]>
git.proxmox.com Git - pve-container.git/blob - src/PVE/LXC/Migrate.pm
82305c001765a4588f96f59c3efde30f6ea74b1d
1 package PVE
::LXC
::Migrate
;
7 use File
::Copy
; # fixme: remove
12 use PVE
::ReplicationConfig
;
13 use PVE
::ReplicationState
;
20 use PVE
::AbstractMigrate
;
21 use base
qw(PVE::AbstractMigrate);
23 # compared against remote end's minimum version
24 our $WS_TUNNEL_VERSION = 2;
27 my ($self, $vmid, $code, @param) = @_;
29 return PVE
::LXC
::Config-
>lock_config($vmid, $code, @param);
33 my ($self, $vmid) = @_;
35 my $online = $self->{opts
}->{online
};
36 my $restart= $self->{opts
}->{restart
};
37 my $remote = $self->{opts
}->{remote
};
39 $self->{storecfg
} = PVE
::Storage
::config
();
42 my $conf = $self->{vmconf
} = PVE
::LXC
::Config-
>load_config($vmid);
44 PVE
::LXC
::Config-
>check_lock($conf);
47 if (PVE
::LXC
::check_running
($vmid)) {
48 die "lxc live migration is currently not implemented\n" if $online;
49 die "running container can only be migrated in restart mode" if !$restart;
52 $self->{was_running
} = $running;
55 PVE
::LXC
::Config-
>foreach_volume_full($conf, { include_unused
=> 1 }, sub {
56 my ($ms, $mountpoint) = @_;
58 my $volid = $mountpoint->{volume
};
59 my $type = $mountpoint->{type
};
61 # skip dev/bind mps when shared
62 if ($type ne 'volume') {
63 if ($mountpoint->{shared
}) {
66 die "cannot migrate local $type mount point '$ms'\n";
70 my ($storage, $volname) = PVE
::Storage
::parse_volume_id
($volid, 1) if $volid;
71 die "can't determine assigned storage for mount point '$ms'\n" if !$storage;
73 # check if storage is available on both nodes
74 my $scfg = PVE
::Storage
::storage_check_enabled
($self->{storecfg
}, $storage);
76 my $targetsid = $storage;
78 die "content type 'rootdir' is not available on storage '$storage'\n"
79 if !$scfg->{content
}->{rootdir
};
81 if ($scfg->{shared
} && !$remote) {
82 # PVE::Storage::activate_storage checks this for non-shared storages
83 my $plugin = PVE
::Storage
::Plugin-
>lookup($scfg->{type
});
84 warn "Used shared storage '$storage' is not online on source node!\n"
85 if !$plugin->check_connection($storage, $scfg);
87 # unless in restart mode because we shut the container down
88 die "unable to migrate local mount point '$volid' while CT is running"
89 if $running && !$restart;
91 $targetsid = PVE
::JSONSchema
::map_id
($self->{opts
}->{storagemap
}, $storage);
95 my $target_scfg = PVE
::Storage
::storage_check_enabled
($self->{storecfg
}, $targetsid, $self->{node
});
97 die "$volid: content type 'rootdir' is not available on storage '$targetsid'\n"
98 if !$target_scfg->{content
}->{rootdir
};
101 $storages->{$targetsid} = 1;
104 # todo: test if VM uses local resources
107 # test & establish websocket connection
108 my $bridges = map_bridges
($conf, $self->{opts
}->{bridgemap
}, 1);
110 my $remote = $self->{opts
}->{remote
};
111 my $conn = $remote->{conn
};
114 my ($level, $msg) = @_;
