PVE/QemuMigrate.pm

   1 package PVE::QemuMigrate;
   2
   3 use strict;
   4 use warnings;
   5 use PVE::AbstractMigrate;
   6 use IO::File;
   7 use IPC::Open2;
   8 use PVE::INotify;
   9 use PVE::Tools;
  10 use PVE::Cluster;
  11 use PVE::Storage;
  12 use PVE::QemuServer;
  13 use Time::HiRes qw( usleep );
  14 use PVE::RPCEnvironment;
  15
  16 use base qw(PVE::AbstractMigrate);
  17
  18 sub fork_command_pipe {
  19     my ($self, $cmd) = @_;
  20
  21     my $reader = IO::File->new();
  22     my $writer = IO::File->new();
  23
  24     my $orig_pid = $$;
  25
  26     my $cpid;
  27
  28     eval { $cpid = open2($reader, $writer, @$cmd); };
  29
  30     my $err = $@;
  31
  32     # catch exec errors
  33     if ($orig_pid != $$) {
  34         $self->log('err', "can't fork command pipe\n");
  35         POSIX::_exit(1);
  36         kill('KILL', $$);
  37     }
  38
  39     die $err if $err;
  40
  41     return { writer => $writer, reader => $reader, pid => $cpid };
  42 }
  43
  44 sub finish_command_pipe {
  45     my ($self, $cmdpipe, $timeout) = @_;
  46
  47     my $writer = $cmdpipe->{writer};
  48     my $reader = $cmdpipe->{reader};
  49
  50     $writer->close();
  51     $reader->close();
  52
  53     my $cpid = $cmdpipe->{pid};
  54
  55     if ($timeout) {
  56         for (my $i = 0; $i < $timeout; $i++) {
  57             return if !PVE::ProcFSTools::check_process_running($cpid);
  58             sleep(1);
  59         }
  60     }
  61
  62     $self->log('info', "ssh tunnel still running - terminating now with SIGTERM\n");
  63     kill(15, $cpid);
  64
  65     # wait again
  66     for (my $i = 0; $i < 10; $i++) {
  67         return if !PVE::ProcFSTools::check_process_running($cpid);
  68         sleep(1);
  69     }
  70
  71     $self->log('info', "ssh tunnel still running - terminating now with SIGKILL\n");
  72     kill 9, $cpid;
  73     sleep 1;
  74 }
  75
  76 sub fork_tunnel {
  77     my ($self, $nodeip, $lport, $rport) = @_;
  78
  79     my $cmd = [@{$self->{rem_ssh}}, '-L', "$lport:localhost:$rport",
  80                'qm', 'mtunnel' ];
  81
  82     my $tunnel = $self->fork_command_pipe($cmd);
  83
  84     my $reader = $tunnel->{reader};
  85
  86     my $helo;
  87     eval {
  88         PVE::Tools::run_with_timeout(60, sub { $helo = <$reader>; });
  89         die "no reply\n" if !$helo;
  90         die "no quorum on target node\n" if $helo =~ m/^no quorum$/;
  91         die "got strange reply from mtunnel ('$helo')\n"
  92             if $helo !~ m/^tunnel online$/;
  93     };
  94     my $err = $@;
  95
  96     if ($err) {
  97         $self->finish_command_pipe($tunnel);
  98         die "can't open migration tunnel - $err";
  99     }
 100     return $tunnel;
 101 }
 102
 103 sub finish_tunnel {
 104     my ($self, $tunnel) = @_;
 105
 106     my $writer = $tunnel->{writer};
 107
 108     eval {
 109         PVE::Tools::run_with_timeout(30, sub {
 110             print $writer "quit\n";
 111             $writer->flush();
 112         });
 113     };
 114     my $err = $@;
 115
 116     $self->finish_command_pipe($tunnel, 30);
 117
 118     die $err if $err;
 119 }
 120
 121 sub lock_vm {
 122     my ($self, $vmid, $code, @param) = @_;
 123
 124     return PVE::QemuServer::lock_config($vmid, $code, @param);
 125 }
 126
 127 sub prepare {
 128     my ($self, $vmid) = @_;
 129
 130     my $online = $self->{opts}->{online};
 131
 132     $self->{storecfg} = PVE::Storage::config();
 133
 134     # test is VM exist
 135     my $conf = $self->{vmconf} = PVE::QemuServer::load_config($vmid);
 136
 137     PVE::QemuServer::check_lock($conf);
 138
 139     my $running = 0;
 140     if (my $pid = PVE::QemuServer::check_running($vmid)) {
