]>
git.proxmox.com Git - qemu-server.git/blob - PVE/QemuMigrate.pm
1 package PVE
::QemuMigrate
;
5 use PVE
::AbstractMigrate
;
8 use POSIX
qw( WNOHANG );
14 use Time
::HiRes
qw( usleep );
15 use PVE
::RPCEnvironment
;
17 use base
qw(PVE::AbstractMigrate);
19 sub fork_command_pipe
{
20 my ($self, $cmd) = @_;
22 my $reader = IO
::File-
>new();
23 my $writer = IO
::File-
>new();
29 eval { $cpid = open2
($reader, $writer, @$cmd); };
34 if ($orig_pid != $$) {
35 $self->log('err', "can't fork command pipe\n");
42 return { writer
=> $writer, reader
=> $reader, pid
=> $cpid };
45 sub finish_command_pipe
{
46 my ($self, $cmdpipe, $timeout) = @_;
48 my $cpid = $cmdpipe->{pid
};
49 return if !defined($cpid);
51 my $writer = $cmdpipe->{writer
};
52 my $reader = $cmdpipe->{reader
};
57 my $collect_child_process = sub {
58 my $res = waitpid($cpid, WNOHANG
);
59 if (defined($res) && ($res == $cpid)) {
60 delete $cmdpipe->{cpid
};
68 for (my $i = 0; $i < $timeout; $i++) {
69 return if &$collect_child_process();
74 $self->log('info', "ssh tunnel still running - terminating now with SIGTERM\n");
78 for (my $i = 0; $i < 10; $i++) {
79 return if &$collect_child_process();
83 $self->log('info', "ssh tunnel still running - terminating now with SIGKILL\n");
87 $self->log('err', "ssh tunnel child process (PID $cpid) couldn't be collected\n")
88 if !&$collect_child_process();
92 my ($self, $tunnel_addr) = @_;
94 my @localtunnelinfo = defined($tunnel_addr) ?
('-L' , $tunnel_addr ) : ();
96 my $cmd = [@{$self->{rem_ssh
}}, '-o ExitOnForwardFailure=yes', @localtunnelinfo, 'qm', 'mtunnel' ];
98 my $tunnel = $self->fork_command_pipe($cmd);
100 my $reader = $tunnel->{reader
};
104 PVE
::Tools
::run_with_timeout
(60, sub { $helo = <$reader>; });
105 die "no reply\n" if !$helo;
106 die "no quorum on target node\n" if $helo =~ m/^no quorum$/;
107 die "got strange reply from mtunnel ('$helo')\n"
108 if $helo !~ m/^tunnel online$/;
113 $self->finish_command_pipe($tunnel);
114 die "can't open migration tunnel - $err";
120 my ($self, $tunnel) = @_;
122 my $writer = $tunnel->{writer
};
125 PVE
::Tools
::run_with_timeout
(30, sub {
126 print $writer "quit\n";
132 $self->finish_command_pipe($tunnel, 30);
134 if ($tunnel->{sock_addr
}) {
135 # ssh does not clean up on local host
136 my $cmd = ['rm', '-f', $tunnel->{sock_addr
}]; #
137 PVE
::Tools
::run_command
($cmd);
139 # .. and just to be sure check on remote side
140 unshift @{$cmd}, @{$self->{rem_ssh
}};
141 PVE
::Tools
::run_command
($cmd);
148 my ($self, $vmid, $code, @param) = @_;
150 return PVE
::QemuConfig-
>lock_config($vmid, $code, @param);
154 my ($self, $vmid) = @_;
156 my $online = $self->{opts
}->{online
};
158 $self->{storecfg
} = PVE
::Storage
::config
();
161 my $conf = $self->{vmconf
} = PVE
::QemuConfig-
>load_config($vmid);
163 PVE
::QemuConfig-
>check_lock($conf);
166 if (my $pid = PVE
::QemuServer
::check_running
($vmid)) {
167 die "can't migrate running VM without --online\n" if !$online;
170 $self->{forcemachine
} = PVE
::QemuServer
::qemu_machine_pxe
($vmid, $conf);
174 if (my $loc_res = PVE
::QemuServer
::check_local_resources
($conf, 1)) {
175 if ($self->{running
} || !$self->{opts
}->{force
}) {
