1 package PVE
::HA
::Sim
::Hardware
;
3 # Simulate Hardware resources
5 # power supply for nodes: on/off
6 # network connection to nodes: on/off
7 # watchdog devices for nodes
11 use POSIX
qw(strftime EINTR);
15 use Fcntl
qw(:DEFAULT :flock);
17 use File
::Path
qw(make_path remove_tree);
20 my $watchdog_timeout = 60;
23 # Status directory layout
27 # $testdir/cmdlist Command list for simulation
28 # $testdir/hardware_status Hardware description (number of nodes, ...)
29 # $testdir/manager_status CRM status (start with {})
30 # $testdir/service_config Service configuration
31 # $testdir/groups HA groups configuration
32 # $testdir/service_status_<node> Service status
35 # runtime status for simulation system
37 # $testdir/status/cluster_locks Cluster locks
38 # $testdir/status/hardware_status Hardware status (power/network on/off)
39 # $testdir/status/watchdog_status Watchdog status
43 # $testdir/status/lrm_status_<node> LRM status
44 # $testdir/status/manager_status CRM status
45 # $testdir/status/crm_commands CRM command queue
46 # $testdir/status/service_config Service configuration
47 # $testdir/status/service_status_<node> Service status
48 # $testdir/status/groups HA groups configuration
51 my ($self, $node) = @_;
53 my $filename = "$self->{statusdir}/lrm_status_$node";
55 return PVE
::HA
::Tools
::read_json_from_file
($filename, {});
58 sub write_lrm_status
{
59 my ($self, $node, $status_obj) = @_;
61 my $filename = "$self->{statusdir}/lrm_status_$node";
63 PVE
::HA
::Tools
::write_json_to_file
($filename, $status_obj);
66 sub read_hardware_status_nolock
{
69 my $filename = "$self->{statusdir}/hardware_status";
71 my $raw = PVE
::Tools
::file_get_contents
($filename);
72 my $cstatus = decode_json
($raw);
77 sub write_hardware_status_nolock
{
78 my ($self, $cstatus) = @_;
80 my $filename = "$self->{statusdir}/hardware_status";
82 PVE
::Tools
::file_set_contents
($filename, encode_json
($cstatus));
85 sub read_service_config
{
88 my $filename = "$self->{statusdir}/service_config";
89 my $conf = PVE
::HA
::Tools
::read_json_from_file
($filename);
91 foreach my $sid (keys %$conf) {
92 my $d = $conf->{$sid};
94 die "service '$sid' without assigned node!" if !$d->{node
};
96 if ($sid =~ m/^vm:(\d+)$/) {
102 $d->{state} = 'disabled' if !$d->{state};
108 sub write_service_config
{
109 my ($self, $conf) = @_;
111 $self->{service_config
} = $conf;
113 my $filename = "$self->{statusdir}/service_config";
114 return PVE
::HA
::Tools
::write_json_to_file
($filename, $conf);
117 sub set_service_state
{
118 my ($self, $sid, $state) = @_;
120 my $conf = $self->read_service_config();
121 die "no such service '$sid'" if !$conf->{$sid};
123 $conf->{$sid}->{state} = $state;
125 $self->write_service_config($conf);
131 my ($self, $sid, $opts) = @_;
133 my $conf = $self->read_service_config();
134 die "resource ID '$sid' already defined\n" if $conf->{$sid};
136 $conf->{$sid} = $opts;
138 $self->write_service_config($conf);
144 my ($self, $sid) = @_;
146 my $conf = $self->read_service_config();
148 die "no such service '$sid'" if !$conf->{$sid};
150 delete $conf->{$sid};
152 $self->write_service_config($conf);
157 sub change_service_location
{
158 my ($self, $sid, $current_node, $new_node) = @_;
160 my $conf = $self->read_service_config();
162 die "no such service '$sid'\n" if !$conf->{$sid};
164 die "current_node for '$sid' does not match ($current_node != $conf->{$sid}->{node})\n"
165 if $current_node ne $conf->{$sid}->{node
};
167 $conf->{$sid}->{node
} = $new_node;
169 $self->write_service_config($conf);
172 sub queue_crm_commands_nolock
{
173 my ($self, $cmd) = @_;
178 my $filename = "$self->{statusdir}/crm_commands";
180 $data = PVE
::Tools
::file_get_contents