 141         die "cant migrate running VM without --online\n" if !$online;
 142         $running = $pid;
 143         $self->{forcemachine} = PVE::QemuServer::get_current_qemu_machine($vmid);
 144     }
 145
 146     if (my $loc_res = PVE::QemuServer::check_local_resources($conf, 1)) {
 147         if ($self->{running} || !$self->{opts}->{force}) {
 148             die "can't migrate VM which uses local devices\n";
 149         } else {
 150             $self->log('info', "migrating VM which uses local devices");
 151         }
 152     }
 153
 154     # activate volumes
 155     my $vollist = PVE::QemuServer::get_vm_volumes($conf);
 156     PVE::Storage::activate_volumes($self->{storecfg}, $vollist);
 157
 158     # fixme: check if storage is available on both nodes
 159
 160     # test ssh connection
 161     my $cmd = [ @{$self->{rem_ssh}}, '/bin/true' ];
 162     eval { $self->cmd_quiet($cmd); };
 163     die "Can't connect to destination address using public key\n" if $@;
 164
 165     return $running;
 166 }
 167
 168 sub sync_disks {
 169     my ($self, $vmid) = @_;
 170
 171     $self->log('info', "copying disk images");
 172
 173     my $conf = $self->{vmconf};
 174
 175     $self->{volumes} = [];
 176
 177     my $res = [];
 178
 179     eval {
 180
 181         my $volhash = {};
 182         my $cdromhash = {};
 183
 184         my $sharedvm = 1;
 185
 186         my @sids = PVE::Storage::storage_ids($self->{storecfg});
 187         foreach my $storeid (@sids) {
 188             my $scfg = PVE::Storage::storage_config($self->{storecfg}, $storeid);
 189             next if $scfg->{shared};
 190             next if !PVE::Storage::storage_check_enabled($self->{storecfg}, $storeid, undef, 1);
 191
 192             # get list from PVE::Storage (for unused volumes)
 193             my $dl = PVE::Storage::vdisk_list($self->{storecfg}, $storeid, $vmid);
 194             PVE::Storage::foreach_volid($dl, sub {
 195                 my ($volid, $sid, $volname) = @_;
 196
 197                 # check if storage is available on target node
 198                 PVE::Storage::storage_check_node($self->{storecfg}, $sid, $self->{node});
 199
 200                 $volhash->{$volid} = 1;
 201                 $sharedvm = 0; # there is a non-shared disk
 202             });
 203         }
 204
 205         # and add used, owned/non-shared disks (just to be sure we have all)
 206
 207         PVE::QemuServer::foreach_volid($conf, sub {
 208             my ($volid, $is_cdrom) = @_;
 209
 210             return if !$volid;
 211
 212             die "cant migrate local file/device '$volid'\n" if $volid =~ m|^/|;
 213
 214             if ($is_cdrom) {
 215                 die "cant migrate local cdrom drive\n" if $volid eq 'cdrom';
 216                 return if $volid eq 'none';
 217                 $cdromhash->{$volid} = 1;
 218             }
 219
 220             my ($sid, $volname) = PVE::Storage::parse_volume_id($volid);
 221
 222             # check if storage is available on both nodes
 223             my $scfg = PVE::Storage::storage_check_node($self->{storecfg}, $sid);
 224             PVE::Storage::storage_check_node($self->{storecfg}, $sid, $self->{node});
 225
 226             return if $scfg->{shared};
 227
 228             die "can't migrate local cdrom '$volid'\n" if $cdromhash->{$volid};
 229
 230             $sharedvm = 0;
 231
 232             my ($path, $owner) = PVE::Storage::path($self->{storecfg}, $volid);
 233
 234             die "can't migrate volume '$volid' - owned by other VM (owner = VM $owner)\n"
 235                 if !$owner || ($owner != $self->{vmid});
 236
 237             $volhash->{$volid} = 1;
 238         });
 239
 240         if ($self->{running} && !$sharedvm) {
 241             die "can't do online migration - VM uses local disks\n";
