]>
Commit | Line | Data |
---|---|---|
3ea94c60 | 1 | package PVE::QemuMigrate; |
1ef75254 | 2 | |
1e3baf05 | 3 | use strict; |
3ea94c60 | 4 | use warnings; |
6d7450cb | 5 | |
3ea94c60 | 6 | use IO::File; |
1e3baf05 | 7 | use IPC::Open2; |
61b04c6d | 8 | use POSIX qw( WNOHANG ); |
6d7450cb TL |
9 | use Time::HiRes qw( usleep ); |
10 | ||
3ea94c60 | 11 | use PVE::Cluster; |
6d7450cb TL |
12 | use PVE::INotify; |
13 | use PVE::RPCEnvironment; | |
14 | use PVE::Replication; | |
15 | use PVE::ReplicationConfig; | |
16 | use PVE::ReplicationState; | |
1e3baf05 | 17 | use PVE::Storage; |
6d7450cb TL |
18 | use PVE::Tools; |
19 | ||
e0fd2b2f | 20 | use PVE::QemuServer::Drive; |
28e6e180 | 21 | use PVE::QemuServer::Helpers qw(min_version); |
3392d6ca | 22 | use PVE::QemuServer::Machine; |
0a13e08e | 23 | use PVE::QemuServer::Monitor qw(mon_cmd); |
6d7450cb | 24 | use PVE::QemuServer; |
1e3baf05 | 25 | |
6d7450cb | 26 | use PVE::AbstractMigrate; |
16e903f2 | 27 | use base qw(PVE::AbstractMigrate); |
1e3baf05 | 28 | |
1ef75254 | 29 | sub fork_command_pipe { |
46a84fd4 | 30 | my ($self, $cmd) = @_; |
19672434 | 31 | |
1ef75254 DM |
32 | my $reader = IO::File->new(); |
33 | my $writer = IO::File->new(); | |
34 | ||
35 | my $orig_pid = $$; | |
36 | ||
37 | my $cpid; | |
38 | ||
39 | eval { $cpid = open2($reader, $writer, @$cmd); }; | |
40 | ||
41 | my $err = $@; | |
42 | ||
43 | # catch exec errors | |
44 | if ($orig_pid != $$) { | |
46a84fd4 | 45 | $self->log('err', "can't fork command pipe\n"); |
19672434 DM |
46 | POSIX::_exit(1); |
47 | kill('KILL', $$); | |
1ef75254 DM |
48 | } |
49 | ||
50 | die $err if $err; | |
51 | ||
52 | return { writer => $writer, reader => $reader, pid => $cpid }; | |
53 | } | |
54 | ||
19672434 | 55 | sub finish_command_pipe { |
97439670 | 56 | my ($self, $cmdpipe, $timeout) = @_; |
1ef75254 | 57 | |
61b04c6d TL |
58 | my $cpid = $cmdpipe->{pid}; |
59 | return if !defined($cpid); | |
60 | ||
1ef75254 DM |
61 | my $writer = $cmdpipe->{writer}; |
62 | my $reader = $cmdpipe->{reader}; | |
63 | ||
64 | $writer->close(); | |
65 | $reader->close(); | |
66 | ||
61b04c6d TL |
67 | my $collect_child_process = sub { |
68 | my $res = waitpid($cpid, WNOHANG); | |
69 | if (defined($res) && ($res == $cpid)) { | |
70 | delete $cmdpipe->{cpid}; | |
71 | return 1; | |
72 | } else { | |
73 | return 0; | |
74 | } | |
75 | }; | |
1ef75254 | 76 | |
97439670 DM |
77 | if ($timeout) { |
78 | for (my $i = 0; $i < $timeout; $i++) { | |
61b04c6d | 79 | return if &$collect_child_process(); |
97439670 DM |
80 | sleep(1); |
81 | } | |
82 | } | |
83 | ||
84 | $self->log('info', "ssh tunnel still running - terminating now with SIGTERM\n"); | |
85 | kill(15, $cpid); | |
1ef75254 | 86 | |
97439670 DM |
87 | # wait again |
88 | for (my $i = 0; $i < 10; $i++) { | |
61b04c6d | 89 | return if &$collect_child_process(); |
97439670 DM |
90 | sleep(1); |
91 | } | |
92 | ||
93 | $self->log('info', "ssh tunnel still running - terminating now with SIGKILL\n"); | |
94 | kill 9, $cpid; | |
95 | sleep 1; | |
61b04c6d TL |
96 | |
97 | $self->log('err', "ssh tunnel child process (PID $cpid) couldn't be collected\n") | |
98 | if !&$collect_child_process(); | |
1ef75254 DM |
99 | } |
100 | ||
e0eb1f76 FG |
101 | sub read_tunnel { |
102 | my ($self, $tunnel, $timeout) = @_; | |
103 | ||
104 | $timeout = 60 if !defined($timeout); | |
105 | ||
106 | my $reader = $tunnel->{reader}; | |
107 | ||
108 | my $output; | |
109 | eval { | |
110 | PVE::Tools::run_with_timeout($timeout, sub { $output = <$reader>; }); | |
111 | }; | |
112 | die "reading from tunnel failed: $@\n" if $@; | |
113 | ||
114 | chomp $output; | |
115 | ||
116 | return $output; | |
117 | } | |
118 | ||
119 | sub write_tunnel { | |
120 | my ($self, $tunnel, $timeout, $command) = @_; | |
121 | ||
122 | $timeout = 60 if !defined($timeout); | |
123 | ||
124 | my $writer = $tunnel->{writer}; | |
125 | ||
126 | eval { | |
127 | PVE::Tools::run_with_timeout($timeout, sub { | |
128 | print $writer "$command\n"; | |
129 | $writer->flush(); | |
130 | }); | |
131 | }; | |
132 | die "writing to tunnel failed: $@\n" if $@; | |
bcb51ae8 FG |
133 | |
134 | if ($tunnel->{version} && $tunnel->{version} >= 1) { | |
135 | my $res = eval { $self->read_tunnel($tunnel, 10); }; | |
136 | die "no reply to command '$command': $@\n" if $@; | |
137 | ||
138 | if ($res eq 'OK') { | |
139 | return; | |
140 | } else { | |
141 | die "tunnel replied '$res' to command '$command'\n"; | |
142 | } | |
143 | } | |
e0eb1f76 FG |
144 | } |
145 | ||
1e3baf05 | 146 | sub fork_tunnel { |
1c9d54bf | 147 | my ($self, $tunnel_addr) = @_; |
1e3baf05 | 148 | |
7827de41 ML |
149 | my @localtunnelinfo = (); |
150 | foreach my $addr (@$tunnel_addr) { | |
151 | push @localtunnelinfo, '-L', $addr; | |
152 | } | |
5bc1e039 | 153 | |
d7b1b24b | 154 | my $cmd = [@{$self->{rem_ssh}}, '-o ExitOnForwardFailure=yes', @localtunnelinfo, '/usr/sbin/qm', 'mtunnel' ]; |
19672434 | 155 | |
46a84fd4 | 156 | my $tunnel = $self->fork_command_pipe($cmd); |
1e3baf05 | 157 | |
19672434 | 158 | eval { |
e0eb1f76 | 159 | my $helo = $self->read_tunnel($tunnel, 60); |
1e3baf05 | 160 | die "no reply\n" if !$helo; |
1ef75254 | 161 | die "no quorum on target node\n" if $helo =~ m/^no quorum$/; |
19672434 | 162 | die "got strange reply from mtunnel ('$helo')\n" |
1e3baf05 DM |
163 | if $helo !~ m/^tunnel online$/; |
164 | }; | |
165 | my $err = $@; | |
166 | ||
58cbe639 FG |
167 | eval { |
168 | my $ver = $self->read_tunnel($tunnel, 10); | |
169 | if ($ver =~ /^ver (\d+)$/) { | |
170 | $tunnel->{version} = $1; | |
171 | $self->log('info', "ssh tunnel $ver\n"); | |
172 | } else { | |
173 | $err = "received invalid tunnel version string '$ver'\n" if !$err; | |
174 | } | |
175 | }; | |
176 | ||
1e3baf05 | 177 | if ($err) { |
46a84fd4 | 178 | $self->finish_command_pipe($tunnel); |
1e3baf05 DM |
179 | die "can't open migration tunnel - $err"; |
180 | } | |
181 | return $tunnel; | |
182 | } | |
183 | ||
19672434 | 184 | sub finish_tunnel { |
16e903f2 | 185 | my ($self, $tunnel) = @_; |
1e3baf05 | 186 | |
e0eb1f76 | 187 | eval { $self->write_tunnel($tunnel, 30, 'quit'); }; |
1e3baf05 | 188 | my $err = $@; |
19672434 | 189 | |
97439670 | 190 | $self->finish_command_pipe($tunnel, 30); |
19672434 | 191 | |
1c9d54bf TL |
192 | if ($tunnel->{sock_addr}) { |
193 | # ssh does not clean up on local host | |
7827de41 | 194 | my $cmd = ['rm', '-f', @{$tunnel->{sock_addr}}]; # |
1c9d54bf TL |
195 | PVE::Tools::run_command($cmd); |
196 | ||
197 | # .. and just to be sure check on remote side | |
198 | unshift @{$cmd}, @{$self->{rem_ssh}}; | |
