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 $cdromhash = {};
 180
 181         my @sids = PVE::Storage::storage_ids($self->{storecfg});
 182         foreach my $storeid (@sids) {
 183             my $scfg = PVE::Storage::storage_config($self->{storecfg}, $storeid);
 184             next if $scfg->{shared};
 185             next if !PVE::Storage::storage_check_enabled($self->{storecfg}, $storeid, undef, 1);
 186
 187             # get list from PVE::Storage (for unused volumes)
 188             my $dl = PVE::Storage::vdisk_list($self->{storecfg}, $storeid, $vmid);
 189             PVE::Storage::foreach_volid($dl, sub {
 190                 my ($volid, $sid, $volname) = @_;
 191
 192                 # check if storage is available on target node
 193                 PVE::Storage::storage_check_node($self->{storecfg}, $sid, $self->{node});
 194
 195                 $volhash->{$volid} = 1;
 196             });
 197         }
 198
 199         # and add used,owned/non-shared disks (just to be sure we have all)
 200
 201         my $sharedvm = 1;
 202         PVE::QemuServer::foreach_drive($conf, sub {
 203             my ($ds, $drive) = @_;
 204
 205             my $volid = $drive->{file};
 206             return if !$volid;
 207
 208             die "cant migrate local file/device '$volid'\n" if $volid =~ m|^/|;
 209
 210             if (PVE::QemuServer::drive_is_cdrom($drive)) {
 211                 die "cant migrate local cdrom drive\n" if $volid eq 'cdrom';
 212                 return if $volid eq 'none';
 213                 $cdromhash->{$volid} = 1;
 214             }
 215
 216             my ($sid, $volname) = PVE::Storage::parse_volume_id($volid);
 217
 218             # check if storage is available on both nodes
 219             my $scfg = PVE::Storage::storage_check_node($self->{storecfg}, $sid);
 220             PVE::Storage::storage_check_node($self->{storecfg}, $sid, $self->{node});
 221
 222             return if $scfg->{shared};
 223
 224             die "can't migrate local cdrom '$volid'\n" if $cdromhash->{$volid};
 225
 226             $sharedvm = 0;
 227
 228             my ($path, $owner) = PVE::Storage::path($self->{storecfg}, $volid);
 229
 230             die "can't migrate volume '$volid' - owned by other VM (owner = VM $owner)\n"
 231                 if !$owner || ($owner != $self->{vmid});
 232
 233             $volhash->{$volid} = 1;
 234         });
 235
 236         if ($self->{running} && !$sharedvm) {
 237             die "can't do online migration - VM uses local disks\n";
 238         }
 239
 240         # do some checks first
 241         foreach my $volid (keys %$volhash) {
 242             my ($sid, $volname) = PVE::Storage::parse_volume_id($volid);
 243             my $scfg =  PVE::Storage::storage_config($self->{storecfg}, $sid);
 244
 245             die "can't migrate '$volid' - storagy type '$scfg->{type}' not supported\n"
 246                 if $scfg->{type} ne 'dir';
 247         }
 248
 249         foreach my $volid (keys %$volhash) {
 250             my ($sid, $volname) = PVE::Storage::parse_volume_id($volid);
 251             push @{$self->{volumes}}, $volid;
 252             PVE::Storage::storage_migrate($self->{storecfg}, $volid, $self->{nodeip}, $sid);
 253         }
 254     };
 255     die "Failed to sync data - $@" if $@;
 256 }
 257
 258 sub phase1 {
 259     my ($self, $vmid) = @_;
 260
 261     $self->log('info', "starting migration of VM $vmid to node '$self->{node}' ($self->{nodeip})");
 262
 263     my $conf = $self->{vmconf};
 264
 265     # set migrate lock in config file
 266     $conf->{lock} = 'migrate';
 267     PVE::QemuServer::update_config_nolock($vmid, $conf, 1);
 268
 269     sync_disks($self, $vmid);
 270
 271 };
 272
 273 sub phase1_cleanup {
 274     my ($self, $vmid, $err) = @_;
 275
 276     $self->log('info', "aborting phase 1 - cleanup resources");
 277
 278     my $conf = $self->{vmconf};
 279     delete $conf->{lock};
 280     eval { PVE::QemuServer::update_config_nolock($vmid, $conf, 1) };
 281     if (my $err = $@) {
 282         $self->log('err', $err);
 283     }
 284
 285     if ($self->{volumes}) {
 286         foreach my $volid (@{$self->{volumes}}) {
 287             $self->log('err', "found stale volume copy '$volid' on node '$self->{node}'");
 288             # fixme: try to remove ?