 242         }
 243
 244         # do some checks first
 245         foreach my $volid (keys %$volhash) {
 246             my ($sid, $volname) = PVE::Storage::parse_volume_id($volid);
 247             my $scfg =  PVE::Storage::storage_config($self->{storecfg}, $sid);
 248
 249             die "can't migrate '$volid' - storagy type '$scfg->{type}' not supported\n"
 250                 if $scfg->{type} ne 'dir';
 251
 252             # if file, check if a backing file exist
 253             if (($scfg->{type} eq 'dir') && (!$sharedvm)) {
 254                 my (undef, undef, undef, $parent) = PVE::Storage::volume_size_info($self->{storecfg}, $volid, 1);
 255                 die "can't migrate '$volid' as it's a clone of '$parent'" if $parent;
 256             }
 257         }
 258
 259         foreach my $volid (keys %$volhash) {
 260             my ($sid, $volname) = PVE::Storage::parse_volume_id($volid);
 261             push @{$self->{volumes}}, $volid;
 262             PVE::Storage::storage_migrate($self->{storecfg}, $volid, $self->{nodeip}, $sid);
 263         }
 264     };
 265     die "Failed to sync data - $@" if $@;
 266 }
 267
 268 sub phase1 {
 269     my ($self, $vmid) = @_;
 270
 271     $self->log('info', "starting migration of VM $vmid to node '$self->{node}' ($self->{nodeip})");
 272
 273     my $conf = $self->{vmconf};
 274
 275     # set migrate lock in config file
 276     $conf->{lock} = 'migrate';
 277     PVE::QemuServer::update_config_nolock($vmid, $conf, 1);
 278
 279     sync_disks($self, $vmid);
 280
 281 };
 282
 283 sub phase1_cleanup {
 284     my ($self, $vmid, $err) = @_;
 285
 286     $self->log('info', "aborting phase 1 - cleanup resources");
 287
 288     my $conf = $self->{vmconf};
 289     delete $conf->{lock};
 290     eval { PVE::QemuServer::update_config_nolock($vmid, $conf, 1) };
 291     if (my $err = $@) {
 292         $self->log('err', $err);
 293     }
 294
 295     if ($self->{volumes}) {
 296         foreach my $volid (@{$self->{volumes}}) {
 297             $self->log('err', "found stale volume copy '$volid' on node '$self->{node}'");
 298             # fixme: try to remove ?
 299         }
 300     }
 301 }
 302
 303 sub phase2 {
 304     my ($self, $vmid) = @_;
 305
 306     my $conf = $self->{vmconf};
 307
 308     $self->log('info', "starting VM $vmid on remote node '$self->{node}'");
 309
 310     my $rport;
 311
 312     my $nodename = PVE::INotify::nodename();
 313
 314     ## start on remote node
 315     my $cmd = [@{$self->{rem_ssh}}];
 316
 317     my $spice_ticket;
 318     if (PVE::QemuServer::vga_conf_has_spice($conf->{vga})) {
 319         my $res = PVE::QemuServer::vm_mon_cmd($vmid, 'query-spice');
 320         $spice_ticket = $res->{ticket};
 321     }
 322
 323     push @$cmd , 'qm', 'start', $vmid, '--stateuri', 'tcp', '--skiplock', '--migratedfrom', $nodename;
 324
 325     if ($self->{forcemachine}) {
 326         push @$cmd, '--machine', $self->{forcemachine};
 327     }
 328
 329     my $spice_port;
 330
 331     # Note: We try to keep $spice_ticket secret (do not pass via command line parameter)
 332     # instead we pipe it through STDIN
 333     PVE::Tools::run_command($cmd, input => $spice_ticket, outfunc => sub {
 334         my $line = shift;
 335
 336         if ($line =~ m/^migration listens on port (\d+)$/) {
 337             $rport = int($1);
 338         }elsif ($line =~ m/^spice listens on port (\d+)$/) {
 339             $spice_port = int($1);
 340         }
 341     }, errfunc => sub {
 342         my $line = shift;
 343         $self->log('info', $line);
 344     });
 345
 346     die "unable to detect remote migration port\n" if !$rport;
 347
 348     $self->log('info', "starting migration tunnel");
 349
 350     ## create tunnel to remote port