199 | PVE::Tools::run_command($cmd); | |
200 | } | |
201 | ||
1e3baf05 DM |
202 | die $err if $err; |
203 | } | |
204 | ||
16e903f2 DM |
205 | sub lock_vm { |
206 | my ($self, $vmid, $code, @param) = @_; | |
f5eb281a | 207 | |
ffda963f | 208 | return PVE::QemuConfig->lock_config($vmid, $code, @param); |
16e903f2 | 209 | } |
ff1a2432 | 210 | |
16e903f2 DM |
211 | sub prepare { |
212 | my ($self, $vmid) = @_; | |
ff1a2432 | 213 | |
16e903f2 | 214 | my $online = $self->{opts}->{online}; |
3ea94c60 | 215 | |
16e903f2 | 216 | $self->{storecfg} = PVE::Storage::config(); |
3ea94c60 | 217 | |
e1fc368d | 218 | # test if VM exists |
ffda963f | 219 | my $conf = $self->{vmconf} = PVE::QemuConfig->load_config($vmid); |
3ea94c60 | 220 | |
ffda963f | 221 | PVE::QemuConfig->check_lock($conf); |
3ea94c60 | 222 | |
16e903f2 DM |
223 | my $running = 0; |
224 | if (my $pid = PVE::QemuServer::check_running($vmid)) { | |
b6adff33 | 225 | die "can't migrate running VM without --online\n" if !$online; |
16e903f2 | 226 | $running = $pid; |
42dbd2ee | 227 | |
3392d6ca | 228 | $self->{forcemachine} = PVE::QemuServer::Machine::qemu_machine_pxe($vmid, $conf); |
7bac824e | 229 | |
3ea94c60 | 230 | } |
ca6abacf TM |
231 | my $loc_res = PVE::QemuServer::check_local_resources($conf, 1); |
232 | if (scalar @$loc_res) { | |
16e903f2 | 233 | if ($self->{running} || !$self->{opts}->{force}) { |
ca6abacf | 234 | die "can't migrate VM which uses local devices: " . join(", ", @$loc_res) . "\n"; |
16e903f2 DM |
235 | } else { |
236 | $self->log('info', "migrating VM which uses local devices"); | |
237 | } | |
3ea94c60 DM |
238 | } |
239 | ||
ff1a2432 | 240 | my $vollist = PVE::QemuServer::get_vm_volumes($conf); |
16e903f2 | 241 | |
73f5ee92 | 242 | my $need_activate = []; |
29701766 FG |
243 | foreach my $volid (@$vollist) { |
244 | my ($sid, $volname) = PVE::Storage::parse_volume_id($volid, 1); | |
245 | ||
246 | # check if storage is available on both nodes | |
bf8fc5a3 | 247 | my $targetsid = PVE::QemuServer::map_storage($self->{opts}->{storagemap}, $sid); |
b74cad8a | 248 | |
29701766 | 249 | my $scfg = PVE::Storage::storage_check_node($self->{storecfg}, $sid); |
b74cad8a | 250 | PVE::Storage::storage_check_node($self->{storecfg}, $targetsid, $self->{node}); |
73f5ee92 FG |
251 | |
252 | if ($scfg->{shared}) { | |
253 | # PVE::Storage::activate_storage checks this for non-shared storages | |
254 | my $plugin = PVE::Storage::Plugin->lookup($scfg->{type}); | |
255 | warn "Used shared storage '$sid' is not online on source node!\n" | |
256 | if !$plugin->check_connection($sid, $scfg); | |
257 | } else { | |
258 | # only activate if not shared | |
9e93a63f | 259 | next if ($volid =~ m/vm-\d+-cloudinit/); |
73f5ee92 FG |
260 | push @$need_activate, $volid; |
261 | } | |
29701766 | 262 | } |
3ea94c60 | 263 | |
73f5ee92 FG |
264 | # activate volumes |
265 | PVE::Storage::activate_volumes($self->{storecfg}, $need_activate); | |
266 | ||
3ea94c60 | 267 | # test ssh connection |
16e903f2 DM |
268 | my $cmd = [ @{$self->{rem_ssh}}, '/bin/true' ]; |
269 | eval { $self->cmd_quiet($cmd); }; | |
3ea94c60 | 270 | die "Can't connect to destination address using public key\n" if $@; |
ff1a2432 | 271 | |
16e903f2 | 272 | return $running; |
3ea94c60 DM |
273 | } |
274 | ||
275 | sub sync_disks { | |
16e903f2 DM |
276 | my ($self, $vmid) = @_; |
277 | ||
16e903f2 DM |
278 | my $conf = $self->{vmconf}; |
279 | ||
dabf2473 | 280 | # local volumes which have been copied |
16e903f2 | 281 | $self->{volumes} = []; |
3ea94c60 | 282 | |
b10afa31 | 283 | my $storecfg = $self->{storecfg}; |
3ea94c60 DM |
284 | eval { |
285 | ||
dabf2473 FG |
286 | # found local volumes and their origin |
287 | my $local_volumes = {}; | |
5bf7f0f1 FG |
288 | my $local_volumes_errors = {}; |
289 | my $other_errors = []; | |
290 | my $abort = 0; | |
3ea94c60 | 291 | |
5bf7f0f1 FG |
292 | my $log_error = sub { |
293 | my ($msg, $volid) = @_; | |
294 | ||
295 | if (defined($volid)) { | |
296 | $local_volumes_errors->{$volid} = $msg; | |
297 | } else { | |
298 | push @$other_errors, $msg; | |
299 | } | |
300 | $abort = 1; | |
301 | }; | |
302 | ||
b10afa31 | 303 | my @sids = PVE::Storage::storage_ids($storecfg); |
86638cc2 | 304 | foreach my $storeid (@sids) { |
b10afa31 | 305 | my $scfg = PVE::Storage::storage_config($storecfg, $storeid); |
86638cc2 | 306 | next if $scfg->{shared}; |
b10afa31 | 307 | next if !PVE::Storage::storage_check_enabled($storecfg, $storeid, undef, 1); |
373ea579 | 308 | |
86638cc2 | 309 | # get list from PVE::Storage (for unused volumes) |
b10afa31 | 310 | my $dl = PVE::Storage::vdisk_list($storecfg, $storeid, $vmid); |
89719f98 FG |
311 | |
312 | next if @{$dl->{$storeid}} == 0; | |
313 | ||
bf8fc5a3 | 314 | my $targetsid = PVE::QemuServer::map_storage($self->{opts}->{storagemap}, $storeid); |
86638cc2 | 315 | # check if storage is available on target node |
b10afa31 | 316 | PVE::Storage::storage_check_node($storecfg, $targetsid, $self->{node}); |
89719f98 | 317 | |
bf8fc5a3 FG |
318 | # grandfather in existing mismatches |
319 | if ($targetsid ne $storeid) { | |
320 | my $target_scfg = PVE::Storage::storage_config($storecfg, $targetsid); | |
321 | die "content type 'images' is not available on storage '$targetsid'\n" | |
322 | if !$target_scfg->{content}->{images}; | |
323 | } | |
324 | ||
86638cc2 | 325 | PVE::Storage::foreach_volid($dl, sub { |
5eca0c36 | 326 | my ($volid, $sid, $volinfo) = @_; |
80b2cbd1 | 327 | |
6f58fce9 | 328 | $local_volumes->{$volid}->{ref} = 'storage'; |
5eca0c36 FE |
329 | |
330 | # If with_snapshots is not set for storage migrate, it tries to use | |
331 | # a raw+size stream, but on-the-fly conversion from qcow2 to raw+size | |
332 | # back to qcow2 is currently not possible. | |
333 | $local_volumes->{$volid}->{snapshots} = ($volinfo->{format} =~ /^(?:qcow2|vmdk)$/); | |
0ad295f9 | 334 | $local_volumes->{$volid}->{format} = $volinfo->{format}; |
86638cc2 FG |
335 | }); |
336 | } | |
3ea94c60 | 337 | |
88126be3 FG |
338 | my $rep_cfg = PVE::ReplicationConfig->new(); |
339 | my $replication_jobcfg = $rep_cfg->find_local_replication_job($vmid, $self->{node}); | |
2cd808d3 TL |
340 | my $replicatable_volumes = !$replication_jobcfg ? {} |
341 | : PVE::QemuConfig->get_replicatable_volumes($storecfg, $vmid, $conf, 0, 1); | |
b9f44d27 | 342 | |
3629c19d | 343 | my $test_volid = sub { |
aee6abe5 | 344 | my ($volid, $attr) = @_; |
3ea94c60 | 345 | |
5bf7f0f1 | 346 | if ($volid =~ m|^/|) { |
ec82e3ee | 347 | return if $attr->{shared}; |
6f58fce9 | 348 | $local_volumes->{$volid}->{ref} = 'config'; |
5bf7f0f1 FG |