115 $self->log($level, $msg);
118 my $websocket_url = "https://$conn->{host}:$conn->{port}/api2/json/nodes/$self->{node}/lxc/$remote->{vmid}/mtunnelwebsocket";
119 my $url = "/nodes/$self->{node}/lxc/$remote->{vmid}/mtunnel";
121 my $tunnel_params = {
122 url
=> $websocket_url,
125 my $storage_list = join(',', keys %$storages);
126 my $bridge_list = join(',', keys %$bridges);
129 storages
=> $storage_list,
130 bridges
=> $bridge_list,
133 my $tunnel = PVE
::Tunnel
::fork_websocket_tunnel
($conn, $url, $req_params, $tunnel_params, $log);
134 my $min_version = $tunnel->{version
} - $tunnel->{age
};
135 $self->log('info', "local WS tunnel version: $WS_TUNNEL_VERSION");
136 $self->log('info', "remote WS tunnel version: $tunnel->{version}");
137 $self->log('info', "minimum required WS tunnel version: $min_version");
138 die "Remote tunnel endpoint not compatible, upgrade required\n"
139 if $WS_TUNNEL_VERSION < $min_version;
140 die "Remote tunnel endpoint too old, upgrade required\n"
141 if $WS_TUNNEL_VERSION > $tunnel->{version
};
143 $self->log('info', "websocket tunnel started\n");
144 $self->{tunnel
} = $tunnel;
146 # test ssh connection
147 my $cmd = [ @{$self->{rem_ssh
}}, '/bin/true' ];
148 eval { $self->cmd_quiet($cmd); };
149 die "Can't connect to destination address using public key\n" if $@;
152 # in restart mode, we shutdown the container before migrating
153 if ($restart && $running) {
154 my $timeout = $self->{opts
}->{timeout
} // 180;
156 $self->log('info', "shutdown CT $vmid\n");
158 PVE
::LXC
::vm_stop
($vmid, 0, $timeout);
167 my ($self, $vmid) = @_;
169 my $remote = $self->{opts
}->{remote
};
171 $self->log('info', "starting migration of CT $self->{vmid} to node '$self->{node}' ($self->{nodeip})");
173 my $conf = $self->{vmconf
};
174 $conf->{lock} = 'migrate';
175 PVE
::LXC
::Config-
>write_config($vmid, $conf);
177 if ($self->{running
}) {
178 $self->log('info', "container is running - using online migration");
181 $self->{volumes
} = []; # list of already migrated volumes
182 my $volhash = {}; # 'config', 'snapshot' or 'storage' for local volumes
183 my $volhash_errors = {};
186 my $log_error = sub {
187 my ($msg, $volid) = @_;
189 $volhash_errors->{$volid} = $msg if !defined($volhash_errors->{$volid});
193 my $test_volid = sub {
194 my ($volid, $snapname) = @_;
198 my ($sid, $volname) = PVE
::Storage
::parse_volume_id
($volid);
200 # check if storage is available on source node
201 my $scfg = PVE
::Storage
::storage_check_enabled
($self->{storecfg
}, $sid);
203 my $targetsid = $sid;
205 if ($scfg->{shared
} && !$remote) {
206 $self->log('info', "volume '$volid' is on shared storage '$sid'")
210 $targetsid = PVE
::JSONSchema
::map_id
($self->{opts
}->{storagemap
}, $sid);
213 PVE
::Storage
::storage_check_enabled
($self->{storecfg
}, $targetsid, $self->{node
})
216 my $bwlimit = $self->get_bwlimit($sid, $targetsid);
218 $volhash->{$volid}->{ref} = defined($snapname) ?