 351     my $lport = PVE::Tools::next_migrate_port();
 352     $self->{tunnel} = $self->fork_tunnel($self->{nodeip}, $lport, $rport);
 353
 354     $self->log('info', "starting online/live migration on port $lport");
 355     # start migration
 356
 357     my $start = time();
 358
 359     # load_defaults
 360     my $defaults = PVE::QemuServer::load_defaults();
 361
 362     # always set migrate speed (overwrite kvm default of 32m)
 363     # we set a very hight default of 8192m which is basically unlimited
 364     my $migrate_speed = $defaults->{migrate_speed} || 8192;
 365     $migrate_speed = $conf->{migrate_speed} || $migrate_speed;
 366     $migrate_speed = $migrate_speed * 1048576;
 367     $self->log('info', "migrate_set_speed: $migrate_speed");
 368     eval {
 369         PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate_set_speed", value => int($migrate_speed));
 370     };
 371     $self->log('info', "migrate_set_speed error: $@") if $@;
 372
 373     my $migrate_downtime = $defaults->{migrate_downtime};
 374     $migrate_downtime = $conf->{migrate_downtime} if defined($conf->{migrate_downtime});
 375     if (defined($migrate_downtime)) {
 376         $self->log('info', "migrate_set_downtime: $migrate_downtime");
 377         eval {
 378             PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate_set_downtime", value => int($migrate_downtime*100)/100);
 379         };
 380         $self->log('info', "migrate_set_downtime error: $@") if $@;
 381     }
 382
 383     my $capabilities = {};
 384     $capabilities->{capability} =  "xbzrle";
 385     $capabilities->{state} = JSON::false;
 386
 387     eval {
 388         PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate-set-capabilities", capabilities => [$capabilities]);
 389     };
 390
 391     #set cachesize 10% of the total memory
 392     my $cachesize = int($conf->{memory}*1048576/10);
 393     eval {
 394         PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate-set-cache-size", value => $cachesize);
 395     };
 396
 397     if (PVE::QemuServer::vga_conf_has_spice($conf->{vga})) {
 398         my $rpcenv = PVE::RPCEnvironment::get();
 399         my $authuser = $rpcenv->get_user();
 400
 401         my (undef, $proxyticket) = PVE::AccessControl::assemble_spice_ticket($authuser, $vmid, $self->{node});
 402
 403         my $filename = "/etc/pve/nodes/$self->{node}/pve-ssl.pem";
 404         my $subject = PVE::QemuServer::read_x509_subject_spice($filename);
 405
 406         $self->log('info', "spice client_migrate_info");
 407
 408         eval {
 409             PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "client_migrate_info", protocol => 'spice',
 410                                                 hostname => $proxyticket, 'tls-port' => $spice_port,
 411                                                 'cert-subject' => $subject);
 412         };
 413         $self->log('info', "client_migrate_info error: $@") if $@;
 414
 415     }
 416
 417     eval {
 418         PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate", uri => "tcp:localhost:$lport");
 419     };
 420     my $merr = $@;
 421
 422     my $lstat = 0;
 423     my $usleep = 2000000;
 424     my $i = 0;
 425     my $err_count = 0;
 426     my $lastrem = undef;
 427     my $downtimecounter = 0;
 428     while (1) {
 429         $i++;
 430         my $avglstat = $lstat/$i if $lstat;
 431
 432         usleep($usleep);
 433         my $stat;
 434         eval {
 435             $stat = PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "query-migrate");
 436         };
 437         if (my $err = $@) {
 438             $err_count++;
 439             warn "query migrate failed: $err\n";
 440             if ($err_count <= 5) {