349 | die "local file/device\n"; |
350 | } | |
3ea94c60 | 351 | |
aee6abe5 DM |
352 | my $snaprefs = $attr->{referenced_in_snapshot}; |
353 | ||
354 | if ($attr->{cdrom}) { | |
5bf7f0f1 FG |
355 | if ($volid eq 'cdrom') { |
356 | my $msg = "can't migrate local cdrom drive"; | |
5009a8c7 | 357 | if (defined($snaprefs) && !$attr->{referenced_in_config}) { |
aee6abe5 | 358 | my $snapnames = join(', ', sort keys %$snaprefs); |
5009a8c7 | 359 | $msg .= " (referenced in snapshot - $snapnames)"; |
aee6abe5 | 360 | } |
5bf7f0f1 FG |
361 | &$log_error("$msg\n"); |
362 | return; | |
363 | } | |
3ea94c60 | 364 | return if $volid eq 'none'; |
3ea94c60 DM |
365 | } |
366 | ||
367 | my ($sid, $volname) = PVE::Storage::parse_volume_id($volid); | |
368 | ||
bf8fc5a3 | 369 | my $targetsid = PVE::QemuServer::map_storage($self->{opts}->{storagemap}, $sid); |
16e903f2 | 370 | # check if storage is available on both nodes |
b10afa31 TL |
371 | my $scfg = PVE::Storage::storage_check_node($storecfg, $sid); |
372 | PVE::Storage::storage_check_node($storecfg, $targetsid, $self->{node}); | |
3ea94c60 DM |
373 | |
374 | return if $scfg->{shared}; | |
375 | ||
6f58fce9 | 376 | $local_volumes->{$volid}->{ref} = $attr->{referenced_in_config} ? 'config' : 'snapshot'; |
d62fcf74 | 377 | |
9e93a63f ML |
378 | if ($attr->{cdrom}) { |
379 | if ($volid =~ /vm-\d+-cloudinit/) { | |
380 | $local_volumes->{$volid}->{ref} = 'generated'; | |
381 | return; | |
382 | } | |
383 | die "local cdrom image\n"; | |
384 | } | |
3629c19d | 385 | |
b10afa31 | 386 | my ($path, $owner) = PVE::Storage::path($storecfg, $volid); |
3ea94c60 | 387 | |
5bf7f0f1 | 388 | die "owned by other VM (owner = VM $owner)\n" |
b10afa31 | 389 | if !$owner || ($owner != $vmid); |
3ea94c60 | 390 | |
aee6abe5 | 391 | if (defined($snaprefs)) { |
5eca0c36 FE |
392 | $local_volumes->{$volid}->{snapshots} = 1; |
393 | ||
3629c19d DM |
394 | # we cannot migrate shapshots on local storage |
395 | # exceptions: 'zfspool' or 'qcow2' files (on directory storage) | |
396 | ||
b74cad8a | 397 | die "online storage migration not possible if snapshot exists\n" if $self->{running}; |
0ad295f9 | 398 | if (!($scfg->{type} eq 'zfspool' || $local_volumes->{$volid}->{format} eq 'qcow2')) { |
5bf7f0f1 | 399 | die "non-migratable snapshot exists\n"; |
3629c19d | 400 | } |
3629c19d | 401 | } |
3a7bc9e2 FG |
402 | |
403 | die "referenced by linked clone(s)\n" | |
b10afa31 | 404 | if PVE::Storage::volume_is_base_and_used($storecfg, $volid); |
3629c19d DM |
405 | }; |
406 | ||
aee6abe5 DM |
407 | PVE::QemuServer::foreach_volid($conf, sub { |
408 | my ($volid, $attr) = @_; | |
409 | eval { $test_volid->($volid, $attr); }; | |
410 | if (my $err = $@) { | |
411 | &$log_error($err, $volid); | |
412 | } | |
413 | }); | |
3ea94c60 | 414 | |
dabf2473 | 415 | foreach my $vol (sort keys %$local_volumes) { |
b9f44d27 | 416 | my $type = $replicatable_volumes->{$vol} ? 'local, replicated' : 'local'; |
6f58fce9 WB |
417 | my $ref = $local_volumes->{$vol}->{ref}; |
418 | if ($ref eq 'storage') { | |
b9f44d27 | 419 | $self->log('info', "found $type disk '$vol' (via storage)\n"); |
6f58fce9 | 420 | } elsif ($ref eq 'config') { |
dbc9420b DM |
421 | &$log_error("can't live migrate attached local disks without with-local-disks option\n", $vol) |
422 | if $self->{running} && !$self->{opts}->{"with-local-disks"}; | |
b9f44d27 | 423 | $self->log('info', "found $type disk '$vol' (in current VM config)\n"); |
6f58fce9 | 424 | } elsif ($ref eq 'snapshot') { |
b9f44d27 | 425 | $self->log('info', "found $type disk '$vol' (referenced by snapshot(s))\n"); |
9e93a63f ML |
426 | } elsif ($ref eq 'generated') { |
427 | $self->log('info', "found generated disk '$vol' (in current VM config)\n"); | |
d62fcf74 | 428 | } else { |
b9f44d27 | 429 | $self->log('info', "found $type disk '$vol'\n"); |
d62fcf74 FG |
430 | } |
431 | } | |
432 | ||
5bf7f0f1 FG |
433 | foreach my $vol (sort keys %$local_volumes_errors) { |
434 | $self->log('warn', "can't migrate local disk '$vol': $local_volumes_errors->{$vol}"); | |
435 | } | |
436 | foreach my $err (@$other_errors) { | |
437 | $self->log('warn', "$err"); | |
438 | } | |
439 | ||
5bf7f0f1 FG |
440 | if ($abort) { |
441 | die "can't migrate VM - check log\n"; | |
442 | } | |
443 | ||
c4d2d6c1 | 444 | # additional checks for local storage |
dabf2473 | 445 | foreach my $volid (keys %$local_volumes) { |
3ea94c60 | 446 | my ($sid, $volname) = PVE::Storage::parse_volume_id($volid); |
b10afa31 | 447 | my $scfg = PVE::Storage::storage_config($storecfg, $sid); |
3ea94c60 | 448 | |
dad06e20 | 449 | my $migratable = $scfg->{type} =~ /^(?:dir|zfspool|lvmthin|lvm)$/; |
c4d2d6c1 | 450 | |
37a6dc78 | 451 | die "can't migrate '$volid' - storage type '$scfg->{type}' not supported\n" |
c4d2d6c1 | 452 | if !$migratable; |
d5604092 | 453 | |
c4d2d6c1 | 454 | # image is a linked clone on local storage, se we can't migrate. |
b10afa31 | 455 | if (my $basename = (PVE::Storage::parse_volname($storecfg, $volid))[3]) { |
c4d2d6c1 | 456 | die "can't migrate '$volid' as it's a clone of '$basename'"; |
d5604092 | 457 | } |
3ea94c60 DM |
458 | } |
459 | ||
88126be3 | 460 | if ($replication_jobcfg) { |
9b6efe43 | 461 | if ($self->{running}) { |
28e6e180 TL |
462 | |
463 | my $version = PVE::QemuServer::kvm_user_version(); | |
464 | if (!min_version($version, 4, 2)) { | |
465 | die "can't live migrate VM with replicated volumes, pve-qemu to old (< 4.2)!\n" | |
466 | } | |
467 | ||
9b6efe43 FG |
468 | my $live_replicatable_volumes = {}; |
469 | PVE::QemuServer::foreach_drive($conf, sub { | |
470 | my ($ds, $drive) = @_; | |
471 | ||
472 | my $volid = $drive->{file}; | |
473 | $live_replicatable_volumes->{$ds} = $volid | |
474 | if defined($replicatable_volumes->{$volid}); | |
475 | }); | |
476 | foreach my $drive (keys %$live_replicatable_volumes) { | |
477 | my $volid = $live_replicatable_volumes->{$drive}; | |
478 | ||
479 | my $bitmap = "repl_$drive"; | |
480 | ||
481 | # start tracking before replication to get full delta + a few duplicates | |
482 | $self->log('info', "$drive: start tracking writes using block-dirty-bitmap '$bitmap'"); | |
483 | mon_cmd($vmid, 'block-dirty-bitmap-add', node => "drive-$drive", name => $bitmap); | |
484 | ||
485 | # other info comes from target node in phase 2 | |
486 | $self->{target_drive}->{$drive}->{bitmap} = $bitmap; | |
487 | } | |
488 | } | |
e85d01f2 | 489 | $self->log('info', "replicating disk images"); |
9b6efe43 | 490 | |
dbc9420b | 491 | my $start_time = time(); |
e85d01f2 TL |
492 | my $logfunc = sub { $self->log('info', shift) }; |