176 die "can't migrate VM which uses local devices\n";
178 $self->log('info', "migrating VM which uses local devices");
182 my $vollist = PVE
::QemuServer
::get_vm_volumes
($conf);
184 my $need_activate = [];
185 foreach my $volid (@$vollist) {
186 my ($sid, $volname) = PVE
::Storage
::parse_volume_id
($volid, 1);
188 # check if storage is available on both nodes
189 my $scfg = PVE
::Storage
::storage_check_node
($self->{storecfg
}, $sid);
190 PVE
::Storage
::storage_check_node
($self->{storecfg
}, $sid, $self->{node
});
192 if ($scfg->{shared
}) {
193 # PVE::Storage::activate_storage checks this for non-shared storages
194 my $plugin = PVE
::Storage
::Plugin-
>lookup($scfg->{type
});
195 warn "Used shared storage '$sid' is not online on source node!\n"
196 if !$plugin->check_connection($sid, $scfg);
198 # only activate if not shared
199 push @$need_activate, $volid;
204 PVE
::Storage
::activate_volumes
($self->{storecfg
}, $need_activate);
206 # test ssh connection
207 my $cmd = [ @{$self->{rem_ssh
}}, '/bin/true' ];
208 eval { $self->cmd_quiet($cmd); };
209 die "Can't connect to destination address using public key\n" if $@;
215 my ($self, $vmid) = @_;
217 $self->log('info', "copying disk images");
219 my $conf = $self->{vmconf
};
221 $self->{volumes
} = [];
231 my @sids = PVE
::Storage
::storage_ids
($self->{storecfg
});
232 foreach my $storeid (@sids) {
233 my $scfg = PVE
::Storage
::storage_config
($self->{storecfg
}, $storeid);
234 next if $scfg->{shared
};
235 next if !PVE
::Storage
::storage_check_enabled
($self->{storecfg
}, $storeid, undef, 1);
237 # get list from PVE::Storage (for unused volumes)
238 my $dl = PVE
::Storage
::vdisk_list
($self->{storecfg
}, $storeid, $vmid);
240 next if @{$dl->{$storeid}} == 0;
242 # check if storage is available on target node
243 PVE
::Storage
::storage_check_node
($self->{storecfg
}, $storeid, $self->{node
});
244 $sharedvm = 0; # there is a non-shared disk
246 PVE
::Storage
::foreach_volid
($dl, sub {
247 my ($volid, $sid, $volname) = @_;
249 $volhash->{$volid} = 1;
253 my $test_volid = sub {
254 my ($volid, $is_cdrom, $snapname) = @_;
258 die "can't migrate local file/device '$volid'\n" if $volid =~ m
|^/|;
261 die "can't migrate local cdrom drive\n" if $volid eq 'cdrom';
262 return if $volid eq 'none';
265 my ($sid, $volname) = PVE
::Storage
::parse_volume_id
($volid);
267 # check if storage is available on both nodes
268 my $scfg = PVE
::Storage
::storage_check_node
($self->{storecfg
}, $sid);
269 PVE
::Storage
::storage_check_node
($self->{storecfg
}, $sid, $self->{node
});
271 return if $scfg->{shared
};
275 die "can't migrate local cdrom '$volid'\n" if $is_cdrom;
277 my ($path, $owner) = PVE
::Storage
::path
($self->{storecfg
}, $volid);
279 die "can't migrate volume '$volid' - owned by other VM (owner = VM $owner)\n"
280 if !$owner || ($owner != $self->{vmid
});
282 if (defined($snapname)) {
283 # we cannot migrate shapshots on local storage
284 # exceptions: 'zfspool' or 'qcow2' files (on directory storage)
286 my $format = PVE
::QemuServer
::qemu_img_format
($scfg, $volname);
288 if (($scfg->{type
} eq 'zfspool') || ($format eq 'qcow2')) {
289 $volhash->{$volid} = 1;
293 die "can't migrate snapshot of local volume '$volid'\n";