 289         }
 290     }
 291 }
 292
 293 sub phase2 {
 294     my ($self, $vmid) = @_;
 295
 296     my $conf = $self->{vmconf};
 297
 298     $self->log('info', "starting VM $vmid on remote node '$self->{node}'");
 299
 300     my $rport;
 301
 302     my $nodename = PVE::INotify::nodename();
 303
 304     ## start on remote node
 305     my $cmd = [@{$self->{rem_ssh}}, 'qm', 'start',
 306                $vmid, '--stateuri', 'tcp', '--skiplock', '--migratedfrom', $nodename];
 307
 308     PVE::Tools::run_command($cmd, outfunc => sub {
 309         my $line = shift;
 310
 311         if ($line =~ m/^migration listens on port (\d+)$/) {
 312             $rport = $1;
 313         }
 314     }, errfunc => sub {});
 315
 316     die "unable to detect remote migration port\n" if !$rport;
 317
 318     $self->log('info', "starting migration tunnel");
 319
 320     ## create tunnel to remote port
 321     my $lport = PVE::QemuServer::next_migrate_port();
 322     $self->{tunnel} = $self->fork_tunnel($self->{nodeip}, $lport, $rport);
 323
 324     $self->log('info', "starting online/live migration on port $lport");
 325     # start migration
 326
 327     my $start = time();
 328
 329     my $capabilities = {};
 330     $capabilities->{capability} =  "xbzrle";
 331     $capabilities->{state} = JSON::false;
 332
 333     eval {
 334         PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate-set-capabilities", capabilities => [$capabilities]);
 335     };
 336
 337     #set cachesize 10% of the total memory
 338     my $cachesize = int($conf->{memory}*1048576/10);
 339     eval {
 340         PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate-set-cache-size", value => $cachesize);
 341     };
 342
 343     eval {
 344         PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate", uri => "tcp:localhost:$lport");
 345     };
 346     my $merr = $@;
 347
 348     my $lstat = 0;
 349     my $usleep = 2000000;
 350     my $i = 0;
 351     my $err_count = 0;
 352     while (1) {
 353         $i++;
 354         my $avglstat = $lstat/$i if $lstat;
 355
 356         usleep($usleep);
 357         my $stat;
 358         eval {
 359             $stat = PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "query-migrate");
 360         };
 361         if (my $err = $@) {
 362             $err_count++;
 363             warn "query migrate failed: $err\n";
 364             if ($err_count <= 5) {
 365                 usleep(1000000);
 366                 next;
 367             }
 368             die "too many query migrate failures - aborting\n";
 369         }
 370         if ($stat->{status} =~ m/^(active|completed|failed|cancelled)$/im) {
 371             $merr = undef;
 372             $err_count = 0;
 373             if ($stat->{status} eq 'completed') {
 374                 my $delay = time() - $start;
 375                 if ($delay > 0) {
 376                     my $mbps = sprintf "%.2f", $conf->{memory}/$delay;
 377                     $self->log('info', "migration speed: $mbps MB/s");
 378                 }
 379             }
 380
 381             if ($stat->{status} eq 'failed' || $stat->{status} eq 'cancelled') {
 382                 die "aborting\n"
 383             }
 384
 385             if ($stat->{status} ne 'active') {
 386                 $self->log('info', "migration status: $stat->{status}");
 387                 last;
 388             }
 389
 390             if ($stat->{ram}->{transferred} ne $lstat) {
 391                 my $trans = $stat->{ram}->{transferred} || 0;
 392                 my $rem = $stat->{ram}->{remaining} || 0;
 393                 my $total = $stat->{ram}->{total} || 0;
 394                 my $xbzrlecachesize = $stat->{"xbzrle-cache"}->{"cache-size"} || 0;
 395                 my $xbzrlebytes = $stat->{"xbzrle-cache"}->{"bytes"} || 0;
 396                 my $xbzrlepages = $stat->{"xbzrle-cache"}->{"pages"} || 0;
 397                 my $xbzrlecachemiss = $stat->{"xbzrle-cache"}->{"cache-miss"} || 0;
 398                 my $xbzrleoverflow = $stat->{"xbzrle-cache"}->{"overflow"} || 0;
 399                 #reduce sleep if remainig memory if lower than the everage transfert
 400                 $usleep = 300000 if $avglstat && $rem < $avglstat;
 401
 402                 $self->log('info', "migration status: $stat->{status} (transferred ${trans}, " .
