]> git.proxmox.com Git - pve-ha-manager.git/blob - src/PVE/HA/Sim/Hardware.pm
allow to simulate container resources
[pve-ha-manager.git] / src / PVE / HA / Sim / Hardware.pm
1 package PVE::HA::Sim::Hardware;
2
3 # Simulate Hardware resources
4
5 # power supply for nodes: on/off
6 # network connection to nodes: on/off
7 # watchdog devices for nodes
8
9 use strict;
10 use warnings;
11 use POSIX qw(strftime EINTR);
12 use Data::Dumper;
13 use JSON;
14 use IO::File;
15 use Fcntl qw(:DEFAULT :flock);
16 use File::Copy;
17 use File::Path qw(make_path remove_tree);
18 use PVE::HA::Config;
19
20 my $watchdog_timeout = 60;
21
22
23 # Status directory layout
24 #
25 # configuration
26 #
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
33
34 #
35 # runtime status for simulation system
36 #
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
40 #
41 # runtime status
42 #
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 # $testdir/status/groups HA groups configuration
49
50 sub read_lrm_status {
51 my ($self, $node) = @_;
52
53 my $filename = "$self->{statusdir}/lrm_status_$node";
54
55 return PVE::HA::Tools::read_json_from_file($filename, {});
56 }
57
58 sub write_lrm_status {
59 my ($self, $node, $status_obj) = @_;
60
61 my $filename = "$self->{statusdir}/lrm_status_$node";
62
63 PVE::HA::Tools::write_json_to_file($filename, $status_obj);
64 }
65
66 sub read_hardware_status_nolock {
67 my ($self) = @_;
68
69 my $filename = "$self->{statusdir}/hardware_status";
70
71 my $raw = PVE::Tools::file_get_contents($filename);
72 my $cstatus = decode_json($raw);
73
74 return $cstatus;
75 }
76
77 sub write_hardware_status_nolock {
78 my ($self, $cstatus) = @_;
79
80 my $filename = "$self->{statusdir}/hardware_status";
81
82 PVE::Tools::file_set_contents($filename, encode_json($cstatus));
83 };
84
85 sub read_service_config {
86 my ($self) = @_;
87
88 my $filename = "$self->{statusdir}/service_config";
89 my $conf = PVE::HA::Tools::read_json_from_file($filename);
90
91 foreach my $sid (keys %$conf) {
92 my $d = $conf->{$sid};
93
94 die "service '$sid' without assigned node!" if !$d->{node};
95
96 if ($sid =~ m/^(vm|ct):(\d+)$/) {
97 $d->{type} = $1;
98 $d->{name} = $2;
99 } else {
100 die "implement me";
101 }
102 $d->{state} = 'disabled' if !$d->{state};
103 }
104
105 return $conf;
106 }
107
108 sub write_service_config {
109 my ($self, $conf) = @_;
110
111 $self->{service_config} = $conf;
112
113 my $filename = "$self->{statusdir}/service_config";
114 return PVE::HA::Tools::write_json_to_file($filename, $conf);
115 }
116
117 sub set_service_state {
118 my ($self, $sid, $state) = @_;
119
120 my $conf = $self->read_service_config();
121 die "no such service '$sid'" if !$conf->{$sid};
122
123 $conf->{$sid}->{state} = $state;
124
125 $self->write_service_config($conf);
126
127 return $conf;
128 }
129
130 sub add_service {
131 my ($self, $sid, $opts) = @_;
132
133 my $conf = $self->read_service_config();
134 die "resource ID '$sid' already defined\n" if $conf->{$sid};
135
136 $conf->{$sid} = $opts;
137
138 $self->write_service_config($conf);
139
140 return $conf;
141 }
142
143 sub delete_service {
144 my ($self, $sid) = @_;
145
146 my $conf = $self->read_service_config();
147
148 die "no such service '$sid'" if !$conf->{$sid};
149
150 delete $conf->{$sid};
151
152 $self->write_service_config($conf);
153
154 return $conf;
155 }
156
157 sub change_service_location {
158 my ($self, $sid, $current_node, $new_node) = @_;
159
160 my $conf = $self->read_service_config();
161
162 die "no such service '$sid'\n" if !$conf->{$sid};
163
164 die "current_node for '$sid' does not match ($current_node != $conf->{$sid}->{node})\n"
165 if $current_node ne $conf->{$sid}->{node};
166
167 $conf->{$sid}->{node} = $new_node;
168
169 $self->write_service_config($conf);
170 }
171
172 sub queue_crm_commands_nolock {
173 my ($self, $cmd) = @_;
174
175 chomp $cmd;
176
177 my $data = '';
178 my $filename = "$self->{statusdir}/crm_commands";
179 if (-f $filename) {
180 $data = PVE::Tools::file_get_contents($filename);
181 }
182 $data .= "$cmd\n";
183 PVE::Tools::file_set_contents($filename, $data);
184
185 return undef;
186 }
187
188 sub queue_crm_commands {
189 my ($self, $cmd) = @_;
190
191 my $code = sub { $self->queue_crm_commands_nolock($cmd); };
192
193 $self->global_lock($code);
194
195 return undef;
196 }
197
198 sub read_crm_commands {
199 my ($self) = @_;
200
201 my $code = sub {
202 my $data = '';
203
204 my $filename = "$self->{statusdir}/crm_commands";
205 if (-f $filename) {
206 $data = PVE::Tools::file_get_contents($filename);
207 }
208 PVE::Tools::file_set_contents($filename, '');
209
210 return $data;
211 };
212
213 return $self->global_lock($code);
214 }
215
216 sub read_group_config {
217 my ($self) = @_;
218
219 my $filename = "$self->{statusdir}/groups";
220 my $raw = '';
221 $raw = PVE::Tools::file_get_contents($filename) if -f $filename;
222
223 return PVE::HA::Config::parse_groups_config($filename, $raw);
224 }
225
226 sub read_service_status {
227 my ($self, $node) = @_;
228
229 my $filename = "$self->{statusdir}/service_status_$node";
230 return PVE::HA::Tools::read_json_from_file($filename);
231 }
232
233 sub write_service_status {
234 my ($self, $node, $data) = @_;
235
236 my $filename = "$self->{statusdir}/service_status_$node";
237 my $res = PVE::HA::Tools::write_json_to_file($filename, $data);
238
239 # fixme: add test if a service runs on two nodes!!!
