - $@" if $@;
	257	}
	258
1e3baf05	259	sub phase1 {
16e903f2	260	my ($self, $vmid) = @_;
1e3baf05	261
16e903f2	262	$self->log('info', "starting migration of VM $vmid to node '$self->{node}' ($self->{nodeip})");
1e3baf05	263
16e903f2	264	my $conf = $self->{vmconf};
1e3baf05 DM	265
1e3baf05 DM	266	# set migrate lock in config file
1858638f DM	267	$conf->{lock} = 'migrate';
1858638f DM	268	PVE::QemuServer::update_config_nolock($vmid, $conf, 1);
1e3baf05	269
16e903f2	270	sync_disks($self, $vmid);
1ef75254	271
1e3baf05 DM	272	};
1e3baf05 DM	273
16e903f2 DM	274	sub phase1_cleanup {
	275	my ($self, $vmid, $err) = @_;
	276
	277	$self->log('info', "aborting phase 1 - cleanup resources");
	278
1858638f DM	279	my $conf = $self->{vmconf};
	280	delete $conf->{lock};
	281	eval { PVE::QemuServer::update_config_nolock($vmid, $conf, 1) };
16e903f2 DM	282	if (my $err = $@) {
	283	$self->log('err', $err);
	284	}
f5eb281a	285
16e903f2 DM	286	if ($self->{volumes}) {
	287	foreach my $volid (@{$self->{volumes}}) {
	288	$self->log('err', "found stale volume copy '$volid' on node '$self->{node}'");
	289	# fixme: try to remove ?
	290	}
	291	}
	292	}
	293
1e3baf05	294	sub phase2 {
16e903f2	295	my ($self, $vmid) = @_;
1e3baf05	296
16e903f2 DM	297	my $conf = $self->{vmconf};
16e903f2 DM	298
46a84fd4	299	$self->log('info', "starting VM $vmid on remote node '$self->{node}'");
1e3baf05 DM	300
	301	my $rport;
	302
7e8dcf2c AD	303	my $nodename = PVE::INotify::nodename();
7e8dcf2c AD	304
19672434	305	## start on remote node
7e8dcf2c AD	306	my $cmd = [@{$self->{rem_ssh}}, 'qm', 'start',
7e8dcf2c AD	307	$vmid, '--stateuri', 'tcp', '--skiplock', '--migratedfrom', $nodename];
1e3baf05	308
72afda82	309	PVE::Tools::run_command($cmd, outfunc => sub {
1e3baf05 DM	310	my $line = shift;
	311
	312	if ($line =~ m/^migration listens on port (\d+)$/) {
	313	$rport = $1;
	314	}
ab399b7c AD	315	}, errfunc => sub {
	316	my $line = shift;
	317	$self->log('info', $line);
	318	});
1e3baf05 DM	319
	320	die "unable to detect remote migration port\n" if !$rport;
	321
16e903f2	322	$self->log('info', "starting migration tunnel");
1ef75254	323
1e3baf05	324	## create tunnel to remote port
1ef75254	325	my $lport = PVE::QemuServer::next_migrate_port();
16e903f2	326	$self->{tunnel} = $self->fork_tunnel($self->{nodeip}, $lport, $rport);
1e3baf05	327
d68afb26	328	$self->log('info', "starting online/live migration on port $lport");
1e3baf05 DM	329	# start migration
	330
	331	my $start = time();
e18b0b99	332
3beb415b AD	333	# load_defaults
	334	my $defaults = PVE::QemuServer::load_defaults();
	335
	336	# always set migrate speed (overwrite kvm default of 32m)
	337	# we set a very hight default of 8192m which is basically unlimited
	338	my $migrate_speed = $defaults->{migrate_speed} \|\| 8192;
	339	$migrate_speed = $conf->{migrate_speed} \|\| $migrate_speed;
	340	$migrate_speed = $migrate_speed * 1048576;
	341	$self->log('info', "migrate_set_speed: $migrate_speed");
	342	eval {
	343	PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate_set_speed", value => int($migrate_speed));
	344	};
	345	$self->log('info', "migrate_set_speed error: $@") if $@;
	346
	347	my $migrate_downtime = $defaults->{migrate_downtime};
	348	$migrate_downtime = $conf->{migrate_downtime} if defined($conf->{migrate_downtime});
	349	if (defined($migrate_downtime)) {