493 | $self->{replicated_volumes} = PVE::Replication::run_replication( | |
88126be3 | 494 | 'PVE::QemuConfig', $replication_jobcfg, $start_time, $start_time, $logfunc); |
dbc9420b DM |
495 | } |
496 | ||
68b108ee SR |
497 | # sizes in config have to be accurate for remote node to correctly |
498 | # allocate disks, rescan to be sure | |
b10afa31 | 499 | my $volid_hash = PVE::QemuServer::scan_volids($storecfg, $vmid); |
68b108ee SR |
500 | PVE::QemuServer::foreach_drive($conf, sub { |
501 | my ($key, $drive) = @_; | |
e0fd2b2f | 502 | my ($updated, $old_size, $new_size) = PVE::QemuServer::Drive::update_disksize($drive, $volid_hash); |
68b108ee SR |
503 | if (defined($updated)) { |
504 | $conf->{$key} = PVE::QemuServer::print_drive($updated); | |
505 | $self->log('info', "size of disk '$updated->{file}' ($key) updated from $old_size to $new_size\n"); | |
506 | } | |
507 | }); | |
508 | ||
818ce80e DC |
509 | # we want to set the efidisk size in the config to the size of the |
510 | # real OVMF_VARS.fd image, else we can create a too big image, which does not work | |
511 | if (defined($conf->{efidisk0})) { | |
512 | PVE::QemuServer::update_efidisk_size($conf); | |
513 | } | |
514 | ||
e85d01f2 TL |
515 | $self->log('info', "copying local disk images") if scalar(%$local_volumes); |
516 | ||
dabf2473 | 517 | foreach my $volid (keys %$local_volumes) { |
3ea94c60 | 518 | my ($sid, $volname) = PVE::Storage::parse_volume_id($volid); |
bf8fc5a3 | 519 | my $targetsid = PVE::QemuServer::map_storage($self->{opts}->{storagemap}, $sid); |
9e93a63f ML |
520 | my $ref = $local_volumes->{$volid}->{ref}; |
521 | if ($self->{running} && $ref eq 'config') { | |
b74cad8a | 522 | push @{$self->{online_local_volumes}}, $volid; |
9e93a63f | 523 | } elsif ($ref eq 'generated') { |
9860fe4e | 524 | die "can't live migrate VM with local cloudinit disk. use a shared storage instead\n" if $self->{running}; |
9e93a63f ML |
525 | # skip all generated volumes but queue them for deletion in phase3_cleanup |
526 | push @{$self->{volumes}}, $volid; | |
527 | next; | |
b74cad8a | 528 | } else { |
e85d01f2 | 529 | next if $self->{replicated_volumes}->{$volid}; |
b74cad8a | 530 | push @{$self->{volumes}}, $volid; |
15a37695 SI |
531 | my $opts = $self->{opts}; |
532 | my $insecure = $opts->{migration_type} eq 'insecure'; | |
6f58fce9 | 533 | my $with_snapshots = $local_volumes->{$volid}->{snapshots}; |
15a37695 | 534 | # use 'migrate' limit for transfer to other node |
29eb909e | 535 | my $bwlimit = PVE::Storage::get_bandwidth_limit('migration', [$targetsid, $sid], $opts->{bwlimit}); |
15a37695 SI |
536 | # JSONSchema and get_bandwidth_limit use kbps - storage_migrate bps |
537 | $bwlimit = $bwlimit * 1024 if defined($bwlimit); | |
538 | ||
b10afa31 | 539 | PVE::Storage::storage_migrate($storecfg, $volid, $self->{ssh_info}, $targetsid, |
15a37695 | 540 | undef, undef, undef, $bwlimit, $insecure, $with_snapshots); |
b74cad8a | 541 | } |
3ea94c60 DM |
542 | } |
543 | }; | |
544 | die "Failed to sync data - $@" if $@; | |
545 | } | |
546 | ||
b74cad8a AD |
547 | sub cleanup_remotedisks { |
548 | my ($self) = @_; | |
549 | ||
550 | foreach my $target_drive (keys %{$self->{target_drive}}) { | |
6a039d06 FG |
551 | my $drivestr = $self->{target_drive}->{$target_drive}->{drivestr}; |
552 | next if !defined($drivestr); | |
553 | ||
554 | my $drive = PVE::QemuServer::parse_drive($target_drive, $drivestr); | |
9b3f5a5c | 555 | |
9b6efe43 | 556 | # don't clean up replicated disks! |
9b3f5a5c | 557 | next if defined($self->{replicated_volumes}->{$drive->{file}}); |
b74cad8a | 558 | |
b74cad8a AD |
559 | my ($storeid, $volname) = PVE::Storage::parse_volume_id($drive->{file}); |
560 | ||
561 | my $cmd = [@{$self->{rem_ssh}}, 'pvesm', 'free', "$storeid:$volname"]; | |
562 | ||
563 | eval{ PVE::Tools::run_command($cmd, outfunc => sub {}, errfunc => sub {}) }; | |
564 | if (my $err = $@) { | |
565 | $self->log('err', $err); | |
566 | $self->{errors} = 1; | |
567 | } | |
568 | } | |
569 | } | |
570 | ||
9b6efe43 FG |
571 | sub cleanup_bitmaps { |
572 | my ($self) = @_; | |
7f5fb49a | 573 | foreach my $drive (keys %{$self->{target_drive}}) { |
9b6efe43 FG |
574 | my $bitmap = $self->{target_drive}->{$drive}->{bitmap}; |
575 | next if !$bitmap; | |
576 | $self->log('info', "$drive: removing block-dirty-bitmap '$bitmap'"); | |
577 | mon_cmd($self->{vmid}, 'block-dirty-bitmap-remove', node => "drive-$drive", name => $bitmap); | |
578 | } | |
579 | } | |
580 | ||
1e3baf05 | 581 | sub phase1 { |
16e903f2 | 582 | my ($self, $vmid) = @_; |
1e3baf05 | 583 | |
16e903f2 | 584 | $self->log('info', "starting migration of VM $vmid to node '$self->{node}' ($self->{nodeip})"); |
1e3baf05 | 585 | |
16e903f2 | 586 | my $conf = $self->{vmconf}; |
1e3baf05 DM |
587 | |
588 | # set migrate lock in config file | |
1858638f | 589 | $conf->{lock} = 'migrate'; |
ffda963f | 590 | PVE::QemuConfig->write_config($vmid, $conf); |
1e3baf05 | 591 | |
16e903f2 | 592 | sync_disks($self, $vmid); |
1ef75254 | 593 | |
68b108ee SR |
594 | # sync_disks fixes disk sizes to match their actual size, write changes so |
595 | # target allocates correct volumes | |
596 | PVE::QemuConfig->write_config($vmid, $conf); | |
1e3baf05 DM |
597 | }; |
598 | ||
16e903f2 DM |
599 | sub phase1_cleanup { |
600 | my ($self, $vmid, $err) = @_; | |
601 | ||
602 | $self->log('info', "aborting phase 1 - cleanup resources"); | |
603 | ||
1858638f DM |
604 | my $conf = $self->{vmconf}; |
605 | delete $conf->{lock}; | |
ffda963f | 606 | eval { PVE::QemuConfig->write_config($vmid, $conf) }; |
16e903f2 DM |
607 | if (my $err = $@) { |
608 | $self->log('err', $err); | |
609 | } | |
f5eb281a | 610 | |
16e903f2 DM |
611 | if ($self->{volumes}) { |
612 | foreach my $volid (@{$self->{volumes}}) { | |
613 | $self->log('err', "found stale volume copy '$volid' on node '$self->{node}'"); | |
614 | # fixme: try to remove ? | |
615 | } | |
616 | } | |
9b6efe43 FG |
617 | |
618 | eval { $self->cleanup_bitmaps() }; | |
619 | if (my $err =$@) { | |
620 | $self->log('err', $err); | |
621 | } | |
622 | ||
16e903f2 DM |
623 | } |
624 | ||
1e3baf05 | 625 | sub phase2 { |
16e903f2 | 626 | my ($self, $vmid) = @_; |
1e3baf05 | 627 | |
16e903f2 DM |
628 | my $conf = $self->{vmconf}; |
629 | ||
46a84fd4 | 630 | $self->log('info', "starting VM $vmid on remote node '$self->{node}'"); |
1e3baf05 | 631 | |
5bc1e039 | 632 | my $raddr; |
1e3baf05 | 633 | my $rport; |
1c9d54bf | 634 | my $ruri; # the whole migration dst. URI (protocol:address[:port]) |
7e8dcf2c AD |
635 | my $nodename = PVE::INotify::nodename(); |
636 | ||
19672434 | 637 | ## start on remote node |