296 $volhash->{$volid} = 1;
300 my $test_drive = sub {
301 my ($ds, $drive, $snapname) = @_;
303 &$test_volid($drive->{file
}, PVE
::QemuServer
::drive_is_cdrom
($drive), $snapname);
306 PVE
::QemuServer
::foreach_drive
($conf, $test_drive);
307 foreach my $snapname (keys %{$conf->{snapshots
}}) {
308 &$test_volid($conf->{snapshots
}->{$snapname}->{'vmstate'}, 0, undef)
309 if defined($conf->{snapshots
}->{$snapname}->{'vmstate'});
310 PVE
::QemuServer
::foreach_drive
($conf->{snapshots
}->{$snapname}, $test_drive, $snapname);
313 if ($self->{running
} && !$sharedvm) {
314 die "can't do online migration - VM uses local disks\n";
317 # additional checks for local storage
318 foreach my $volid (keys %$volhash) {
319 my ($sid, $volname) = PVE
::Storage
::parse_volume_id
($volid);
320 my $scfg = PVE
::Storage
::storage_config
($self->{storecfg
}, $sid);
322 my $migratable = ($scfg->{type
} eq 'dir') || ($scfg->{type
} eq 'zfspool') ||
323 ($scfg->{type
} eq 'lvmthin') || ($scfg->{type
} eq 'lvm');
325 die "can't migrate '$volid' - storage type '$scfg->{type}' not supported\n"
328 # image is a linked clone on local storage, se we can't migrate.
329 if (my $basename = (PVE
::Storage
::parse_volname
($self->{storecfg
}, $volid))[3]) {
330 die "can't migrate '$volid' as it's a clone of '$basename'";
334 foreach my $volid (keys %$volhash) {
335 my ($sid, $volname) = PVE
::Storage
::parse_volume_id
($volid);
336 push @{$self->{volumes
}}, $volid;
337 PVE
::Storage
::storage_migrate
($self->{storecfg
}, $volid, $self->{nodeip
}, $sid);
340 die "Failed to sync data - $@" if $@;
344 my ($self, $vmid) = @_;
346 $self->log('info', "starting migration of VM $vmid to node '$self->{node}' ($self->{nodeip})");
348 my $conf = $self->{vmconf
};
350 # set migrate lock in config file
351 $conf->{lock} = 'migrate';
352 PVE
::QemuConfig-
>write_config($vmid, $conf);
354 sync_disks
($self, $vmid);
359 my ($self, $vmid, $err) = @_;
361 $self->log('info', "aborting phase 1 - cleanup resources");
363 my $conf = $self->{vmconf
};
364 delete $conf->{lock};
365 eval { PVE
::QemuConfig-
>write_config($vmid, $conf) };
367 $self->log('err', $err);
370 if ($self->{volumes
}) {
371 foreach my $volid (@{$self->{volumes
}}) {
372 $self->log('err', "found stale volume copy '$volid' on node '$self->{node}'");
373 # fixme: try to remove ?
379 my ($self, $vmid) = @_;
381 my $conf = $self->{vmconf
};
383 $self->log('info', "starting VM $vmid on remote node '$self->{node}'");
387 my $ruri; # the whole migration dst. URI (protocol:address[:port])
388 my $nodename = PVE
::INotify
::nodename
();
390 ## start on remote node
391 my $cmd = [@{$self->{rem_ssh
}}];
394 if (PVE
::QemuServer
::vga_conf_has_spice
($conf->{vga
})) {
395 my $res = PVE
::QemuServer
::vm_mon_cmd
($vmid, 'query-spice');
396 $spice_ticket = $res->{ticket
};
399 push @$cmd , 'qm', 'start', $vmid, '--skiplock', '--migratedfrom', $nodename;
401 # we use TCP only for unsecure migrations as TCP ssh forward tunnels often
402 # did appeared to late (they are hard, if not impossible, to check for)
403 # secure migration use UNIX sockets now, this *breaks* compatibilty when trying
404 # to migrate from new to old but *not* from old to new.