	350	$self->log('info', "migrate_set_downtime: $migrate_downtime");
	351	eval {
865ef132	352	PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate_set_downtime", value => int($migrate_downtime*100)/100);
3beb415b AD	353	};
	354	$self->log('info', "migrate_set_downtime error: $@") if $@;
	355	}
	356
e18b0b99 AD	357	my $capabilities = {};
e18b0b99 AD	358	$capabilities->{capability} = "xbzrle";
972511a0	359	$capabilities->{state} = JSON::false;
e18b0b99 AD	360
	361	eval {
	362	PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate-set-capabilities", capabilities => [$capabilities]);
	363	};
	364
	365	#set cachesize 10% of the total memory
	366	my $cachesize = int($conf->{memory}*1048576/10);
	367	eval {
	368	PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate-set-cache-size", value => $cachesize);
	369	};
	370
5a7835f5 AD	371	eval {
	372	PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate", uri => "tcp:localhost:$lport");
	373	};
	374	my $merr = $@;
1e3baf05	375
a05b47a8	376	my $lstat = 0;
e52bd94c AD	377	my $usleep = 2000000;
e52bd94c AD	378	my $i = 0;
b0b756c1	379	my $err_count = 0;
865ef132 SP	380	my $lastrem = undef;
865ef132 SP	381	my $downtimecounter = 0;
1e3baf05	382	while (1) {
e52bd94c AD	383	$i++;
	384	my $avglstat = $lstat/$i if $lstat;
	385
b0b756c1 DM	386	usleep($usleep);
	387	my $stat;
	388	eval {
	389	$stat = PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "query-migrate");
	390	};
	391	if (my $err = $@) {
	392	$err_count++;
	393	warn "query migrate failed: $err\n";
	394	if ($err_count <= 5) {
	395	usleep(1000000);
	396	next;
	397	}
	398	die "too many query migrate failures - aborting\n";
	399	}
5a7835f5	400	if ($stat->{status} =~ m/^(active\|completed\|failed\|cancelled)$/im) {
d68afb26	401	$merr = undef;
b0b756c1	402	$err_count = 0;
5a7835f5	403	if ($stat->{status} eq 'completed') {
1e3baf05 DM	404	my $delay = time() - $start;
	405	if ($delay > 0) {
	406	my $mbps = sprintf "%.2f", $conf->{memory}/$delay;
135007c0 AD	407	my $downtime = $stat->{downtime} \|\| 0;
135007c0 AD	408	$self->log('info', "migration speed: $mbps MB/s - downtime $downtime ms");
1e3baf05 DM	409	}
1e3baf05 DM	410	}
f5eb281a	411
5a7835f5	412	if ($stat->{status} eq 'failed' \|\| $stat->{status} eq 'cancelled') {
1e3baf05 DM	413	die "aborting\n"
	414	}
	415
a05b47a8 DM	416	if ($stat->{status} ne 'active') {
	417	$self->log('info', "migration status: $stat->{status}");
	418	last;
	419	}
	420
	421	if ($stat->{ram}->{transferred} ne $lstat) {
	422	my $trans = $stat->{ram}->{transferred} \|\| 0;
	423	my $rem = $stat->{ram}->{remaining} \|\| 0;
	424	my $total = $stat->{ram}->{total} \|\| 0;
e18b0b99 AD	425	my $xbzrlecachesize = $stat->{"xbzrle-cache"}->{"cache-size"} \|\| 0;
	426	my $xbzrlebytes = $stat->{"xbzrle-cache"}->{"bytes"} \|\| 0;
	427	my $xbzrlepages = $stat->{"xbzrle-cache"}->{"pages"} \|\| 0;
	428	my $xbzrlecachemiss = $stat->{"xbzrle-cache"}->{"cache-miss"} \|\| 0;
	429	my $xbzrleoverflow = $stat->{"xbzrle-cache"}->{"overflow"} \|\| 0;
135007c0	430	my $expected_downtime = $stat->{"expected-downtime"} \|\| 0;
e52bd94c	431	#reduce sleep if remainig memory if lower than the everage transfert
94235c59	432	$usleep = 300000 if $avglstat && $rem < $avglstat;
a05b47a8 DM	433
a05b47a8 DM	434	$self->log('info', "migration status: $stat->{status} (transferred ${trans}, " .