95a4b4a9 AD |
638 | my $cmd = [@{$self->{rem_ssh}}]; |
639 | ||
7c14dcae | 640 | my $spice_ticket; |
86b8228b | 641 | if (PVE::QemuServer::vga_conf_has_spice($conf->{vga})) { |
0a13e08e | 642 | my $res = mon_cmd($vmid, 'query-spice'); |
7c14dcae | 643 | $spice_ticket = $res->{ticket}; |
95a4b4a9 AD |
644 | } |
645 | ||
1c9d54bf TL |
646 | push @$cmd , 'qm', 'start', $vmid, '--skiplock', '--migratedfrom', $nodename; |
647 | ||
f1c2a53a | 648 | my $migration_type = $self->{opts}->{migration_type}; |
2de2d6f7 TL |
649 | |
650 | push @$cmd, '--migration_type', $migration_type; | |
651 | ||
652 | push @$cmd, '--migration_network', $self->{opts}->{migration_network} | |
653 | if $self->{opts}->{migration_network}; | |
654 | ||
655 | if ($migration_type eq 'insecure') { | |
1c9d54bf TL |
656 | push @$cmd, '--stateuri', 'tcp'; |
657 | } else { | |
658 | push @$cmd, '--stateuri', 'unix'; | |
659 | } | |
95a4b4a9 | 660 | |
42668529 DM |
661 | if ($self->{forcemachine}) { |
662 | push @$cmd, '--machine', $self->{forcemachine}; | |
663 | } | |
664 | ||
4530494b TL |
665 | if ($self->{online_local_volumes}) { |
666 | push @$cmd, '--targetstorage', ($self->{opts}->{targetstorage} // '1'); | |
b74cad8a AD |
667 | } |
668 | ||
86b8228b | 669 | my $spice_port; |
7827de41 ML |
670 | my $tunnel_addr = []; |
671 | my $sock_addr = []; | |
672 | # version > 0 for unix socket support | |
673 | my $nbd_protocol_version = 1; | |
ff09c795 ML |
674 | # TODO change to 'spice_ticket: <ticket>\n' in 7.0 |
675 | my $input = $spice_ticket ? "$spice_ticket\n" : "\n"; | |
676 | $input .= "nbd_protocol_version: $nbd_protocol_version\n"; | |
88126be3 FG |
677 | foreach my $volid (keys %{$self->{replicated_volumes}}) { |
678 | $input .= "replicated_volume: $volid\n"; | |
679 | } | |
680 | ||
681 | my $target_replicated_volumes = {}; | |
86b8228b | 682 | |
7c14dcae DM |
683 | # Note: We try to keep $spice_ticket secret (do not pass via command line parameter) |
684 | # instead we pipe it through STDIN | |
7827de41 | 685 | my $exitcode = PVE::Tools::run_command($cmd, input => $input, outfunc => sub { |
1e3baf05 DM |
686 | my $line = shift; |
687 | ||
407e0b8b | 688 | if ($line =~ m/^migration listens on tcp:(localhost|[\d\.]+|\[[\d\.:a-fA-F]+\]):(\d+)$/) { |
5bc1e039 SP |
689 | $raddr = $1; |
690 | $rport = int($2); | |
1c9d54bf TL |
691 | $ruri = "tcp:$raddr:$rport"; |
692 | } | |
693 | elsif ($line =~ m!^migration listens on unix:(/run/qemu-server/(\d+)\.migrate)$!) { | |
694 | $raddr = $1; | |
695 | die "Destination UNIX sockets VMID does not match source VMID" if $vmid ne $2; | |
696 | $ruri = "unix:$raddr"; | |
5bc1e039 SP |
697 | } |
698 | elsif ($line =~ m/^migration listens on port (\d+)$/) { | |
699 | $raddr = "localhost"; | |
86b8228b | 700 | $rport = int($1); |
1c9d54bf | 701 | $ruri = "tcp:$raddr:$rport"; |
5bc1e039 | 702 | } |
f3a483b6 | 703 | elsif ($line =~ m/^spice listens on port (\d+)$/) { |
86b8228b | 704 | $spice_port = int($1); |
1e3baf05 | 705 | } |
769f187d | 706 | elsif ($line =~ m/^storage migration listens on nbd:(localhost|[\d\.]+|\[[\d\.:a-fA-F]+\]):(\d+):exportname=(\S+) volume:(\S+)$/) { |
8b02e568 | 707 | my $drivestr = $4; |
b74cad8a AD |
708 | my $nbd_uri = "nbd:$1:$2:exportname=$3"; |
709 | my $targetdrive = $3; | |
710 | $targetdrive =~ s/drive-//g; | |
711 | ||
8b02e568 | 712 | $self->{target_drive}->{$targetdrive}->{drivestr} = $drivestr; |
b74cad8a | 713 | $self->{target_drive}->{$targetdrive}->{nbd_uri} = $nbd_uri; |
7827de41 ML |
714 | } elsif ($line =~ m!^storage migration listens on nbd:unix:(/run/qemu-server/(\d+)_nbd\.migrate):exportname=(\S+) volume:(\S+)$!) { |
715 | my $drivestr = $4; | |
716 | die "Destination UNIX socket's VMID does not match source VMID" if $vmid ne $2; | |
717 | my $nbd_unix_addr = $1; | |
718 | my $nbd_uri = "nbd:unix:$nbd_unix_addr:exportname=$3"; | |
719 | my $targetdrive = $3; | |
720 | $targetdrive =~ s/drive-//g; | |
b74cad8a | 721 | |
7827de41 ML |
722 | $self->{target_drive}->{$targetdrive}->{drivestr} = $drivestr; |
723 | $self->{target_drive}->{$targetdrive}->{nbd_uri} = $nbd_uri; | |
724 | push @$tunnel_addr, "$nbd_unix_addr:$nbd_unix_addr"; | |
725 | push @$sock_addr, $nbd_unix_addr; | |
88126be3 FG |
726 | } elsif ($line =~ m/^re-using replicated volume: (\S+) - (.*)$/) { |
727 | my $drive = $1; | |
728 | my $volid = $2; | |
729 | $target_replicated_volumes->{$volid} = $drive; | |
8bf30c2a SR |
730 | } elsif ($line =~ m/^QEMU: (.*)$/) { |
731 | $self->log('info', "[$self->{node}] $1\n"); | |
b74cad8a | 732 | } |
ab399b7c AD |
733 | }, errfunc => sub { |
734 | my $line = shift; | |
8bf30c2a | 735 | $self->log('info', "[$self->{node}] $line"); |
6e0216d8 SR |
736 | }, noerr => 1); |
737 | ||
738 | die "remote command failed with exit code $exitcode\n" if $exitcode; | |
1e3baf05 | 739 | |
5bc1e039 | 740 | die "unable to detect remote migration address\n" if !$raddr; |
1ef75254 | 741 | |
88126be3 FG |
742 | if (scalar(keys %$target_replicated_volumes) != scalar(keys %{$self->{replicated_volumes}})) { |
743 | die "number of replicated disks on source and target node do not match - target node too old?\n" | |
744 | } | |
745 | ||
d296ed08 AD |
746 | $self->log('info', "start remote tunnel"); |
747 | ||
2de2d6f7 | 748 | if ($migration_type eq 'secure') { |
1c9d54bf TL |
749 | |
750 | if ($ruri =~ /^unix:/) { | |
54323eed | 751 | unlink $raddr; |
7827de41 ML |
752 | push @$tunnel_addr, "$raddr:$raddr"; |
753 | $self->{tunnel} = $self->fork_tunnel($tunnel_addr); | |
754 | push @$sock_addr, $raddr; | |
1c9d54bf TL |
755 | |
756 | my $unix_socket_try = 0; # wait for the socket to become ready | |
7827de41 | 757 | while ($unix_socket_try <= 100) { |
1c9d54bf | 758 | $unix_socket_try++; |
7827de41 ML |
759 | my $available = 0; |
760 | foreach my $sock (@$sock_addr) { | |
761 | if (-S $sock) { | |
762 | $available++; | |
763 | } | |
764 | } | |
765 | ||
766 | if ($available == @$sock_addr) { | |
767 | last; | |
1c9d54bf TL |
768 | } |
769 | ||
770 | usleep(50000); | |
771 | } | |
7827de41 ML |
772 | if ($unix_socket_try > 100) { |
773 | $self->{errors} = 1; | |
774 | $self->finish_tunnel($self->{tunnel}); | |
775 | die "Timeout, migration socket $ruri did not get ready"; | |
776 | } | |
1c9d54bf TL |
777 | |
778 | } elsif ($ruri =~ /^tcp:/) { | |
e858e9d2 TL |
779 | my $tunnel_addr; |
780 | if ($raddr eq "localhost") { | |
781 | # for backwards compatibility with older qemu-server versions | |
782 | my $pfamily = PVE::Tools::get_host_address_family($nodename); | |
783 | my $lport = PVE::Tools::next_migrate_port($pfamily); | |
