]>
git.proxmox.com Git - qemu-server.git/blob - PVE/QemuMigrate.pm
1 package PVE
::QemuMigrate
;
5 use PVE
::AbstractMigrate
;
12 use Time
::HiRes
qw( usleep );
14 use base
qw(PVE::AbstractMigrate);
16 sub fork_command_pipe
{
17 my ($self, $cmd) = @_;
19 my $reader = IO
::File-
>new();
20 my $writer = IO
::File-
>new();
26 eval { $cpid = open2
($reader, $writer, @$cmd); };
31 if ($orig_pid != $$) {
32 $self->log('err', "can't fork command pipe\n");
39 return { writer
=> $writer, reader
=> $reader, pid
=> $cpid };
42 sub finish_command_pipe
{
43 my ($self, $cmdpipe, $timeout) = @_;
45 my $writer = $cmdpipe->{writer
};
46 my $reader = $cmdpipe->{reader
};
51 my $cpid = $cmdpipe->{pid
};
54 for (my $i = 0; $i < $timeout; $i++) {
55 return if !PVE
::ProcFSTools
::check_process_running
($cpid);
60 $self->log('info', "ssh tunnel still running - terminating now with SIGTERM\n");
64 for (my $i = 0; $i < 10; $i++) {
65 return if !PVE
::ProcFSTools
::check_process_running
($cpid);
69 $self->log('info', "ssh tunnel still running - terminating now with SIGKILL\n");
75 my ($self, $nodeip, $lport, $rport) = @_;
77 my $cmd = [@{$self->{rem_ssh
}}, '-L', "$lport:localhost:$rport",
80 my $tunnel = $self->fork_command_pipe($cmd);
82 my $reader = $tunnel->{reader
};
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 $helo !~ m/^tunnel online$/;
95 $self->finish_command_pipe($tunnel);
96 die "can't open migration tunnel - $err";
102 my ($self, $tunnel) = @_;
104 my $writer = $tunnel->{writer
};
107 PVE
::Tools
::run_with_timeout
(30, sub {
108 print $writer "quit\n";
114 $self->finish_command_pipe($tunnel, 30);
120 my ($self, $vmid, $code, @param) = @_;
122 return PVE
::QemuServer
::lock_config
($vmid, $code, @param);
126 my ($self, $vmid) = @_;
128 my $online = $self->{opts
}->{online
};
130 $self->{storecfg
} = PVE
::Storage
::config
();
133 my $conf = $self->{vmconf
} = PVE
::QemuServer
::load_config
($vmid);
135 PVE
::QemuServer
::check_lock
($conf);
138 if (my $pid = PVE
::QemuServer
::check_running
($vmid)) {
139 die "cant migrate running VM without --online\n" if !$online;
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";
147 $self->log('info', "migrating VM which uses local devices");
152 my $vollist = PVE
::QemuServer
::get_vm_volumes
($conf);
153 PVE
::Storage
::activate_volumes
($self->{storecfg
}, $vollist);
155 # fixme: check if storage is available on both nodes
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 $@;
166 my ($self, $vmid) = @_;
168 $self->log('info', "copying disk images");
170 my $conf = $self->{vmconf
};
172 $self->{volumes
} = [];
183 my @sids = PVE
::Storage
::storage_ids
($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);
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, $volname) = @_;
194 # check if storage is available on target node
195 PVE
::Storage
::storage_check_node
($self->{storecfg
}, $sid, $self->{node
});
197 $volhash->{$volid} = 1;
198 $sharedvm = 0; # there is a non-shared disk
202 # and add used, owned/non-shared disks (just to be sure we have all)
204 PVE
::QemuServer
::foreach_volid
($conf, sub {
205 my ($volid, $is_cdrom) = @_;
209 die "cant migrate local file/device '$volid'\n" if $volid =~ m
|^/|;
212 die "cant migrate local cdrom drive\n" if $volid eq 'cdrom';
213 return if $volid eq 'none';
214 $cdromhash->{$volid} = 1;
217 my ($sid, $volname) = PVE
::Storage
::parse_volume_id
($volid);
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
});
223 return if $scfg->{shared
};
225 die "can't migrate local cdrom '$volid'\n" if $cdromhash->{$volid};
229 my ($path, $owner) = PVE
::Storage
::path
($self->{storecfg
}, $volid);
231 die "can't migrate volume '$volid' - owned by other VM (owner = VM $owner)\n"
232 if !$owner || ($owner != $self->{vmid
});
234 $volhash->{$volid} = 1;
237 if ($self->{running
} && !$sharedvm) {
238 die "can't do online migration - VM uses local disks\n";
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
}, $sid);
246 die "can't migrate '$volid' - storagy type '$scfg->{type}' not supported\n"
247 if $scfg->{type
} ne 'dir';
250 foreach my $volid (keys %$volhash) {
251 my ($sid, $volname) = PVE
::Storage
::parse_volume_id
($volid);
252 push @{$self->{volumes
}}, $volid;
253 PVE
::Storage
::storage_migrate
($self->{storecfg
}, $volid, $self->{nodeip
}, $sid);
256 die "Failed to sync data - $@" if $@;
260 my ($self, $vmid) = @_;
262 $self->log('info', "starting migration of VM $vmid to node '$self->{node}' ($self->{nodeip})");
264 my $conf = $self->{vmconf
};
266 # set migrate lock in config file
267 $conf->{lock} = 'migrate';
268 PVE
::QemuServer
::update_config_nolock
($vmid, $conf, 1);
270 sync_disks
($self, $vmid);
275 my ($self, $vmid, $err) = @_;
277 $self->log('info', "aborting phase 1 - cleanup resources");
279 my $conf = $self->{vmconf
};
280 delete $conf->{lock};
281 eval { PVE
::QemuServer
::update_config_nolock
($vmid, $conf, 1) };
283 $self->log('err', $err);
286 if ($self->{volumes
}) {
287 foreach my $volid (@{$self->{volumes
}}) {
288 $self->log('err', "found stale volume copy '$volid' on node '$self->{node}'");
289 # fixme: try to remove ?