135007c0	435	"remaining ${rem}), total ${total}) , expected downtime ${expected_downtime}");
e18b0b99	436
972511a0	437	#$self->log('info', "migration xbzrle cachesize: ${xbzrlecachesize} transferred ${xbzrlebytes} pages ${xbzrlepages} cachemiss ${xbzrlecachemiss} overflow ${xbzrleoverflow}");
865ef132 SP	438	if (($lastrem && $rem > $lastrem ) \|\| ($rem == 0)) {
	439	$downtimecounter++;
	440	}
	441	$lastrem = $rem;
	442
	443	if ($downtimecounter > 5) {
	444	$downtimecounter = 0;
	445	$migrate_downtime *= 2;
	446	$self->log('info', "migrate_set_downtime: $migrate_downtime");
	447	eval {
	448	PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate_set_downtime", value => int($migrate_downtime*100)/100);
	449	};
	450	$self->log('info', "migrate_set_downtime error: $@") if $@;
	451	}
	452
a05b47a8 DM	453	}
a05b47a8 DM	454
865ef132	455
a05b47a8	456	$lstat = $stat->{ram}->{transferred};
e52bd94c	457
1e3baf05	458	} else {
d68afb26	459	die $merr if $merr;
5a7835f5	460	die "unable to parse migration status '$stat->{status}' - aborting\n";
1e3baf05	461	}
a05b47a8	462	}
1e3baf05	463	}
16e903f2	464
c04b5b04 AD	465	sub phase2_cleanup {
	466	my ($self, $vmid, $err) = @_;
	467
af30308f DM	468	return if !$self->{errors};
	469	$self->{phase2errors} = 1;
	470
c04b5b04 AD	471	$self->log('info', "aborting phase 2 - cleanup resources");
	472
	473	my $conf = $self->{vmconf};
	474	delete $conf->{lock};
	475	eval { PVE::QemuServer::update_config_nolock($vmid, $conf, 1) };
	476	if (my $err = $@) {
	477	$self->log('err', $err);
	478	}
	479
af30308f DM	480	# cleanup ressources on target host
	481	my $nodename = PVE::INotify::nodename();
	482
	483	my $cmd = [@{$self->{rem_ssh}}, 'qm', 'stop', $vmid, '--skiplock', '--migratedfrom', $nodename];
	484	eval{ PVE::Tools::run_command($cmd, outfunc => sub {}, errfunc => sub {}) };
	485	if (my $err = $@) {
	486	$self->log('err', $err);
	487	$self->{errors} = 1;
	488	}
c04b5b04 AD	489	}
c04b5b04 AD	490
16e903f2 DM	491	sub phase3 {
16e903f2 DM	492	my ($self, $vmid) = @_;
f5eb281a	493
16e903f2	494	my $volids = $self->{volumes};
af30308f	495	return if $self->{phase2errors};
16e903f2 DM	496
	497	# destroy local copies
	498	foreach my $volid (@$volids) {
	499	eval { PVE::Storage::vdisk_free($self->{storecfg}, $volid); };
	500	if (my $err = $@) {
	501	$self->log('err', "removing local copy of '$volid' failed - $err");
	502	$self->{errors} = 1;
	503	last if $err =~ /^interrupted by signal$/;
	504	}
	505	}
16e903f2 DM	506	}
	507
	508	sub phase3_cleanup {
	509	my ($self, $vmid, $err) = @_;
	510
	511	my $conf = $self->{vmconf};
af30308f	512	return if $self->{phase2errors};
16e903f2	513
b8d20802 AD	514	# move config to remote node
	515	my $conffile = PVE::QemuServer::config_file($vmid);
	516	my $newconffile = PVE::QemuServer::config_file($vmid, $self->{node});
	517
	518	die "Failed to move config to node '$self->{node}' - rename failed: $!\n"
	519	if !rename($conffile, $newconffile);
	520
f5eb281a	521	# now that config file is move, we can resume vm on target if livemigrate
b67900f1	522	if ($self->{tunnel}) {
b67900f1 AD	523	my $cmd = [@{$self->{rem_ssh}}, 'qm', 'resume', $vmid, '--skiplock'];
b67900f1 AD	524	eval{ PVE::Tools::run_command($cmd, outfunc => sub {}, errfunc => sub {}) };
f5eb281a	525	if (my $err = $@) {
b67900f1 AD	526	$self->log('err', $err);
	527	$self->{errors} = 1;
	528	}
	529	}
	530
16e903f2 DM	531	# always stop local VM
	532	eval { PVE::QemuServer::vm_stop($self->{storecfg}, $vmid, 1, 1); };
	533	if (my $err = $@) {
	534	$self->log('err', "stopping vm failed - $err");
	535	$self->{errors} = 1;
	536	}
	537
97439670 DM	538	if ($self->{tunnel}) {
	539	eval { finish_tunnel($self, $self->{tunnel}); };
	540	if (my $err = $@) {
	541	$self->log('err', $err);
	542	$self->{errors} = 1;
	543	}
	544	}
	545
16e903f2 DM	546	# always deactivate volumes - avoid lvm LVs to be active on several nodes
	547	eval {
	548	my $vollist = PVE::QemuServer::get_vm_volumes($conf);
	549	PVE::Storage::deactivate_volumes($self->{storecfg}, $vollist);
	550	};
	551	if (my $err = $@) {
	552	$self->log('err', $err);
	553	$self->{errors} = 1;
	554	}
	555
	556	# clear migrate lock
	557	my $cmd = [ @{$self->{rem_ssh}}, 'qm', 'unlock', $vmid ];
	558	$self->cmd_logerr($cmd, errmsg => "failed to clear migrate lock");
	559	}
	560
	561	sub final_cleanup {
	562	my ($self, $vmid) = @_;
	563
	564	# nothing to do
	565	}
	566
	567	1;