 441                 usleep(1000000);
 442                 next;
 443             }
 444             die "too many query migrate failures - aborting\n";
 445         }
 446         if ($stat->{status} =~ m/^(active|completed|failed|cancelled)$/im) {
 447             $merr = undef;
 448             $err_count = 0;
 449             if ($stat->{status} eq 'completed') {
 450                 my $delay = time() - $start;
 451                 if ($delay > 0) {
 452                     my $mbps = sprintf "%.2f", $conf->{memory}/$delay;
 453                     my $downtime = $stat->{downtime} || 0;
 454                     $self->log('info', "migration speed: $mbps MB/s - downtime $downtime ms");
 455                 }
 456             }
 457
 458             if ($stat->{status} eq 'failed' || $stat->{status} eq 'cancelled') {
 459                 die "aborting\n"
 460             }
 461
 462             if ($stat->{status} ne 'active') {
 463                 $self->log('info', "migration status: $stat->{status}");
 464                 last;
 465             }
 466
 467             if ($stat->{ram}->{transferred} ne $lstat) {
 468                 my $trans = $stat->{ram}->{transferred} || 0;
 469                 my $rem = $stat->{ram}->{remaining} || 0;
 470                 my $total = $stat->{ram}->{total} || 0;
 471                 my $xbzrlecachesize = $stat->{"xbzrle-cache"}->{"cache-size"} || 0;
 472                 my $xbzrlebytes = $stat->{"xbzrle-cache"}->{"bytes"} || 0;
 473                 my $xbzrlepages = $stat->{"xbzrle-cache"}->{"pages"} || 0;
 474                 my $xbzrlecachemiss = $stat->{"xbzrle-cache"}->{"cache-miss"} || 0;
 475                 my $xbzrleoverflow = $stat->{"xbzrle-cache"}->{"overflow"} || 0;
 476                 #reduce sleep if remainig memory if lower than the everage transfert
 477                 $usleep = 300000 if $avglstat && $rem < $avglstat;
 478
 479                 $self->log('info', "migration status: $stat->{status} (transferred ${trans}, " .
 480                            "remaining ${rem}), total ${total})");
 481
 482                 #$self->log('info', "migration xbzrle cachesize: ${xbzrlecachesize} transferred ${xbzrlebytes} pages ${xbzrlepages} cachemiss ${xbzrlecachemiss} overflow ${xbzrleoverflow}");
 483                 if (($lastrem  && $rem > $lastrem ) || ($rem == 0)) {
 484                     $downtimecounter++;
 485                 }
 486                 $lastrem = $rem;
 487
 488                 if ($downtimecounter > 5) {
 489                     $downtimecounter = 0;
 490                     $migrate_downtime *= 2;
 491                     $self->log('info', "migrate_set_downtime: $migrate_downtime");
 492                     eval {
 493                         PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate_set_downtime", value => int($migrate_downtime*100)/100);
 494                     };
 495                     $self->log('info', "migrate_set_downtime error: $@") if $@;
 496                 }
 497
 498             }
 499
 500
 501             $lstat = $stat->{ram}->{transferred};
 502
 503         } else {
 504             die $merr if $merr;
 505             die "unable to parse migration status '$stat->{status}' - aborting\n";
 506         }
 507     }
 508 }
 509
 510 sub phase2_cleanup {
 511     my ($self, $vmid, $err) = @_;
 512
 513     return if !$self->{errors};
 514     $self->{phase2errors} = 1;
 515
 516     $self->log('info', "aborting phase 2 - cleanup resources");
 517
 518     $self->log('info', "migrate_cancel");
 519     eval {
 520         PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate_cancel");
 521     };
 522     $self->log('info', "migrate_cancel error: $@") if $@;
 523
 524     my $conf = $self->{vmconf};
 525     delete $conf->{lock};