784 | $tunnel_addr = "$lport:localhost:$rport"; | |
785 | } | |
1c9d54bf | 786 | |
e858e9d2 | 787 | $self->{tunnel} = $self->fork_tunnel($tunnel_addr); |
1c9d54bf TL |
788 | |
789 | } else { | |
790 | die "unsupported protocol in migration URI: $ruri\n"; | |
791 | } | |
d296ed08 AD |
792 | } else { |
793 | #fork tunnel for insecure migration, to send faster commands like resume | |
794 | $self->{tunnel} = $self->fork_tunnel(); | |
1c9d54bf | 795 | } |
7827de41 | 796 | $self->{tunnel}->{sock_addr} = $sock_addr if (@$sock_addr); |
1e3baf05 | 797 | |
1e3baf05 | 798 | my $start = time(); |
b74cad8a | 799 | |
d189e590 SI |
800 | my $opt_bwlimit = $self->{opts}->{bwlimit}; |
801 | ||
c7789f54 | 802 | if (defined($self->{online_local_volumes})) { |
b74cad8a AD |
803 | $self->{storage_migration} = 1; |
804 | $self->{storage_migration_jobs} = {}; | |
805 | $self->log('info', "starting storage migration"); | |
806 | ||
bd2d5fe6 | 807 | die "The number of local disks does not match between the source and the destination.\n" |
3b4cf0f0 | 808 | if (scalar(keys %{$self->{target_drive}}) != scalar @{$self->{online_local_volumes}}); |
b74cad8a | 809 | foreach my $drive (keys %{$self->{target_drive}}){ |
d189e590 SI |
810 | my $target = $self->{target_drive}->{$drive}; |
811 | my $nbd_uri = $target->{nbd_uri}; | |
683ab654 | 812 | |
1764fa05 | 813 | my $source_drive = PVE::QemuServer::parse_drive($drive, $conf->{$drive}); |
683ab654 FG |
814 | my $target_drive = PVE::QemuServer::parse_drive($drive, $target->{drivestr}); |
815 | ||
1764fa05 FE |
816 | my $source_sid = PVE::Storage::Plugin::parse_volume_id($source_drive->{file}); |
817 | my $target_sid = PVE::Storage::Plugin::parse_volume_id($target_drive->{file}); | |
683ab654 | 818 | |
29eb909e | 819 | my $bwlimit = PVE::Storage::get_bandwidth_limit('migration', [$source_sid, $target_sid], $opt_bwlimit); |
9b6efe43 | 820 | my $bitmap = $target->{bitmap}; |
d189e590 | 821 | |
d108cb1e | 822 | $self->log('info', "$drive: start migration to $nbd_uri"); |
9b6efe43 | 823 | PVE::QemuServer::qemu_drive_mirror($vmid, $drive, $nbd_uri, $vmid, undef, $self->{storage_migration_jobs}, 'skip', undef, $bwlimit, $bitmap); |
b74cad8a AD |
824 | } |
825 | } | |
826 | ||
1c9d54bf | 827 | $self->log('info', "starting online/live migration on $ruri"); |
5bc1e039 | 828 | $self->{livemigration} = 1; |
e18b0b99 | 829 | |
3beb415b AD |
830 | # load_defaults |
831 | my $defaults = PVE::QemuServer::load_defaults(); | |
832 | ||
485449e3 SR |
833 | $self->log('info', "set migration_caps"); |
834 | eval { | |
835 | PVE::QemuServer::set_migration_caps($vmid); | |
836 | }; | |
837 | warn $@ if $@; | |
838 | ||
839 | my $qemu_migrate_params = {}; | |
840 | ||
ddd664d7 SI |
841 | # migrate speed can be set via bwlimit (datacenter.cfg and API) and via the |
842 | # migrate_speed parameter in qm.conf - take the lower of the two. | |
29eb909e | 843 | my $bwlimit = PVE::Storage::get_bandwidth_limit('migration', undef, $opt_bwlimit) // 0; |
ddd664d7 SI |
844 | my $migrate_speed = $conf->{migrate_speed} // $bwlimit; |
845 | # migrate_speed is in MB/s, bwlimit in KB/s | |
846 | $migrate_speed *= 1024; | |
847 | ||
848 | $migrate_speed = ($bwlimit < $migrate_speed) ? $bwlimit : $migrate_speed; | |
849 | ||
850 | # always set migrate speed (overwrite kvm default of 32m) we set a very high | |
851 | # default of 8192m which is basically unlimited | |
852 | $migrate_speed ||= ($defaults->{migrate_speed} || 8192) * 1024; | |
853 | ||
854 | # qmp takes migrate_speed in B/s. | |
855 | $migrate_speed *= 1024; | |
485449e3 SR |
856 | $self->log('info', "migration speed limit: $migrate_speed B/s"); |
857 | $qemu_migrate_params->{'max-bandwidth'} = int($migrate_speed); | |
3beb415b AD |
858 | |
859 | my $migrate_downtime = $defaults->{migrate_downtime}; | |
860 | $migrate_downtime = $conf->{migrate_downtime} if defined($conf->{migrate_downtime}); | |
c05f1b33 SR |
861 | # migrate-set-parameters expects limit in ms |
862 | $migrate_downtime *= 1000; | |
863 | $self->log('info', "migration downtime limit: $migrate_downtime ms"); | |
864 | $qemu_migrate_params->{'downtime-limit'} = int($migrate_downtime); | |
3beb415b | 865 | |
171ed95c EK |
866 | # set cachesize to 10% of the total memory |
867 | my $memory = $conf->{memory} || $defaults->{memory}; | |
868 | my $cachesize = int($memory * 1048576 / 10); | |
50d8dd5d AD |
869 | $cachesize = round_powerof2($cachesize); |
870 | ||
485449e3 SR |
871 | $self->log('info', "migration cachesize: $cachesize B"); |
872 | $qemu_migrate_params->{'xbzrle-cache-size'} = int($cachesize); | |
873 | ||
874 | $self->log('info', "set migration parameters"); | |
e18b0b99 | 875 | eval { |
485449e3 | 876 | mon_cmd($vmid, "migrate-set-parameters", %{$qemu_migrate_params}); |
e18b0b99 | 877 | }; |
485449e3 | 878 | $self->log('info', "migrate-set-parameters error: $@") if $@; |
f34d1466 | 879 | |
86b8228b | 880 | if (PVE::QemuServer::vga_conf_has_spice($conf->{vga})) { |
95a4b4a9 AD |
881 | my $rpcenv = PVE::RPCEnvironment::get(); |
882 | my $authuser = $rpcenv->get_user(); | |
883 | ||
86b8228b | 884 | my (undef, $proxyticket) = PVE::AccessControl::assemble_spice_ticket($authuser, $vmid, $self->{node}); |
95a4b4a9 | 885 | |
86b8228b | 886 | my $filename = "/etc/pve/nodes/$self->{node}/pve-ssl.pem"; |
769f187d | 887 | my $subject = PVE::AccessControl::read_x509_subject_spice($filename); |
95a4b4a9 AD |
888 | |
889 | $self->log('info', "spice client_migrate_info"); | |
890 | ||
891 | eval { | |
0a13e08e | 892 | mon_cmd($vmid, "client_migrate_info", protocol => 'spice', |
ccab68c2 | 893 | hostname => $proxyticket, 'port' => 0, 'tls-port' => $spice_port, |
86b8228b | 894 | 'cert-subject' => $subject); |
95a4b4a9 AD |
895 | }; |
896 | $self->log('info', "client_migrate_info error: $@") if $@; | |
897 | ||
898 | } | |
899 | ||
f34d1466 | 900 | $self->log('info', "start migrate command to $ruri"); |
5a7835f5 | 901 | eval { |
0a13e08e | 902 | mon_cmd($vmid, "migrate", uri => $ruri); |
5a7835f5 AD |
903 | }; |
904 | my $merr = $@; | |
1c9d54bf | 905 | $self->log('info', "migrate uri => $ruri failed: $merr") if $merr; |
1e3baf05 | 906 | |
a05b47a8 | 907 | my $lstat = 0; |
4305207d | 908 | my $usleep = 1000000; |
e52bd94c | 909 | my $i = 0; |
b0b756c1 | 910 | my $err_count = 0; |
865ef132 SP |
911 | my $lastrem = undef; |
912 | my $downtimecounter = 0; | |
1e3baf05 | 913 | while (1) { |
e52bd94c AD |
914 | $i++; |
915 | my $avglstat = $lstat/$i if $lstat; | |
916 | ||
b0b756c1 DM |
917 | usleep($usleep); |
918 | my $stat; | |
919 | eval { | |
0a13e08e | 920 | $stat = mon_cmd($vmid, "query-migrate"); |
