]>
Commit | Line | Data |
---|---|---|
8b3f9144 DM |
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 | |
0cfd8f5b DM |
8 | |
9 | use strict; | |
10 | use warnings; | |
11 | use POSIX qw(strftime EINTR); | |
289e4784 | 12 | use JSON; |
0cfd8f5b DM |
13 | use IO::File; |
14 | use Fcntl qw(:DEFAULT :flock); | |
787b66eb DM |
15 | use File::Copy; |
16 | use File::Path qw(make_path remove_tree); | |
c982dfee | 17 | use PVE::HA::FenceConfig; |
7d33cb12 | 18 | use PVE::HA::Groups; |
f5a14b93 | 19 | |
17b5cf98 | 20 | my $watchdog_timeout = 60; |
0bba8f60 | 21 | |
0cfd8f5b | 22 | |
787b66eb DM |
23 | # Status directory layout |
24 | # | |
25 | # configuration | |
26 | # | |
8456bde2 DM |
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 | |
abc920b4 | 31 | # $testdir/groups HA groups configuration |
8456bde2 | 32 | # $testdir/service_status_<node> Service status |
ed408b44 | 33 | # $testdir/datacenter.cfg Datacenter wide HA configuration |
3c36cbca | 34 | |
9329c1e2 DM |
35 | # |
36 | # runtime status for simulation system | |
37 | # | |
38 | # $testdir/status/cluster_locks Cluster locks | |
39 | # $testdir/status/hardware_status Hardware status (power/network on/off) | |
40 | # $testdir/status/watchdog_status Watchdog status | |
787b66eb DM |
41 | # |
42 | # runtime status | |
9329c1e2 | 43 | # |
8456bde2 DM |
44 | # $testdir/status/lrm_status_<node> LRM status |
45 | # $testdir/status/manager_status CRM status | |
abc920b4 | 46 | # $testdir/status/crm_commands CRM command queue |
8456bde2 DM |
47 | # $testdir/status/service_config Service configuration |
48 | # $testdir/status/service_status_<node> Service status | |
abc920b4 | 49 | # $testdir/status/groups HA groups configuration |
c4a221bc DM |
50 | |
51 | sub read_lrm_status { | |
52 | my ($self, $node) = @_; | |
53 | ||
54 | my $filename = "$self->{statusdir}/lrm_status_$node"; | |
55 | ||
289e4784 | 56 | return PVE::HA::Tools::read_json_from_file($filename, {}); |
c4a221bc DM |
57 | } |
58 | ||
59 | sub write_lrm_status { | |
60 | my ($self, $node, $status_obj) = @_; | |
61 | ||
62 | my $filename = "$self->{statusdir}/lrm_status_$node"; | |
63 | ||
289e4784 | 64 | PVE::HA::Tools::write_json_to_file($filename, $status_obj); |
c4a221bc | 65 | } |
787b66eb | 66 | |
8b3f9144 | 67 | sub read_hardware_status_nolock { |
0cfd8f5b DM |
68 | my ($self) = @_; |
69 | ||
8b3f9144 | 70 | my $filename = "$self->{statusdir}/hardware_status"; |
0cfd8f5b DM |
71 | |
72 | my $raw = PVE::Tools::file_get_contents($filename); | |
73 | my $cstatus = decode_json($raw); | |
74 | ||
75 | return $cstatus; | |
76 | } | |
77 | ||
8b3f9144 | 78 | sub write_hardware_status_nolock { |
0cfd8f5b DM |
79 | my ($self, $cstatus) = @_; |
80 | ||
8b3f9144 | 81 | my $filename = "$self->{statusdir}/hardware_status"; |
0cfd8f5b DM |
82 | |
83 | PVE::Tools::file_set_contents($filename, encode_json($cstatus)); | |
84 | }; | |
85 | ||
95360669 DM |
86 | sub read_service_config { |
87 | my ($self) = @_; | |
88 | ||
89 | my $filename = "$self->{statusdir}/service_config"; | |
289e4784 | 90 | my $conf = PVE::HA::Tools::read_json_from_file($filename); |