($filename);
183 PVE
::Tools
::file_set_contents
($filename, $data);
188 sub queue_crm_commands
{
189 my ($self, $cmd) = @_;
191 my $code = sub { $self->queue_crm_commands_nolock($cmd); };
193 $self->global_lock($code);
198 sub read_crm_commands
{
204 my $filename = "$self->{statusdir}/crm_commands";
206 $data = PVE
::Tools
::file_get_contents
($filename);
208 PVE
::Tools
::file_set_contents
($filename, '');
213 return $self->global_lock($code);
216 sub read_group_config
{
219 my $filename = "$self->{statusdir}/groups";
221 $raw = PVE
::Tools
::file_get_contents
($filename) if -f
$filename;
223 return PVE
::HA
::Config
::parse_groups_config
($filename, $raw);
226 sub read_service_status
{
227 my ($self, $node) = @_;
229 my $filename = "$self->{statusdir}/service_status_$node";
230 return PVE
::HA
::Tools
::read_json_from_file
($filename);
233 sub write_service_status
{
234 my ($self, $node, $data) = @_;
236 my $filename = "$self->{statusdir}/service_status_$node";
237 my $res = PVE
::HA
::Tools
::write_json_to_file
($filename, $data);
239 # fixme: add test if a service runs on two nodes!!!
244 my $default_group_config = <<__EOD;
259 my ($this, $testdir) = @_;
261 die "missing testdir" if !$testdir;
263 my $class = ref($this) || $this;
265 my $self = bless {}, $class;
267 my $statusdir = $self->{statusdir
} = "$testdir/status";
269 remove_tree
($statusdir);
272 # copy initial configuartion
273 copy
("$testdir/manager_status", "$statusdir/manager_status"); # optional
275 if (-f
"$testdir/groups") {
276 copy
("$testdir/groups", "$statusdir/groups");
278 PVE
::Tools
::file_set_contents
("$statusdir/groups", $default_group_config);
281 if (-f
"$testdir/service_config") {
282 copy
("$testdir/service_config", "$statusdir/service_config");
285 'vm:101' => { node
=> 'node1', group
=> 'prefer_node1' },
286 'vm:102' => { node
=> 'node2', group
=> 'prefer_node2' },
287 'vm:103' => { node
=> 'node3', group
=> 'prefer_node3' },
288 'vm:104' => { node
=> 'node1', group
=> 'prefer_node1' },
289 'vm:105' => { node
=> 'node2', group
=> 'prefer_node2' },
290 'vm:106' => { node
=> 'node3', group
=> 'prefer_node3' },
292 $self->write_service_config($conf);
295 if (-f
"$testdir/hardware_status") {
296 copy
("$testdir/hardware_status", "$statusdir/hardware_status") ||
297 die "Copy failed: $!\n";
300 node1
=> { power
=> 'off', network
=> 'off' },
301 node2
=> { power
=> 'off', network
=> 'off' },
302 node3
=> { power
=> 'off', network
=> 'off' },
304 $self->write_hardware_status_nolock($cstatus);
308 my $cstatus = $self->read_hardware_status_nolock();
310 foreach my $node (sort keys %$cstatus) {
311 $self->{nodes
}->{$node} = {};
313 if (-f
"$testdir/service_status_$node") {
314 copy
("$testdir/service_status_$node", "$statusdir/service_status_$node");
316 $self->write_service_status($node, {});
320 $self->{service_config
} = $self->read_service_config();
328 die "implement in subclass";
332 my ($self, $level, $msg, $id) = @_;
336 my $time = $self->get_time();
338 $id = 'hardware' if !$id;
340 printf("%-5s %5d %12s: $msg\n", $level, $time, $id);
344 my ($self, $node) = @_;
346 return $self->{statusdir
};
350 my ($self, $code, @param) = @_;
352 my $lockfile = "$self->{statusdir}/hardware.lck";
353 my $fh = IO
::File-
>new(">>$lockfile") ||
354 die "unable to open '$lockfile'\n";
358 $success = flock($fh, LOCK_EX
);
359 if ($success || ($! != EINTR
)) {
364 die "can't acquire lock '$lockfile' - $!\n";
370 eval { $res = &$code($fh, @param) };
380 my $compute_node_info = sub {
381 my ($self, $cstatus) = @_;
386 my $online_count = 0;
388 foreach my $node (keys %$cstatus) {
389 my $d = $cstatus->{$node};
391 my $online = ($d->{power
} eq 'on' && $d->{network
} eq 'on') ?