295 my ($self, $vmid) = @_;
297 my $conf = $self->{vmconf
};
299 $self->log('info', "starting VM $vmid on remote node '$self->{node}'");
303 my $nodename = PVE
::INotify
::nodename
();
305 ## start on remote node
306 my $cmd = [@{$self->{rem_ssh
}}, 'qm', 'start',
307 $vmid, '--stateuri', 'tcp', '--skiplock', '--migratedfrom', $nodename];
309 PVE
::Tools
::run_command
($cmd, outfunc
=> sub {
312 if ($line =~ m/^migration listens on port (\d+)$/) {
317 $self->log('info', $line);
320 die "unable to detect remote migration port\n" if !$rport;
322 $self->log('info', "starting migration tunnel");
324 ## create tunnel to remote port
325 my $lport = PVE
::QemuServer
::next_migrate_port
();
326 $self->{tunnel
} = $self->fork_tunnel($self->{nodeip
}, $lport, $rport);
328 $self->log('info', "starting online/live migration on port $lport");
334 my $defaults = PVE
::QemuServer
::load_defaults
();
336 # always set migrate speed (overwrite kvm default of 32m)
337 # we set a very hight default of 8192m which is basically unlimited
338 my $migrate_speed = $defaults->{migrate_speed
} || 8192;
339 $migrate_speed = $conf->{migrate_speed
} || $migrate_speed;
340 $migrate_speed = $migrate_speed * 1048576;
341 $self->log('info', "migrate_set_speed: $migrate_speed");
343 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "migrate_set_speed", value
=> int($migrate_speed));
345 $self->log('info', "migrate_set_speed error: $@") if $@;
347 my $migrate_downtime = $defaults->{migrate_downtime
};
348 $migrate_downtime = $conf->{migrate_downtime
} if defined($conf->{migrate_downtime
});
349 if (defined($migrate_downtime)) {
350 $self->log('info', "migrate_set_downtime: $migrate_downtime");
352 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "migrate_set_downtime", value
=> int($migrate_downtime*100)/100);
354 $self->log('info', "migrate_set_downtime error: $@") if $@;
357 my $capabilities = {};
358 $capabilities->{capability
} = "xbzrle";
359 $capabilities->{state} = JSON
::false
;
362 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "migrate-set-capabilities", capabilities
=> [$capabilities]);
365 #set cachesize 10% of the total memory
366 my $cachesize = int($conf->{memory
}*1048576/10);
368 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "migrate-set-cache-size", value
=> $cachesize);
372 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "migrate", uri
=> "tcp:localhost:$lport");
377 my $usleep = 2000000;
381 my $downtimecounter = 0;
384 my $avglstat = $lstat/$i if $lstat;
389 $stat = PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "query-migrate");
393 warn "query migrate failed: $err\n";
394 if ($err_count <= 5) {
398 die "too many query migrate failures - aborting\n";
400 if ($stat->{status
} =~ m/^(active|completed|failed|cancelled)$/im) {
403 if ($stat->{status
} eq 'completed') {
404 my $delay = time() - $start;
406 my $mbps = sprintf "%.2f", $conf->{memory
}/$delay;
407 my $downtime = $stat->{downtime
} || 0;
408 $self->log('info', "migration speed: $mbps MB/s - downtime $downtime ms");
412 if ($stat->{status
} eq 'failed' || $stat->{status
} eq 'cancelled') {
416 if ($stat->{status
} ne 'active') {
417 $self->log('info', "migration status: $stat->{status}");
421 if ($stat->{ram
}->{transferred
} ne $lstat) {
422 my $trans = $stat->{ram
}->{transferred
} || 0;
423 my $rem = $stat->{ram
}->{remaining
} || 0;
424 my $total = $stat->{ram
}->{total
} || 0;
425 my $xbzrlecachesize = $stat->{"xbzrle-cache"}->{"cache-size"} || 0;
426 my $xbzrlebytes = $stat->{"xbzrle-cache"}->{"bytes"} || 0;
427 my $xbzrlepages = $stat->{"xbzrle-cache"}->{"pages"} || 0;
428 my $xbzrlecachemiss = $stat->{"xbzrle-cache"}->{"cache-miss"} || 0;
429 my $xbzrleoverflow = $stat->{"xbzrle-cache"}->{"overflow"} || 0;
430 #reduce sleep if remainig memory if lower than the everage transfert
431 $usleep = 300000 if $avglstat && $rem < $avglstat;
433 $self->log('info', "migration status: $stat->{status} (transferred ${trans}, " .