95360669 DM |
91 | |
92 | foreach my $sid (keys %$conf) { | |
93 | my $d = $conf->{$sid}; | |
8456bde2 DM |
94 | |
95 | die "service '$sid' without assigned node!" if !$d->{node}; | |
96 | ||
ec368d74 | 97 | if ($sid =~ m/^(vm|ct|fa):(\d+)$/) { |
b026c8c9 DM |
98 | $d->{type} = $1; |
99 | $d->{name} = $2; | |
95360669 DM |
100 | } else { |
101 | die "implement me"; | |
102 | } | |
103 | $d->{state} = 'disabled' if !$d->{state}; | |
bb07bd2c | 104 | $d->{state} = 'started' if $d->{state} eq 'enabled'; # backward compatibility |
ec368d74 TL |
105 | $d->{max_restart} = 1 if !defined($d->{max_restart}); |
106 | $d->{max_relocate} = 1 if !defined($d->{max_relocate}); | |
95360669 DM |
107 | } |
108 | ||
109 | return $conf; | |
110 | } | |
111 | ||
76b83c72 FE |
112 | sub update_service_config { |
113 | my ($self, $sid, $param) = @_; | |
114 | ||
115 | my $conf = $self->read_service_config(); | |
116 | ||
117 | my $sconf = $conf->{$sid} || die "no such resource '$sid'\n"; | |
118 | ||
119 | foreach my $k (%$param) { | |
120 | $sconf->{$k} = $param->{$k}; | |
121 | } | |
122 | ||
123 | $self->write_service_config($conf); | |
124 | } | |
125 | ||
79e0e005 DM |
126 | sub write_service_config { |
127 | my ($self, $conf) = @_; | |
128 | ||
95360669 DM |
129 | $self->{service_config} = $conf; |
130 | ||
79e0e005 DM |
131 | my $filename = "$self->{statusdir}/service_config"; |
132 | return PVE::HA::Tools::write_json_to_file($filename, $conf); | |
c982dfee TL |
133 | } |
134 | ||
135 | sub read_fence_config { | |
136 | my ($self) = @_; | |
137 | ||
138 | my $raw = undef; | |
139 | ||
140 | my $filename = "$self->{statusdir}/fence.cfg"; | |
141 | if (-e $filename) { | |
142 | $raw = PVE::Tools::file_get_contents($filename); | |
143 | } | |
144 | ||
145 | return PVE::HA::FenceConfig::parse_config($filename, $raw); | |
146 | } | |
147 | ||
148 | sub exec_fence_agent { | |
149 | my ($self, $agent, $node, @param) = @_; | |
150 | ||
151 | # let all agent succeed and behave the same for now | |
152 | $self->sim_hardware_cmd("power $node off", $agent); | |
153 | ||
154 | return 0; # EXIT_SUCCESS | |
155 | } | |
79e0e005 | 156 | |
e5f43426 TL |
157 | sub set_service_state { |
158 | my ($self, $sid, $state) = @_; | |
159 | ||
160 | my $conf = $self->read_service_config(); | |
161 | die "no such service '$sid'" if !$conf->{$sid}; | |
162 | ||
163 | $conf->{$sid}->{state} = $state; | |
164 | ||
165 | $self->write_service_config($conf); | |
166 | ||
167 | return $conf; | |
168 | } | |
169 | ||
27ccc95c TL |
170 | sub add_service { |
171 | my ($self, $sid, $opts) = @_; | |
172 | ||
173 | my $conf = $self->read_service_config(); | |
174 | die "resource ID '$sid' already defined\n" if $conf->{$sid}; | |
175 | ||
176 | $conf->{$sid} = $opts; | |
177 | ||
178 | $self->write_service_config($conf); | |
179 | ||
180 | return $conf; | |
181 | } | |
182 | ||
183 | sub delete_service { | |
184 | my ($self, $sid) = @_; | |
185 | ||
186 | my $conf = $self->read_service_config(); | |
187 | ||
188 | die "no such service '$sid'" if !$conf->{$sid}; | |