240
241 return $res;
242 }
243
244 my $default_group_config = <<__EOD;
245 group: prefer_node1
246 nodes node1
247 nofailback 1
248
249 group: prefer_node2
250 nodes node2
251 nofailback 1
252
253 group: prefer_node3
254 nodes node3
255 nofailback 1
256 __EOD
257
258 sub new {
259 my ($this, $testdir) = @_;
260
261 die "missing testdir" if !$testdir;
262
263 my $class = ref($this) || $this;
264
265 my $self = bless {}, $class;
266
267 my $statusdir = $self->{statusdir} = "$testdir/status";
268
269 remove_tree($statusdir);
270 mkdir $statusdir;
271
272 # copy initial configuartion
273 copy("$testdir/manager_status", "$statusdir/manager_status"); # optional
274
275 if (-f "$testdir/groups") {
276 copy("$testdir/groups", "$statusdir/groups");
277 } else {
278 PVE::Tools::file_set_contents("$statusdir/groups", $default_group_config);
279 }
280
281 if (-f "$testdir/service_config") {
282 copy("$testdir/service_config", "$statusdir/service_config");
283 } else {
284 my $conf = {
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' },
291 };
292 $self->write_service_config($conf);
293 }
294
295 if (-f "$testdir/hardware_status") {
296 copy("$testdir/hardware_status", "$statusdir/hardware_status") ||
297 die "Copy failed: $!\n";
298 } else {
299 my $cstatus = {
300 node1 => { power => 'off', network => 'off' },
301 node2 => { power => 'off', network => 'off' },
302 node3 => { power => 'off', network => 'off' },
303 };
304 $self->write_hardware_status_nolock($cstatus);
305 }
306
307
308 my $cstatus = $self->read_hardware_status_nolock();
309
310 foreach my $node (sort keys %$cstatus) {
311 $self->{nodes}->{$node} = {};
312
313 if (-f "$testdir/service_status_$node") {
314 copy("$testdir/service_status_$node", "$statusdir/service_status_$node");
315 } else {
316 $self->write_service_status($node, {});
317 }
318 }
319
320 $self->{service_config} = $self->read_service_config();
321
322 return $self;
323 }
324
325 sub get_time {
326 my ($self) = @_;
327
328 die "implement in subclass";
329 }
330
331 sub log {
332 my ($self, $level, $msg, $id) = @_;
333
334 chomp $msg;
335
336 my $time = $self->get_time();
337
338 $id = 'hardware' if !$id;
339
340 printf("%-5s %5d %12s: $msg\n", $level, $time, $id);
341 }
342
343 sub statusdir {
344 my ($self, $node) = @_;
345
346 return $self->{statusdir};
347 }
348
349 sub global_lock {
350 my ($self, $code, @param) = @_;
351
352 my $lockfile = "$self->{statusdir}/hardware.lck";
353 my $fh = IO::File->new(">>$lockfile") ||
354 die "unable to open '$lockfile'\n";
355
356 my $success;
357 for (;;) {
358 $success = flock($fh, LOCK_EX);
359 if ($success || ($! != EINTR)) {
360 last;
361 }
362 if (!$success) {
363 close($fh);
364 die "can't acquire lock '$lockfile' - $!\n";
365 }
366 }
367
368 my $res;
369
370 eval { $res = &$code($fh, @param) };
371 my $err = $@;
372
373 close($fh);
374
375 die $err if $err;
376
377 return $res;
378 }
379
380 my $compute_node_info = sub {
381 my ($self, $cstatus) = @_;
382
383 my $node_info = {};
384
385 my $node_count = 0;
386 my $online_count = 0;
387
388 foreach my $node (keys %$cstatus) {
389 my $d = $cstatus->{$node};
390
391 my $online = ($d->{power} eq 'on' && $d->{network} eq 'on') ? 1 : 0;
392 $node_info->{$node}->{online} = $online;
393
394 $node_count++;
395 $online_count++ if $online;
396 }
397
398 my $quorate = ($online_count > int($node_count/2)) ? 1 : 0;
399
400 if (!$quorate) {
401 foreach my $node (keys %$cstatus) {
402 my $d = $cstatus->{$node};