434 "remaining ${rem}), total ${total})");
436 #$self->log('info', "migration xbzrle cachesize: ${xbzrlecachesize} transferred ${xbzrlebytes} pages ${xbzrlepages} cachemiss ${xbzrlecachemiss} overflow ${xbzrleoverflow}");
437 if (($lastrem && $rem > $lastrem ) || ($rem == 0)) {
442 if ($downtimecounter > 5) {
443 $downtimecounter = 0;
444 $migrate_downtime *= 2;
445 $self->log('info', "migrate_set_downtime: $migrate_downtime");
447 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "migrate_set_downtime", value
=> int($migrate_downtime*100)/100);
449 $self->log('info', "migrate_set_downtime error: $@") if $@;
455 $lstat = $stat->{ram
}->{transferred
};
459 die "unable to parse migration status '$stat->{status}' - aborting\n";
465 my ($self, $vmid, $err) = @_;
467 return if !$self->{errors
};
468 $self->{phase2errors
} = 1;
470 $self->log('info', "aborting phase 2 - cleanup resources");
472 $self->log('info', "migrate_cancel");
474 PVE
::QemuServer
::vm_mon_cmd_nocheck
($vmid, "migrate_cancel");
476 $self->log('info', "migrate_cancel error: $@") if $@;
478 my $conf = $self->{vmconf
};
479 delete $conf->{lock};
480 eval { PVE
::QemuServer
::update_config_nolock
($vmid, $conf, 1) };
482 $self->log('err', $err);
485 # cleanup ressources on target host
486 my $nodename = PVE
::INotify
::nodename
();
488 my $cmd = [@{$self->{rem_ssh
}}, 'qm', 'stop', $vmid, '--skiplock', '--migratedfrom', $nodename];
489 eval{ PVE
::Tools
::run_command
($cmd, outfunc
=> sub {}, errfunc
=> sub {}) };
491 $self->log('err', $err);
497 my ($self, $vmid) = @_;
499 my $volids = $self->{volumes
};
500 return if $self->{phase2errors
};
502 # destroy local copies
503 foreach my $volid (@$volids) {
504 eval { PVE
::Storage
::vdisk_free
($self->{storecfg
}, $volid); };
506 $self->log('err', "removing local copy of '$volid' failed - $err");
508 last if $err =~ /^interrupted by signal$/;
514 my ($self, $vmid, $err) = @_;
516 my $conf = $self->{vmconf
};
517 return if $self->{phase2errors
};
519 # move config to remote node
520 my $conffile = PVE
::QemuServer
::config_file
($vmid);
521 my $newconffile = PVE
::QemuServer
::config_file
($vmid, $self->{node
});
523 die "Failed to move config to node '$self->{node}' - rename failed: $!\n"
524 if !rename($conffile, $newconffile);
526 # now that config file is move, we can resume vm on target if livemigrate
527 if ($self->{tunnel
}) {
528 my $cmd = [@{$self->{rem_ssh
}}, 'qm', 'resume', $vmid, '--skiplock'];
529 eval{ PVE
::Tools
::run_command
($cmd, outfunc
=> sub {}, errfunc
=> sub {}) };
531 $self->log('err', $err);
536 # always stop local VM
537 eval { PVE
::QemuServer
::vm_stop
($self->{storecfg
}, $vmid, 1, 1); };
539 $self->log('err', "stopping vm failed - $err");
543 if ($self->{tunnel
}) {
544 eval { finish_tunnel
($self, $self->{tunnel
}); };
546 $self->log('err', $err);
551 # always deactivate volumes - avoid lvm LVs to be active on several nodes
553 my $vollist = PVE
::QemuServer
::get_vm_volumes
($conf);
554 PVE
::Storage
::deactivate_volumes
($self->{storecfg
}, $vollist);
557 $self->log('err', $err);
562 my $cmd = [ @{$self->{rem_ssh
}}, 'qm', 'unlock', $vmid ];
563 $self->cmd_logerr($cmd, errmsg
=> "failed to clear migrate lock");
567 my ($self, $vmid) = @_;