189 | ||
190 | delete $conf->{$sid}; | |
191 | ||
192 | $self->write_service_config($conf); | |
193 | ||
194 | return $conf; | |
195 | } | |
196 | ||
8456bde2 | 197 | sub change_service_location { |
6da27e23 | 198 | my ($self, $sid, $current_node, $new_node) = @_; |
8456bde2 DM |
199 | |
200 | my $conf = $self->read_service_config(); | |
201 | ||
202 | die "no such service '$sid'\n" if !$conf->{$sid}; | |
203 | ||
289e4784 | 204 | die "current_node for '$sid' does not match ($current_node != $conf->{$sid}->{node})\n" |
6da27e23 | 205 | if $current_node ne $conf->{$sid}->{node}; |
289e4784 | 206 | |
6da27e23 | 207 | $conf->{$sid}->{node} = $new_node; |
8456bde2 DM |
208 | |
209 | $self->write_service_config($conf); | |
210 | } | |
211 | ||
cde11324 TL |
212 | sub service_has_lock { |
213 | my ($self, $sid) = @_; | |
214 | ||
215 | my $conf = $self->read_service_config(); | |
216 | ||
217 | die "no such service '$sid'\n" if !$conf->{$sid}; | |
218 | ||
219 | return $conf->{$sid}->{lock}; | |
220 | } | |
221 | ||
222 | sub lock_service { | |
223 | my ($self, $sid, $lock) = @_; | |
224 | ||
225 | my $conf = $self->read_service_config(); | |
226 | ||
227 | die "no such service '$sid'\n" if !$conf->{$sid}; | |
228 | ||
229 | $conf->{$sid}->{lock} = $lock || 'backup'; | |
230 | ||
231 | $self->write_service_config($conf); | |
232 | ||
233 | return $conf; | |
234 | } | |
235 | ||
236 | sub unlock_service { | |
237 | my ($self, $sid, $lock) = @_; | |
238 | ||
239 | my $conf = $self->read_service_config(); | |
240 | ||
241 | die "no such service '$sid'\n" if !$conf->{$sid}; | |
242 | ||
243 | if (!defined($conf->{$sid}->{lock})) { | |
cde11324 TL |
244 | return undef; |
245 | } | |
246 | ||
247 | if (defined($lock) && $conf->{$sid}->{lock} ne $lock) { | |
248 | warn "found lock '$conf->{$sid}->{lock}' trying to remove '$lock' lock\n"; | |
249 | return undef; | |
250 | } | |
251 | ||
252 | my $removed_lock = delete $conf->{$sid}->{lock}; | |
253 | ||
254 | $self->write_service_config($conf); | |
255 | ||
256 | return $removed_lock; | |
257 | } | |
258 | ||
b70aa69e | 259 | sub queue_crm_commands_nolock { |
3b996922 DM |
260 | my ($self, $cmd) = @_; |
261 | ||
262 | chomp $cmd; | |
263 | ||
b70aa69e DM |
264 | my $data = ''; |
265 | my $filename = "$self->{statusdir}/crm_commands"; | |
266 | if (-f $filename) { | |
267 | $data = PVE::Tools::file_get_contents($filename); | |
268 | } | |
269 | $data .= "$cmd\n"; | |
270 | PVE::Tools::file_set_contents($filename, $data); | |
271 | ||
272 | return undef; | |
273 | } | |
274 | ||
275 | sub queue_crm_commands { | |
276 | my ($self, $cmd) = @_; | |
277 | ||
278 | my $code = sub { $self->queue_crm_commands_nolock($cmd); }; | |
289e4784 | 279 | |
3b996922 DM |
280 | $self->global_lock($code); |
281 | ||
282 | return undef; | |
283 | } | |
284 | ||
285 | sub read_crm_commands { | |
286 | my ($self) = @_; | |
287 | ||
288 | my $code = sub { | |
289 | my $data = ''; | |
290 | ||
291 | my $filename = "$self->{statusdir}/crm_commands"; | |
292 | if (-f $filename) { | |