1 : 0;
392 $node_info->{$node}->{online
} = $online;
395 $online_count++ if $online;
398 my $quorate = ($online_count > int($node_count/2)) ?
1 : 0;
401 foreach my $node (keys %$cstatus) {
402 my $d = $cstatus->{$node};
403 $node_info->{$node}->{online
} = 0;
407 return ($node_info, $quorate);
413 my $cstatus = $self->read_hardware_status_nolock();
414 my ($node_info, $quorate) = &$compute_node_info($self, $cstatus);
416 return ($node_info, $quorate);
419 # simulate hardware commands
420 # power <node> <on|off>
421 # network <node> <on|off>
423 sub sim_hardware_cmd
{
424 my ($self, $cmdstr, $logid) = @_;
426 die "implement in subclass";
432 die "implement in subclass";
435 my $modify_watchog = sub {
436 my ($self, $code) = @_;
438 my $update_cmd = sub {
440 my $filename = "$self->{statusdir}/watchdog_status";
442 my ($res, $wdstatus);
445 my $raw = PVE
::Tools
::file_get_contents
($filename);
446 $wdstatus = decode_json
($raw);
451 ($wdstatus, $res) = &$code($wdstatus);
453 PVE
::Tools
::file_set_contents
($filename, encode_json
($wdstatus));
458 return $self->global_lock($update_cmd);
461 sub watchdog_reset_nolock
{
462 my ($self, $node) = @_;
464 my $filename = "$self->{statusdir}/watchdog_status";
467 my $raw = PVE
::Tools
::file_get_contents
($filename);
468 my $wdstatus = decode_json
($raw);
470 foreach my $id (keys %$wdstatus) {
471 delete $wdstatus->{$id} if $wdstatus->{$id}->{node
} eq $node;
474 PVE
::Tools
::file_set_contents
($filename, encode_json
($wdstatus));
479 my ($self, $node) = @_;
486 foreach my $wfh (keys %$wdstatus) {
487 my $wd = $wdstatus->{$wfh};
488 next if $wd->{node
} ne $node;
490 my $ctime = $self->get_time();
491 my $tdiff = $ctime - $wd->{update_time
};
493 if ($tdiff > $watchdog_timeout) { # expired
495 delete $wdstatus->{$wfh};
499 return ($wdstatus, $res);
502 return &$modify_watchog($self, $code);
508 my ($self, $node) = @_;
515 my $id = "WD:$node:$$:$wdcounter";
517 die "internal error" if defined($wdstatus->{$id});
521 update_time
=> $self->get_time(),
524 return ($wdstatus, $id);
527 return &$modify_watchog($self, $code);
531 my ($self, $wfh) = @_;
536 my $wd = $wdstatus->{$wfh};
537 die "no such watchdog handle '$wfh'\n" if !defined($wd);
539 my $tdiff = $self->get_time() - $wd->{update_time
};
540 die "watchdog expired" if $tdiff > $watchdog_timeout;
542 delete $wdstatus->{$wfh};
547 return &$modify_watchog($self, $code);
550 sub watchdog_update
{
551 my ($self, $wfh) = @_;
556 my $wd = $wdstatus->{$wfh};
558 die "no such watchdog handle '$wfh'\n" if !defined($wd);
560 my $ctime = $self->get_time();
561 my $tdiff = $ctime - $wd->{update_time
};
563 die "watchdog expired" if $tdiff > $watchdog_timeout;
565 $wd->{update_time
} = $ctime;
570 return &$modify_watchog($self, $code);