 403                            "remaining ${rem}), total ${total})");
 404
 405                 #$self->log('info', "migration xbzrle cachesize: ${xbzrlecachesize} transferred ${xbzrlebytes} pages ${xbzrlepages} cachemiss ${xbzrlecachemiss} overflow ${xbzrleoverflow}");
 406             }
 407
 408             $lstat = $stat->{ram}->{transferred};
 409
 410         } else {
 411             die $merr if $merr;
 412             die "unable to parse migration status '$stat->{status}' - aborting\n";
 413         }
 414     }
 415 }
 416
 417 sub phase2_cleanup {
 418     my ($self, $vmid, $err) = @_;
 419
 420     return if !$self->{errors};
 421     $self->{phase2errors} = 1;
 422
 423     $self->log('info', "aborting phase 2 - cleanup resources");
 424
 425     my $conf = $self->{vmconf};
 426     delete $conf->{lock};
 427     eval { PVE::QemuServer::update_config_nolock($vmid, $conf, 1) };
 428     if (my $err = $@) {
 429         $self->log('err', $err);
 430     }
 431
 432     # cleanup ressources on target host
 433     my $nodename = PVE::INotify::nodename();
 434
 435     my $cmd = [@{$self->{rem_ssh}}, 'qm', 'stop', $vmid, '--skiplock', '--migratedfrom', $nodename];
 436     eval{ PVE::Tools::run_command($cmd, outfunc => sub {}, errfunc => sub {}) };
 437     if (my $err = $@) {
 438         $self->log('err', $err);
 439         $self->{errors} = 1;
 440     }
 441 }
 442
 443 sub phase3 {
 444     my ($self, $vmid) = @_;
 445
 446     my $volids = $self->{volumes};
 447     return if $self->{phase2errors};
 448
 449     # destroy local copies
 450     foreach my $volid (@$volids) {
 451         eval { PVE::Storage::vdisk_free($self->{storecfg}, $volid); };
 452         if (my $err = $@) {
 453             $self->log('err', "removing local copy of '$volid' failed - $err");
 454             $self->{errors} = 1;
 455             last if $err =~ /^interrupted by signal$/;
 456         }
 457     }
 458 }
 459
 460 sub phase3_cleanup {
 461     my ($self, $vmid, $err) = @_;
 462
 463     my $conf = $self->{vmconf};
 464     return if $self->{phase2errors};
 465
 466     # move config to remote node
 467     my $conffile = PVE::QemuServer::config_file($vmid);
 468     my $newconffile = PVE::QemuServer::config_file($vmid, $self->{node});
 469
 470     die "Failed to move config to node '$self->{node}' - rename failed: $!\n"
 471         if !rename($conffile, $newconffile);
 472
 473     # now that config file is move, we can resume vm on target if livemigrate
 474     if ($self->{tunnel}) {
 475         my $cmd = [@{$self->{rem_ssh}}, 'qm', 'resume', $vmid, '--skiplock'];
 476         eval{ PVE::Tools::run_command($cmd, outfunc => sub {}, errfunc => sub {}) };
 477         if (my $err = $@) {
 478             $self->log('err', $err);
 479             $self->{errors} = 1;
 480         }
 481     }
 482
 483     # always stop local VM
 484     eval { PVE::QemuServer::vm_stop($self->{storecfg}, $vmid, 1, 1); };
 485     if (my $err = $@) {
 486         $self->log('err', "stopping vm failed - $err");
 487         $self->{errors} = 1;
 488     }
 489
 490     if ($self->{tunnel}) {
 491         eval { finish_tunnel($self, $self->{tunnel});  };
 492         if (my $err = $@) {
 493             $self->log('err', $err);
 494             $self->{errors} = 1;
 495         }
 496     }
 497
 498     # always deactivate volumes - avoid lvm LVs to be active on several nodes
 499     eval {
 500         my $vollist = PVE::QemuServer::get_vm_volumes($conf);
 501         PVE::Storage::deactivate_volumes($self->{storecfg}, $vollist);
 502     };
 503     if (my $err = $@) {
 504         $self->log('err', $err);
 505         $self->{errors} = 1;
 506     }
 507
 508     # clear migrate lock
 509     my $cmd = [ @{$self->{rem_ssh}}, 'qm', 'unlock', $vmid ];
 510     $self->cmd_logerr($cmd, errmsg => "failed to clear migrate lock");
 511 }
 512
 513 sub final_cleanup {
 514     my ($self, $vmid) = @_;
 515
 516     # nothing to do
 517 }
 518
 519 1;