'snapshot' : 'config';
219 $volhash->{$volid}->{snapshots
} = 1 if defined($snapname);
220 $volhash->{$volid}->{targetsid
} = $targetsid;
221 $volhash->{$volid}->{bwlimit
} = $bwlimit;
223 my ($path, $owner) = PVE
::Storage
::path
($self->{storecfg
}, $volid);
225 die "owned by other guest (owner = $owner)\n"
226 if !$owner || ($owner != $self->{vmid
});
228 if (defined($snapname)) {
229 # we cannot migrate shapshots on local storage
230 # exceptions: 'zfspool', 'btrfs'
231 if ($scfg->{type
} eq 'zfspool' || $scfg->{type
} eq 'btrfs') {
234 die "non-migratable snapshot exists\n";
239 my ($ms, $mountpoint, $snapname) = @_;
241 my $volid = $mountpoint->{volume
};
242 # already checked in prepare
243 if ($mountpoint->{type
} ne 'volume') {
244 $self->log('info', "ignoring shared '$mountpoint->{type}' mount point '$ms' ('$volid')")
250 &$test_volid($volid, $snapname);
252 die "remote migration with snapshots not supported yet\n"
253 if $remote && $snapname;
256 &$log_error($@, $volid) if $@;
259 # first unused / lost volumes owned by this container
260 my @sids = PVE
::Storage
::storage_ids
($self->{storecfg
});
261 foreach my $storeid (@sids) {
262 my $scfg = PVE
::Storage
::storage_config
($self->{storecfg
}, $storeid);
263 next if $scfg->{shared
} && !$remote;
264 next if !PVE
::Storage
::storage_check_enabled
($self->{storecfg
}, $storeid, undef, 1);
266 # get list from PVE::Storage (for unreferenced volumes)
267 my $dl = PVE
::Storage
::vdisk_list
($self->{storecfg
}, $storeid, $vmid, undef, 'rootdir');
269 next if @{$dl->{$storeid}} == 0;
271 # check if storage is available on target node
272 my $targetsid = PVE
::JSONSchema
::map_id
($self->{opts
}->{storagemap
}, $storeid);
274 my $target_scfg = PVE
::Storage
::storage_check_enabled
($self->{storecfg
}, $targetsid, $self->{node
});
276 die "content type 'rootdir' is not available on storage '$targetsid'\n"
277 if !$target_scfg->{content
}->{rootdir
};
280 PVE
::Storage
::foreach_volid
($dl, sub {
281 my ($volid, $sid, $volname) = @_;
283 $volhash->{$volid}->{ref} = 'storage';
284 $volhash->{$volid}->{targetsid
} = $targetsid;
288 # then all volumes referenced in snapshots
289 foreach my $snapname (keys %{$conf->{snapshots
}}) {
290 &$test_volid($conf->{snapshots
}->{$snapname}->{'vmstate'}, 0, undef)
291 if defined($conf->{snapshots
}->{$snapname}->{'vmstate'});
292 PVE
::LXC
::Config-
>foreach_volume($conf->{snapshots
}->{$snapname}, $test_mp, $snapname);
295 # finally all current volumes
296 PVE
::LXC
::Config-
>foreach_volume_full($conf, { include_unused
=> 1 }, $test_mp);
298 # additional checks for local storage
299 foreach my $volid (keys %$volhash) {
301 my ($sid, $volname) = PVE
::Storage
::parse_volume_id
($volid);
302 my $scfg = PVE
::Storage
::storage_config
($self->{storecfg
}, $sid);
304 # TODO move to storage plugin layer?
305 my $migratable_storages = [
313 push @$migratable_storages, 'cifs';
314 push @$migratable_storages, 'nfs';
317 die "storage type '$scfg->{type}' not supported\n"
318 if !grep { $_ eq $scfg->{type
} } @$migratable_storages;
320 # image is a linked clone on local storage, se we can't migrate.