 526     eval { PVE::QemuServer::update_config_nolock($vmid, $conf, 1) };
 527     if (my $err = $@) {
 528         $self->log('err', $err);
 529     }
 530
 531     # cleanup ressources on target host
 532     my $nodename = PVE::INotify::nodename();
 533
 534     my $cmd = [@{$self->{rem_ssh}}, 'qm', 'stop', $vmid, '--skiplock', '--migratedfrom', $nodename];
 535     eval{ PVE::Tools::run_command($cmd, outfunc => sub {}, errfunc => sub {}) };
 536     if (my $err = $@) {
 537         $self->log('err', $err);
 538         $self->{errors} = 1;
 539     }
 540 }
 541
 542 sub phase3 {
 543     my ($self, $vmid) = @_;
 544
 545     my $volids = $self->{volumes};
 546     return if $self->{phase2errors};
 547
 548     # destroy local copies
 549     foreach my $volid (@$volids) {
 550         eval { PVE::Storage::vdisk_free($self->{storecfg}, $volid); };
 551         if (my $err = $@) {
 552             $self->log('err', "removing local copy of '$volid' failed - $err");
 553             $self->{errors} = 1;
 554             last if $err =~ /^interrupted by signal$/;
 555         }
 556     }
 557 }
 558
 559 sub phase3_cleanup {
 560     my ($self, $vmid, $err) = @_;
 561
 562     my $conf = $self->{vmconf};
 563     return if $self->{phase2errors};
 564
 565     # move config to remote node
 566     my $conffile = PVE::QemuServer::config_file($vmid);
 567     my $newconffile = PVE::QemuServer::config_file($vmid, $self->{node});
 568
 569     die "Failed to move config to node '$self->{node}' - rename failed: $!\n"
 570         if !rename($conffile, $newconffile);
 571
 572     # now that config file is move, we can resume vm on target if livemigrate
 573     if ($self->{tunnel}) {
 574         my $cmd = [@{$self->{rem_ssh}}, 'qm', 'resume', $vmid, '--skiplock'];
 575         eval{ PVE::Tools::run_command($cmd, outfunc => sub {},
 576                 errfunc => sub {
 577                     my $line = shift;
 578                     $self->log('err', $line);
 579                 });
 580         };
 581         if (my $err = $@) {
 582             $self->log('err', $err);
 583             $self->{errors} = 1;
 584         }
 585     }
 586
 587     my $timer = 0;
 588     if (PVE::QemuServer::vga_conf_has_spice($conf->{vga})) {
 589         $self->log('info', "Waiting for spice server migration");
 590         while (1) {
 591             my $res = PVE::QemuServer::vm_mon_cmd_nocheck($vmid, 'query-spice');
 592             last if int($res->{'migrated'}) == 1;
 593             last if $timer > 50;
 594             $timer ++;
 595             usleep(200000);
 596         }
 597     }
 598
 599     # always stop local VM
 600     eval { PVE::QemuServer::vm_stop($self->{storecfg}, $vmid, 1, 1); };
 601     if (my $err = $@) {
 602         $self->log('err', "stopping vm failed - $err");
 603         $self->{errors} = 1;
 604     }
 605
 606     if ($self->{tunnel}) {
 607         eval { finish_tunnel($self, $self->{tunnel});  };
 608         if (my $err = $@) {
 609             $self->log('err', $err);
 610             $self->{errors} = 1;
 611         }
 612     }
 613
 614     # always deactivate volumes - avoid lvm LVs to be active on several nodes
 615     eval {
 616         my $vollist = PVE::QemuServer::get_vm_volumes($conf);
 617         PVE::Storage::deactivate_volumes($self->{storecfg}, $vollist);
 618     };
 619     if (my $err = $@) {
 620         $self->log('err', $err);
 621         $self->{errors} = 1;
 622     }
 623
 624     # clear migrate lock
 625     my $cmd = [ @{$self->{rem_ssh}}, 'qm', 'unlock', $vmid ];
 626     $self->cmd_logerr($cmd, errmsg => "failed to clear migrate lock");
 627 }
 628
 629 sub final_cleanup {
 630     my ($self, $vmid) = @_;
 631
 632     # nothing to do
 633 }
 634
 635 1;