405 my $datacenterconf = PVE
::Cluster
::cfs_read_file
('datacenter.cfg');
406 my $secure_migration = ($datacenterconf->{migration_unsecure
}) ?
0 : 1;
408 if (!$secure_migration) {
409 push @$cmd, '--stateuri', 'tcp';
411 push @$cmd, '--stateuri', 'unix';
414 if ($self->{forcemachine
}) {
415 push @$cmd, '--machine', $self->{forcemachine
};
420 # Note: We try to keep $spice_ticket secret (do not pass via command line parameter)
421 # instead we pipe it through STDIN
422 PVE
::Tools
::run_command
($cmd, input
=> $spice_ticket, outfunc
=> sub {
425 if ($line =~ m/^migration listens on tcp:(localhost|[\d\.]+|\[[\d\.:a-fA-F]+\]):(\d+)$/) {
428 $ruri = "tcp:$raddr:$rport";
430 elsif ($line =~ m!^migration listens on unix:(/run/qemu-server/(\d+)\.migrate)$!) {
432 die "Destination UNIX sockets VMID does not match source VMID" if $vmid ne $2;
433 $ruri = "unix:$raddr";
435 elsif ($line =~ m/^migration listens on port (\d+)$/) {
436 $raddr = "localhost";
438 $ruri = "tcp:$raddr:$rport";
440 elsif ($line =~ m/^spice listens on port (\d+)$/) {
441 $spice_port = int($1);
445 $self->log('info', $line);
448 die "unable to detect remote migration address\n" if !$raddr;
450 if ($secure_migration) {
451 $self->log('info', "start remote tunnel");
453 if ($ruri =~ /^unix:/) {
455 $self->{tunnel
} = $self->fork_tunnel("$raddr:$raddr");
456 $self->{tunnel
}->{sock_addr
} = $raddr;
458 my $unix_socket_try = 0; # wait for the socket to become ready
459 while (! -S
$raddr) {
461 if ($unix_socket_try > 100) {
463 $self->finish_tunnel($self->{tunnel
});
464 die "Timeout, migration socket $ruri did not get ready";
470 } elsif ($ruri =~ /^tcp:/) {
472 if ($raddr eq "localhost") {
473 # for backwards compatibility with older qemu-server versions
474 my $pfamily = PVE
::Tools
::get_host_address_family
($nodename);
475 my $lport = PVE
::Tools
::next_migrate_port
($pfamily);
476 $tunnel_addr = "$lport:localhost:$rport";
479 $self->{tunnel
} = $self->fork_tunnel($tunnel_addr);
482 die "unsupported protocol in migration URI: $ruri\n";
487 $self->log('info', "starting online/live migration on $ruri");
488 $self->{livemigration
} = 1;
491 my $defaults = PVE
::QemuServer
::load_defaults
();
493 # always set migrate speed (overwrite kvm default of 32m)
494 # we set a very hight default of 8192m which is basically unlimited
495 my $migrate_speed = $defaults->{migrate_speed
} || 8192;
496 $migrate_speed = $conf->{migrate_speed
} || $migrate_speed;
497 $migrate_speed = $migrate_speed * 1048576;
498 $self->log('info', "migrate_set_speed: $migrate_speed");
500 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "migrate_set_speed", value
=> int($migrate_speed));
502 $self->log('info', "migrate_set_speed error: $@") if $@;
504 my $migrate_downtime = $defaults->{migrate_downtime
};
505 $migrate_downtime = $conf->{migrate_downtime
} if defined($conf->{migrate_downtime
});
506 if (defined($migrate_downtime)) {
507 $self->log('info', "migrate_set_downtime: $migrate_downtime");
509 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "migrate_set_downtime", value
=> int($migrate_downtime*100)/100);
511 $self->log('info', "migrate_set_downtime error: $@") if $@;