321 if (my $basename = (PVE
::Storage
::parse_volname
($self->{storecfg
}, $volid))[3]) {
322 die "clone of '$basename'";
325 &$log_error($@, $volid) if $@;
328 foreach my $volid (sort keys %$volhash) {
329 my $ref = $volhash->{$volid}->{ref};
330 if ($ref eq 'storage') {
331 $self->log('info', "found local volume '$volid' (via storage)\n");
332 } elsif ($ref eq 'config') {
333 $self->log('info', "found local volume '$volid' (in current VM config)\n");
334 } elsif ($ref eq 'snapshot') {
335 $self->log('info', "found local volume '$volid' (referenced by snapshot(s))\n");
337 $self->log('info', "found local volume '$volid'\n");
341 foreach my $volid (sort keys %$volhash_errors) {
342 $self->log('warn', "can't migrate local volume '$volid': $volhash_errors->{$volid}");
346 die "can't migrate CT - check log\n";
351 my $rep_cfg = PVE
::ReplicationConfig-
>new();
354 die "cannot remote-migrate replicated VM\n"
355 if $rep_cfg->check_for_existing_jobs($vmid, 1);
356 } elsif (my $jobcfg = $rep_cfg->find_local_replication_job($vmid, $self->{node
})) {
357 die "can't live migrate VM with replicated volumes\n" if $self->{running
};
358 my $start_time = time();
359 my $logfunc = sub { my ($msg) = @_; $self->log('info', $msg); };
360 $rep_volumes = PVE
::Replication
::run_replication
(
361 'PVE::LXC::Config', $jobcfg, $start_time, $start_time, $logfunc);
364 my $opts = $self->{opts
};
365 foreach my $volid (keys %$volhash) {
366 next if $rep_volumes->{$volid};
367 push @{$self->{volumes
}}, $volid;
369 # JSONSchema and get_bandwidth_limit use kbps - storage_migrate bps
370 my $bwlimit = $volhash->{$volid}->{bwlimit
};
371 $bwlimit = $bwlimit * 1024 if defined($bwlimit);
373 my $targetsid = $volhash->{$volid}->{targetsid
};
375 my $new_volid = eval {
378 my ($level, $msg) = @_;
379 $self->log($level, $msg);
382 return PVE
::StorageTunnel
::storage_migrate
(
392 my $storage_migrate_opts = {
393 'ratelimit_bps' => $bwlimit,
394 'insecure' => $opts->{migration_type
} eq 'insecure',
395 'with_snapshots' => $volhash->{$volid}->{snapshots
},
399 my $logfunc = sub { $self->log('info', $_[0]); };
400 return PVE
::Storage
::storage_migrate
(
405 $storage_migrate_opts,
412 die "storage migration for '$volid' to storage '$targetsid' failed - $err\n";
415 $self->{volume_map
}->{$volid} = $new_volid;
416 $self->log('info', "volume '$volid' is '$new_volid' on the target\n");
418 eval { PVE
::Storage
::deactivate_volumes
($self->{storecfg
}, [$volid]); };
420 $self->log('warn', $err);
424 if ($self->{running
}) {
428 # make sure everything on (shared) storage is unmounted
429 # Note: we must be 100% sure, else we get data corruption because
430 # non-shared file system could be mounted twice (on shared storage)
432 PVE
::LXC
::umount_all
($vmid, $self->{storecfg
}, $conf);
434 #to be sure there are no active volumes
435 my $vollist = PVE
::LXC
::Config-
>get_vm_volumes($conf);
436 PVE
::Storage
::deactivate_volumes
($self->{storecfg
}, $vollist);
439 my $remote_conf = PVE
::LXC
::Config-
>load_config($vmid);
440 PVE
::LXC
::Config-
>update_volume_ids($remote_conf, $self->{volume_map
});
442 my $bridges = map_bridges
($remote_conf, $self->{opts
}->{bridgemap
});
443 for my $target (keys $bridges->%*) {
444 for my $nic (keys $bridges->{$target}->%*) {
445 $self->log('info', "mapped: $nic from $bridges->{$target}->{$nic} to $target");
448 my $conf_str = PVE
::LXC
::Config
::write_pct_config
("remote", $remote_conf);
450 # TODO expose in PVE::Firewall?