403 $node_info->{$node}->{online} = 0;
404 }
405 }
406
407 return ($node_info, $quorate);
408 };
409
410 sub get_node_info {
411 my ($self) = @_;
412
413 my $cstatus = $self->read_hardware_status_nolock();
414 my ($node_info, $quorate) = &$compute_node_info($self, $cstatus);
415
416 return ($node_info, $quorate);
417 }
418
419 # simulate hardware commands
420 # power <node> <on|off>
421 # network <node> <on|off>
422
423 sub sim_hardware_cmd {
424 my ($self, $cmdstr, $logid) = @_;
425
426 die "implement in subclass";
427 }
428
429 sub run {
430 my ($self) = @_;
431
432 die "implement in subclass";
433 }
434
435 my $modify_watchog = sub {
436 my ($self, $code) = @_;
437
438 my $update_cmd = sub {
439
440 my $filename = "$self->{statusdir}/watchdog_status";
441
442 my ($res, $wdstatus);
443
444 if (-f $filename) {
445 my $raw = PVE::Tools::file_get_contents($filename);
446 $wdstatus = decode_json($raw);
447 } else {
448 $wdstatus = {};
449 }
450
451 ($wdstatus, $res) = &$code($wdstatus);
452
453 PVE::Tools::file_set_contents($filename, encode_json($wdstatus));
454
455 return $res;
456 };
457
458 return $self->global_lock($update_cmd);
459 };
460
461 sub watchdog_reset_nolock {
462 my ($self, $node) = @_;
463
464 my $filename = "$self->{statusdir}/watchdog_status";
465
466 if (-f $filename) {
467 my $raw = PVE::Tools::file_get_contents($filename);
468 my $wdstatus = decode_json($raw);
469
470 foreach my $id (keys %$wdstatus) {
471 delete $wdstatus->{$id} if $wdstatus->{$id}->{node} eq $node;
472 }
473
474 PVE::Tools::file_set_contents($filename, encode_json($wdstatus));
475 }
476 }
477
478 sub watchdog_check {
479 my ($self, $node) = @_;
480
481 my $code = sub {
482 my ($wdstatus) = @_;
483
484 my $res = 1;
485
486 foreach my $wfh (keys %$wdstatus) {
487 my $wd = $wdstatus->{$wfh};
488 next if $wd->{node} ne $node;
489
490 my $ctime = $self->get_time();
491 my $tdiff = $ctime - $wd->{update_time};
492
493 if ($tdiff > $watchdog_timeout) { # expired
494 $res = 0;
495 delete $wdstatus->{$wfh};
496 }
497 }
498
499 return ($wdstatus, $res);
500 };
501
502 return &$modify_watchog($self, $code);
503 }
504
505 my $wdcounter = 0;
506
507 sub watchdog_open {
508 my ($self, $node) = @_;
509
510 my $code = sub {
511 my ($wdstatus) = @_;
512
513 ++$wdcounter;
514
515 my $id = "WD:$node:$$:$wdcounter";
516
517 die "internal error" if defined($wdstatus->{$id});
518
519 $wdstatus->{$id} = {
520 node => $node,
521 update_time => $self->get_time(),
522 };
523
524 return ($wdstatus, $id);
525 };
526
527 return &$modify_watchog($self, $code);
528 }
529
530 sub watchdog_close {
531 my ($self, $wfh) = @_;
532
533 my $code = sub {
534 my ($wdstatus) = @_;
535
536 my $wd = $wdstatus->{$wfh};
537 die "no such watchdog handle '$wfh'\n" if !defined($wd);
538
539 my $tdiff = $self->get_time() - $wd->{update_time};
540 die "watchdog expired" if $tdiff > $watchdog_timeout;
541
542 delete $wdstatus->{$wfh};
543
544 return ($wdstatus);
545 };
546
547 return &$modify_watchog($self, $code);
548 }
549
550 sub watchdog_update {
551 my ($self, $wfh) = @_;
552
553 my $code = sub {
554 my ($wdstatus) = @_;
555
556 my $wd = $wdstatus->{$wfh};
557
558 die "no such watchdog handle '$wfh'\n" if !defined($wd);
559
560 my $ctime = $self->get_time();
561 my $tdiff = $ctime - $wd->{update_time};
562
563 die "watchdog expired" if $tdiff > $watchdog_timeout;
564
565 $wd->{update_time} = $ctime;
566
567 return ($wdstatus);
568 };
569
570 return &$modify_watchog($self, $code);
571 }
572
573 1;