b0b756c1 DM |
921 | }; |
922 | if (my $err = $@) { | |
923 | $err_count++; | |
924 | warn "query migrate failed: $err\n"; | |
f34d1466 | 925 | $self->log('info', "query migrate failed: $err"); |
b0b756c1 DM |
926 | if ($err_count <= 5) { |
927 | usleep(1000000); | |
928 | next; | |
929 | } | |
930 | die "too many query migrate failures - aborting\n"; | |
931 | } | |
985a5f48 | 932 | |
f34d1466 | 933 | if (defined($stat->{status}) && $stat->{status} =~ m/^(setup)$/im) { |
985a5f48 AD |
934 | sleep(1); |
935 | next; | |
936 | } | |
937 | ||
f34d1466 | 938 | if (defined($stat->{status}) && $stat->{status} =~ m/^(active|completed|failed|cancelled)$/im) { |
d68afb26 | 939 | $merr = undef; |
b0b756c1 | 940 | $err_count = 0; |
5a7835f5 | 941 | if ($stat->{status} eq 'completed') { |
1e3baf05 DM |
942 | my $delay = time() - $start; |
943 | if ($delay > 0) { | |
171ed95c | 944 | my $mbps = sprintf "%.2f", $memory / $delay; |
135007c0 AD |
945 | my $downtime = $stat->{downtime} || 0; |
946 | $self->log('info', "migration speed: $mbps MB/s - downtime $downtime ms"); | |
1e3baf05 DM |
947 | } |
948 | } | |
f5eb281a | 949 | |
5a7835f5 | 950 | if ($stat->{status} eq 'failed' || $stat->{status} eq 'cancelled') { |
f34d1466 | 951 | $self->log('info', "migration status error: $stat->{status}"); |
1e3baf05 DM |
952 | die "aborting\n" |
953 | } | |
954 | ||
a05b47a8 DM |
955 | if ($stat->{status} ne 'active') { |
956 | $self->log('info', "migration status: $stat->{status}"); | |
957 | last; | |
958 | } | |
959 | ||
960 | if ($stat->{ram}->{transferred} ne $lstat) { | |
961 | my $trans = $stat->{ram}->{transferred} || 0; | |
962 | my $rem = $stat->{ram}->{remaining} || 0; | |
963 | my $total = $stat->{ram}->{total} || 0; | |
e18b0b99 AD |
964 | my $xbzrlecachesize = $stat->{"xbzrle-cache"}->{"cache-size"} || 0; |
965 | my $xbzrlebytes = $stat->{"xbzrle-cache"}->{"bytes"} || 0; | |
966 | my $xbzrlepages = $stat->{"xbzrle-cache"}->{"pages"} || 0; | |
967 | my $xbzrlecachemiss = $stat->{"xbzrle-cache"}->{"cache-miss"} || 0; | |
968 | my $xbzrleoverflow = $stat->{"xbzrle-cache"}->{"overflow"} || 0; | |
4305207d FG |
969 | # reduce sleep if remainig memory is lower than the average transfer speed |
970 | $usleep = 100000 if $avglstat && $rem < $avglstat; | |
a05b47a8 DM |
971 | |
972 | $self->log('info', "migration status: $stat->{status} (transferred ${trans}, " . | |
0302101c | 973 | "remaining ${rem}), total ${total})"); |
e18b0b99 | 974 | |
2e787b18 SP |
975 | if (${xbzrlecachesize}) { |
976 | $self->log('info', "migration xbzrle cachesize: ${xbzrlecachesize} transferred ${xbzrlebytes} pages ${xbzrlepages} cachemiss ${xbzrlecachemiss} overflow ${xbzrleoverflow}"); | |
977 | } | |
978 | ||
865ef132 SP |
979 | if (($lastrem && $rem > $lastrem ) || ($rem == 0)) { |
980 | $downtimecounter++; | |
981 | } | |
982 | $lastrem = $rem; | |
983 | ||
984 | if ($downtimecounter > 5) { | |
985 | $downtimecounter = 0; | |
986 | $migrate_downtime *= 2; | |
c05f1b33 | 987 | $self->log('info', "auto-increased downtime to continue migration: $migrate_downtime ms"); |
865ef132 | 988 | eval { |
c05f1b33 SR |
989 | # migrate-set-parameters does not touch values not |
990 | # specified, so this only changes downtime-limit | |
991 | mon_cmd($vmid, "migrate-set-parameters", 'downtime-limit' => int($migrate_downtime)); | |
865ef132 | 992 | }; |
c05f1b33 | 993 | $self->log('info', "migrate-set-parameters error: $@") if $@; |
865ef132 SP |
994 | } |
995 | ||
a05b47a8 DM |
996 | } |
997 | ||
865ef132 | 998 | |
a05b47a8 | 999 | $lstat = $stat->{ram}->{transferred}; |
370b05e7 | 1000 | |
1e3baf05 | 1001 | } else { |
d68afb26 | 1002 | die $merr if $merr; |
5a7835f5 | 1003 | die "unable to parse migration status '$stat->{status}' - aborting\n"; |
1e3baf05 | 1004 | } |
a05b47a8 | 1005 | } |
1e3baf05 | 1006 | } |
16e903f2 | 1007 | |
c04b5b04 AD |
1008 | sub phase2_cleanup { |
1009 | my ($self, $vmid, $err) = @_; | |
1010 | ||
af30308f DM |
1011 | return if !$self->{errors}; |
1012 | $self->{phase2errors} = 1; | |
1013 | ||
c04b5b04 AD |
1014 | $self->log('info', "aborting phase 2 - cleanup resources"); |
1015 | ||
19168b91 SP |
1016 | $self->log('info', "migrate_cancel"); |
1017 | eval { | |
0a13e08e | 1018 | mon_cmd($vmid, "migrate_cancel"); |
19168b91 SP |
1019 | }; |
1020 | $self->log('info', "migrate_cancel error: $@") if $@; | |
1021 | ||
c04b5b04 AD |
1022 | my $conf = $self->{vmconf}; |
1023 | delete $conf->{lock}; | |
ffda963f | 1024 | eval { PVE::QemuConfig->write_config($vmid, $conf) }; |
c04b5b04 AD |
1025 | if (my $err = $@) { |
1026 | $self->log('err', $err); | |
1027 | } | |
1028 | ||
af30308f | 1029 | # cleanup ressources on target host |
3b4cf0f0 | 1030 | if ($self->{storage_migration}) { |
b74cad8a AD |
1031 | eval { PVE::QemuServer::qemu_blockjobs_cancel($vmid, $self->{storage_migration_jobs}) }; |
1032 | if (my $err = $@) { | |
1033 | $self->log('err', $err); | |
1034 | } | |
9b3f5a5c | 1035 | } |
b74cad8a | 1036 | |
9b3f5a5c FG |
1037 | eval { $self->cleanup_bitmaps() }; |
1038 | if (my $err =$@) { | |
1039 | $self->log('err', $err); | |
b74cad8a AD |
1040 | } |
1041 | ||
af30308f | 1042 | my $nodename = PVE::INotify::nodename(); |
370b05e7 | 1043 | |
af30308f DM |
1044 | my $cmd = [@{$self->{rem_ssh}}, 'qm', 'stop', $vmid, '--skiplock', '--migratedfrom', $nodename]; |
1045 | eval{ PVE::Tools::run_command($cmd, outfunc => sub {}, errfunc => sub {}) }; | |
1046 | if (my $err = $@) { | |
1047 | $self->log('err', $err); | |
1048 | $self->{errors} = 1; | |
1049 | } | |
386c6ba7 | 1050 | |
9b3f5a5c FG |
1051 | # cleanup after stopping, otherwise disks might be in-use by target VM! |
1052 | eval { PVE::QemuMigrate::cleanup_remotedisks($self) }; | |
1053 | if (my $err = $@) { | |
1054 | $self->log('err', $err); | |
1055 | } | |
1056 | ||
1057 | ||
386c6ba7 WL |
1058 | if ($self->{tunnel}) { |
1059 | eval { finish_tunnel($self, $self->{tunnel}); }; | |
1060 | if (my $err = $@) { | |
1061 | $self->log('err', $err); | |
1062 | $self->{errors} = 1; | |
1063 | } | |
1064 | } | |
c04b5b04 AD |
1065 | } |
1066 | ||
16e903f2 DM |
1067 | sub phase3 { |
1068 | my ($self, $vmid) = @_; | |
f5eb281a | 1069 | |
16e903f2 | 1070 | my $volids = $self->{volumes}; |
af30308f | 1071 | return if $self->{phase2errors}; |
16e903f2 DM |
1072 | |
1073 | # destroy local copies | |
1074 | foreach my $volid (@$volids) { | |
46883f80 DM |
1075 | eval { PVE::Storage::vdisk_free($self->{storecfg}, $volid); }; |
1076 | if (my $err = $@) { | |
1077 | $self->log('err', "removing local copy of '$volid' failed - $err"); | |
1078 | $self->{errors} = 1; | |