451 my $vm_fw_conf_path = "/etc/pve/firewall/$vmid.fw";
453 $fw_conf_str = PVE
::Tools
::file_get_contents
($vm_fw_conf_path)
454 if -e
$vm_fw_conf_path;
457 'firewall-config' => $fw_conf_str,
460 PVE
::Tunnel
::write_tunnel
($self->{tunnel
}, 10, 'config', $params);
462 # transfer replication state before moving config
463 $self->transfer_replication_state() if $rep_volumes;
464 PVE
::LXC
::Config-
>update_volume_ids($conf, $self->{volume_map
});
465 PVE
::LXC
::Config-
>write_config($vmid, $conf);
466 PVE
::LXC
::Config-
>move_config_to_node($vmid, $self->{node
});
467 $self->switch_replication_job_target() if $rep_volumes;
469 $self->{conf_migrated
} = 1;
473 my ($self, $vmid, $err) = @_;
475 $self->log('info', "aborting phase 1 - cleanup resources");
477 if ($self->{volumes
}) {
478 foreach my $volid (@{$self->{volumes
}}) {
479 $self->log('err', "found stale volume copy '$volid' on node '$self->{node}'");
480 # fixme: try to remove ?
484 if ($self->{opts
}->{remote
}) {
485 # cleans up remote volumes
486 PVE
::Tunnel
::finish_tunnel
($self->{tunnel
}, 1);
487 delete $self->{tunnel
};
492 my ($self, $vmid) = @_;
494 my $volids = $self->{volumes
};
496 # handled below in final_cleanup
497 return if $self->{opts
}->{remote
};
499 # destroy local copies
500 foreach my $volid (@$volids) {
501 eval { PVE
::Storage
::vdisk_free
($self->{storecfg
}, $volid); };
503 $self->log('err', "removing local copy of '$volid' failed - $err");
505 last if $err =~ /^interrupted by signal$/;
511 my ($self, $vmid) = @_;
513 $self->log('info', "start final cleanup");
515 if (!$self->{conf_migrated
}) {
516 eval { PVE
::LXC
::Config-
>remove_lock($vmid, 'migrate'); };
518 $self->log('err', $err);
520 # in restart mode, we start the container on the source node on migration error
521 if ($self->{opts
}->{restart
} && $self->{was_running
}) {
522 $self->log('info', "start container on source node");
524 PVE
::LXC
::vm_start
($vmid, $self->{vmconf
}, $skiplock);
526 } elsif ($self->{opts
}->{remote
}) {
527 eval { PVE
::Tunnel
::write_tunnel
($self->{tunnel
}, 10, 'unlock') };
528 $self->log('err', "Failed to clear migrate lock - $@\n") if $@;
530 if ($self->{opts
}->{restart
} && $self->{was_running
}) {
531 $self->log('info', "start container on target node");
532 PVE
::Tunnel
::write_tunnel
($self->{tunnel
}, 60, 'start');
534 if ($self->{opts
}->{delete}) {
535 PVE
::LXC
::destroy_lxc_container
(
536 PVE
::Storage
::config
(),
538 PVE
::LXC
::Config-
>load_config($vmid),
543 PVE
::Tunnel
::finish_tunnel
($self->{tunnel
});
545 my $cmd = [ @{$self->{rem_ssh
}}, 'pct', 'unlock', $vmid ];
546 $self->cmd_logerr($cmd, errmsg
=> "failed to clear migrate lock");
548 # in restart mode, we start the container on the target node after migration
549 if ($self->{opts
}->{restart
} && $self->{was_running
}) {
550 $self->log('info', "start container on target node");
551 my $cmd = [ @{$self->{rem_ssh
}}, 'pct', 'start', $vmid];
558 my ($conf, $map, $scan_only) = @_;
562 foreach my $opt (keys %$conf) {
563 next if $opt !~ m/^net\d+$/;
565 next if !$conf->{$opt};
566 my $d = PVE
::LXC
::Config-
>parse_lxc_network($conf->{$opt});
567 next if !$d || !$d->{bridge
};
569 my $target_bridge = PVE
::JSONSchema
::map_id
($map, $d->{bridge
});
570 $bridges->{$target_bridge}->{$opt} = $d->{bridge
};
574 $d->{bridge
} = $target_bridge;
575 $conf->{$opt} = PVE
::LXC
::Config-
>print_lxc_network($d);