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