1079 | last if $err =~ /^interrupted by signal$/; | |
16e903f2 DM |
1080 | } |
1081 | } | |
16e903f2 DM |
1082 | } |
1083 | ||
1084 | sub phase3_cleanup { | |
1085 | my ($self, $vmid, $err) = @_; | |
1086 | ||
1087 | my $conf = $self->{vmconf}; | |
af30308f | 1088 | return if $self->{phase2errors}; |
16e903f2 | 1089 | |
1d5aaa1d FG |
1090 | my $tunnel = $self->{tunnel}; |
1091 | ||
b74cad8a | 1092 | if ($self->{storage_migration}) { |
db1f8b39 | 1093 | # finish block-job with block-job-cancel, to disconnect source VM from NBD |
1e0074c4 TL |
1094 | # to avoid it trying to re-establish it. We are in blockjob ready state, |
1095 | # thus, this command changes to it to blockjob complete (see qapi docs) | |
db1f8b39 | 1096 | eval { PVE::QemuServer::qemu_drive_mirror_monitor($vmid, undef, $self->{storage_migration_jobs}, 'cancel'); }; |
b74cad8a AD |
1097 | |
1098 | if (my $err = $@) { | |
1099 | eval { PVE::QemuServer::qemu_blockjobs_cancel($vmid, $self->{storage_migration_jobs}) }; | |
1100 | eval { PVE::QemuMigrate::cleanup_remotedisks($self) }; | |
40a572f7 | 1101 | die "Failed to complete storage migration: $err\n"; |
b74cad8a | 1102 | } else { |
b74cad8a | 1103 | foreach my $target_drive (keys %{$self->{target_drive}}) { |
8b02e568 | 1104 | my $drive = PVE::QemuServer::parse_drive($target_drive, $self->{target_drive}->{$target_drive}->{drivestr}); |
71c58bb7 | 1105 | $conf->{$target_drive} = PVE::QemuServer::print_drive($drive); |
b74cad8a AD |
1106 | PVE::QemuConfig->write_config($vmid, $conf); |
1107 | } | |
1108 | } | |
1109 | } | |
1110 | ||
dbc9420b | 1111 | # transfer replication state before move config |
4bdd20ab | 1112 | $self->transfer_replication_state() if $self->{replicated_volumes}; |
dbc9420b | 1113 | |
b8d20802 | 1114 | # move config to remote node |
ffda963f FG |
1115 | my $conffile = PVE::QemuConfig->config_file($vmid); |
1116 | my $newconffile = PVE::QemuConfig->config_file($vmid, $self->{node}); | |
b8d20802 AD |
1117 | |
1118 | die "Failed to move config to node '$self->{node}' - rename failed: $!\n" | |
1119 | if !rename($conffile, $newconffile); | |
1120 | ||
4bdd20ab | 1121 | $self->switch_replication_job_target() if $self->{replicated_volumes}; |
dbc9420b | 1122 | |
5bc1e039 | 1123 | if ($self->{livemigration}) { |
504105c6 FG |
1124 | if ($self->{storage_migration}) { |
1125 | # stop nbd server on remote vm - requirement for resume since 2.9 | |
1126 | my $cmd = [@{$self->{rem_ssh}}, 'qm', 'nbdstop', $vmid]; | |
1127 | ||
1128 | eval{ PVE::Tools::run_command($cmd, outfunc => sub {}, errfunc => sub {}) }; | |
1129 | if (my $err = $@) { | |
1130 | $self->log('err', $err); | |
1131 | $self->{errors} = 1; | |
1132 | } | |
1133 | } | |
1d5aaa1d | 1134 | |
877e2ea7 | 1135 | # config moved and nbd server stopped - now we can resume vm on target |
1d5aaa1d FG |
1136 | if ($tunnel && $tunnel->{version} && $tunnel->{version} >= 1) { |
1137 | eval { | |
1138 | $self->write_tunnel($tunnel, 30, "resume $vmid"); | |
1139 | }; | |
1140 | if (my $err = $@) { | |
1141 | $self->log('err', $err); | |
1142 | $self->{errors} = 1; | |
1143 | } | |
1144 | } else { | |
1145 | my $cmd = [@{$self->{rem_ssh}}, 'qm', 'resume', $vmid, '--skiplock', '--nocheck']; | |
1146 | my $logf = sub { | |
1147 | my $line = shift; | |
1148 | $self->log('err', $line); | |
1149 | }; | |
1150 | eval { PVE::Tools::run_command($cmd, outfunc => sub {}, errfunc => $logf); }; | |
1151 | if (my $err = $@) { | |
1152 | $self->log('err', $err); | |
1153 | $self->{errors} = 1; | |
1154 | } | |
b67900f1 | 1155 | } |
ca662131 SI |
1156 | |
1157 | if ($self->{storage_migration} && PVE::QemuServer::parse_guest_agent($conf)->{fstrim_cloned_disks} && $self->{running}) { | |
1158 | my $cmd = [@{$self->{rem_ssh}}, 'qm', 'guest', 'cmd', $vmid, 'fstrim']; | |
1159 | eval{ PVE::Tools::run_command($cmd, outfunc => sub {}, errfunc => sub {}) }; | |
1160 | } | |
b67900f1 AD |
1161 | } |
1162 | ||
2e7fee87 FG |
1163 | # close tunnel on successful migration, on error phase2_cleanup closed it |
1164 | if ($tunnel) { | |
1165 | eval { finish_tunnel($self, $tunnel); }; | |
1166 | if (my $err = $@) { | |
1167 | $self->log('err', $err); | |
1168 | $self->{errors} = 1; | |
1169 | } | |
1170 | } | |
1171 | ||
fd8469f7 | 1172 | eval { |
fd8469f7 AD |
1173 | my $timer = 0; |
1174 | if (PVE::QemuServer::vga_conf_has_spice($conf->{vga}) && $self->{running}) { | |
1175 | $self->log('info', "Waiting for spice server migration"); | |
1176 | while (1) { | |
0a13e08e | 1177 | my $res = mon_cmd($vmid, 'query-spice'); |
fd8469f7 AD |
1178 | last if int($res->{'migrated'}) == 1; |
1179 | last if $timer > 50; | |
1180 | $timer ++; | |
1181 | usleep(200000); | |
769f187d | 1182 | } |
fd8469f7 AD |
1183 | } |
1184 | }; | |
95a4b4a9 | 1185 | |
16e903f2 DM |
1186 | # always stop local VM |
1187 | eval { PVE::QemuServer::vm_stop($self->{storecfg}, $vmid, 1, 1); }; | |
1188 | if (my $err = $@) { | |
1189 | $self->log('err', "stopping vm failed - $err"); | |
1190 | $self->{errors} = 1; | |
1191 | } | |
1192 | ||
1193 | # always deactivate volumes - avoid lvm LVs to be active on several nodes | |
1194 | eval { | |
1195 | my $vollist = PVE::QemuServer::get_vm_volumes($conf); | |
1196 | PVE::Storage::deactivate_volumes($self->{storecfg}, $vollist); | |
1197 | }; | |
1198 | if (my $err = $@) { | |
1199 | $self->log('err', $err); | |
1200 | $self->{errors} = 1; | |
1201 | } | |
1202 | ||
b74cad8a AD |
1203 | if($self->{storage_migration}) { |
1204 | # destroy local copies | |
1205 | my $volids = $self->{online_local_volumes}; | |
1206 | ||
1207 | foreach my $volid (@$volids) { | |
9b6efe43 FG |
1208 | # keep replicated volumes! |
1209 | next if $self->{replicated_volumes}->{$volid}; | |
1210 | ||
b74cad8a AD |
1211 | eval { PVE::Storage::vdisk_free($self->{storecfg}, $volid); }; |
1212 | if (my $err = $@) { | |
1213 | $self->log('err', "removing local copy of '$volid' failed - $err"); | |
1214 | $self->{errors} = 1; | |
1215 | last if $err =~ /^interrupted by signal$/; | |
1216 | } | |
1217 | } | |
1218 | ||
b74cad8a AD |
1219 | } |
1220 | ||
16e903f2 DM |
1221 | # clear migrate lock |
1222 | my $cmd = [ @{$self->{rem_ssh}}, 'qm', 'unlock', $vmid ]; | |
1223 | $self->cmd_logerr($cmd, errmsg => "failed to clear migrate lock"); | |
1224 | } | |
1225 | ||
1226 | sub final_cleanup { | |
1227 | my ($self, $vmid) = @_; | |
1228 | ||
1229 | # nothing to do | |
1230 | } | |
1231 | ||
50d8dd5d AD |
1232 | sub round_powerof2 { |
1233 | return 1 if $_[0] < 2; | |
1234 | return 2 << int(log($_[0]-1)/log(2)); | |
1235 | } | |
1236 | ||
16e903f2 | 1237 | 1; |