514 $self->log('info', "set migration_caps");
516 PVE
::QemuServer
::set_migration_caps
($vmid);
520 #set cachesize 10% of the total memory
521 my $cachesize = int($conf->{memory
}*1048576/10);
522 $self->log('info', "set cachesize: $cachesize");
524 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "migrate-set-cache-size", value
=> int($cachesize));
526 $self->log('info', "migrate-set-cache-size error: $@") if $@;
528 if (PVE
::QemuServer
::vga_conf_has_spice
($conf->{vga
})) {
529 my $rpcenv = PVE
::RPCEnvironment
::get
();
530 my $authuser = $rpcenv->get_user();
532 my (undef, $proxyticket) = PVE
::AccessControl
::assemble_spice_ticket
($authuser, $vmid, $self->{node
});
534 my $filename = "/etc/pve/nodes/$self->{node}/pve-ssl.pem";
535 my $subject = PVE
::AccessControl
::read_x509_subject_spice
($filename);
537 $self->log('info', "spice client_migrate_info");
540 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "client_migrate_info", protocol
=> 'spice',
541 hostname
=> $proxyticket, 'tls-port' => $spice_port,
542 'cert-subject' => $subject);
544 $self->log('info', "client_migrate_info error: $@") if $@;
548 $self->log('info', "start migrate command to $ruri");
550 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "migrate", uri
=> $ruri);
553 $self->log('info', "migrate uri => $ruri failed: $merr") if $merr;
556 my $usleep = 2000000;
560 my $downtimecounter = 0;
563 my $avglstat = $lstat/$i if $lstat;
568 $stat = PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "query-migrate");
572 warn "query migrate failed: $err\n";
573 $self->log('info', "query migrate failed: $err");
574 if ($err_count <= 5) {
578 die "too many query migrate failures - aborting\n";
581 if (defined($stat->{status
}) && $stat->{status
} =~ m/^(setup)$/im) {
586 if (defined($stat->{status
}) && $stat->{status
} =~ m/^(active|completed|failed|cancelled)$/im) {
589 if ($stat->{status
} eq 'completed') {
590 my $delay = time() - $start;
592 my $mbps = sprintf "%.2f", $conf->{memory
}/$delay;
593 my $downtime = $stat->{downtime
} || 0;
594 $self->log('info', "migration speed: $mbps MB/s - downtime $downtime ms");
598 if ($stat->{status
} eq 'failed' || $stat->{status
} eq 'cancelled') {
599 $self->log('info', "migration status error: $stat->{status}");
603 if ($stat->{status
} ne 'active') {
604 $self->log('info', "migration status: $stat->{status}");
608 if ($stat->{ram
}->{transferred
} ne $lstat) {
609 my $trans = $stat->{ram
}->{transferred
} || 0;
610 my $rem = $stat->{ram
}->{remaining
} || 0;
611 my $total = $stat->{ram
}->{total
} || 0;
612 my $xbzrlecachesize = $stat->{"xbzrle-cache"}->{"cache-size"} || 0;
613 my $xbzrlebytes = $stat->{"xbzrle-cache"}->{"bytes"} || 0;
614 my $xbzrlepages = $stat->{"xbzrle-cache"}->{"pages"} || 0;
615 my $xbzrlecachemiss = $stat->{"xbzrle-cache"}->{"cache-miss"} || 0;
616 my $xbzrleoverflow = $stat->{"xbzrle-cache"}->{"overflow"} || 0;
617 #reduce sleep if remainig memory if lower than the everage transfert
618 $usleep = 300000 if $avglstat && $rem < $avglstat;
620 $self->log('info', "migration status: $stat->{status} (transferred ${trans}, " .
621 "remaining ${rem}), total ${total})");
623 if (${xbzrlecachesize
}) {
624 $self->log('info', "migration xbzrle cachesize: ${xbzrlecachesize} transferred ${xbzrlebytes} pages ${xbzrlepages} cachemiss ${xbzrlecachemiss} overflow ${xbzrleoverflow}");
627 if (($lastrem && $rem > $lastrem ) || ($rem == 0)) {
632 if ($downtimecounter > 5) {
633 $downtimecounter = 0;
634 $migrate_downtime *= 2;
635 $self->log('info', "migrate_set_downtime: $migrate_downtime");
637 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "migrate_set_downtime", value
=> int($migrate_downtime*100)/100);
639 $self->log('info', "migrate_set_downtime error: $@") if $@;
645 $lstat = $stat->{ram
}->{transferred
};
649 die "unable to parse migration status '$stat->{status}' - aborting\n";
653 # just to be sure that the tunnel gets closed on successful migration, on error
654 # phase2_cleanup closes it *after* stopping the remote waiting VM
655 if (!$self->{errors
} && $self->{tunnel
}) {
656 eval { finish_tunnel
($self, $self->{tunnel
}); };
658 $self->log('err', $err);
665 my ($self, $vmid, $err) = @_;
667 return if !$self->{errors
};
668 $self->{phase2errors
} = 1;
670 $self->log('info', "aborting phase 2 - cleanup resources");
672 $self->log('info', "migrate_cancel");
674 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "migrate_cancel");
676 $self->log('info', "migrate_cancel error: $@") if $@;
678 my $conf = $self->{vmconf
};
679 delete $conf->{lock};
680 eval { PVE
::QemuConfig-
>write_config($vmid, $conf) };
682 $self->log('err', $err);
685 # cleanup ressources on target host
686 my $nodename = PVE
::INotify
::nodename
();
688 my $cmd = [@{$self->{rem_ssh
}}, 'qm', 'stop', $vmid, '--skiplock', '--migratedfrom', $nodename];
689 eval{ PVE
::Tools
::run_command
($cmd, outfunc
=> sub {}, errfunc
=> sub {}) };
691 $self->log('err', $err);
695 if ($self->{tunnel
}) {
696 eval { finish_tunnel
($self, $self->{tunnel
}); };
698 $self->log('err', $err);
705 my ($self, $vmid) = @_;
707 my $volids = $self->{volumes
};
708 return if $self->{phase2errors
};
710 # destroy local copies
711 foreach my $volid (@$volids) {
712 eval { PVE
::Storage
::vdisk_free
($self->{storecfg
}, $volid); };
714 $self->log('err', "removing local copy of '$volid' failed - $err");
716 last if $err =~ /^interrupted by signal$/;
722 my ($self, $vmid, $err) = @_;
724 my $conf = $self->{vmconf
};
725 return if $self->{phase2errors
};
727 # move config to remote node
728 my $conffile = PVE
::QemuConfig-
>config_file($vmid);
729 my $newconffile = PVE
::QemuConfig-
>config_file($vmid, $self->{node
});
731 die "Failed to move config to node '$self->{node}' - rename failed: $!\n"
732 if !rename($conffile, $newconffile);
734 if ($self->{livemigration
}) {
735 # now that config file is move, we can resume vm on target if livemigrate
736 my $cmd = [@{$self->{rem_ssh
}}, 'qm', 'resume', $vmid, '--skiplock', '--nocheck'];
737 eval{ PVE
::Tools
::run_command
($cmd, outfunc
=> sub {},
740 $self->log('err', $line);
744 $self->log('err', $err);
752 if (PVE
::QemuServer
::vga_conf_has_spice
($conf->{vga
}) && $self->{running
}) {
753 $self->log('info', "Waiting for spice server migration");
755 my $res = PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, 'query-spice');
756 last if int($res->{'migrated'}) == 1;
764 # always stop local VM
765 eval { PVE
::QemuServer
::vm_stop
($self->{storecfg
}, $vmid, 1, 1); };
767 $self->log('err', "stopping vm failed - $err");
771 # always deactivate volumes - avoid lvm LVs to be active on several nodes
773 my $vollist = PVE
::QemuServer
::get_vm_volumes
($conf);
774 PVE
::Storage
::deactivate_volumes
($self->{storecfg
}, $vollist);
777 $self->log('err', $err);
782 my $cmd = [ @{$self->{rem_ssh
}}, 'qm', 'unlock', $vmid ];
783 $self->cmd_logerr($cmd, errmsg
=> "failed to clear migrate lock");
787 my ($self, $vmid) = @_;