293 | $data = PVE::Tools::file_get_contents($filename); | |
294 | } | |
295 | PVE::Tools::file_set_contents($filename, ''); | |
296 | ||
297 | return $data; | |
298 | }; | |
289e4784 | 299 | |
3b996922 DM |
300 | return $self->global_lock($code); |
301 | } | |
302 | ||
abc920b4 DM |
303 | sub read_group_config { |
304 | my ($self) = @_; | |
305 | ||
306 | my $filename = "$self->{statusdir}/groups"; | |
307 | my $raw = ''; | |
308 | $raw = PVE::Tools::file_get_contents($filename) if -f $filename; | |
309 | ||
7d33cb12 | 310 | return PVE::HA::Groups->parse_config($filename, $raw); |
abc920b4 DM |
311 | } |
312 | ||
c4a221bc | 313 | sub read_service_status { |
8456bde2 | 314 | my ($self, $node) = @_; |
c4a221bc | 315 | |
8456bde2 | 316 | my $filename = "$self->{statusdir}/service_status_$node"; |
289e4784 | 317 | return PVE::HA::Tools::read_json_from_file($filename); |
c4a221bc DM |
318 | } |
319 | ||
320 | sub write_service_status { | |
8456bde2 DM |
321 | my ($self, $node, $data) = @_; |
322 | ||
323 | my $filename = "$self->{statusdir}/service_status_$node"; | |
324 | my $res = PVE::HA::Tools::write_json_to_file($filename, $data); | |
325 | ||
326 | # fixme: add test if a service runs on two nodes!!! | |
c4a221bc | 327 | |
8456bde2 | 328 | return $res; |
289e4784 | 329 | } |
c4a221bc | 330 | |
abc920b4 DM |
331 | my $default_group_config = <<__EOD; |
332 | group: prefer_node1 | |
333 | nodes node1 | |
e941bdc5 | 334 | nofailback 1 |
abc920b4 DM |
335 | |
336 | group: prefer_node2 | |
337 | nodes node2 | |
e941bdc5 | 338 | nofailback 1 |
abc920b4 DM |
339 | |
340 | group: prefer_node3 | |
7a294ad4 | 341 | nodes node3 |
e941bdc5 | 342 | nofailback 1 |
abc920b4 DM |
343 | __EOD |
344 | ||
0cfd8f5b DM |
345 | sub new { |
346 | my ($this, $testdir) = @_; | |
347 | ||
348 | die "missing testdir" if !$testdir; | |
349 | ||
ba9e808e TL |
350 | die "testdir '$testdir' does not exist or is not a directory!\n" |
351 | if !-d $testdir; | |
352 | ||
0cfd8f5b DM |
353 | my $class = ref($this) || $this; |
354 | ||
355 | my $self = bless {}, $class; | |
356 | ||
787b66eb DM |
357 | my $statusdir = $self->{statusdir} = "$testdir/status"; |
358 | ||
359 | remove_tree($statusdir); | |
360 | mkdir $statusdir; | |
0cfd8f5b | 361 | |
787b66eb DM |
362 | # copy initial configuartion |
363 | copy("$testdir/manager_status", "$statusdir/manager_status"); # optional | |
79e0e005 | 364 | |
abc920b4 DM |
365 | if (-f "$testdir/groups") { |
366 | copy("$testdir/groups", "$statusdir/groups"); | |
367 | } else { | |
368 | PVE::Tools::file_set_contents("$statusdir/groups", $default_group_config); | |
369 | } | |
370 | ||
79e0e005 DM |
371 | if (-f "$testdir/service_config") { |
372 | copy("$testdir/service_config", "$statusdir/service_config"); | |
373 | } else { | |
374 | my $conf = { | |
eda9314d DM |
375 | 'vm:101' => { node => 'node1', group => 'prefer_node1' }, |
376 | 'vm:102' => { node => 'node2', group => 'prefer_node2' }, | |
377 | 'vm:103' => { node => 'node3', group => 'prefer_node3' }, | |
378 | 'vm:104' => { node => 'node1', group => 'prefer_node1' }, | |
379 | 'vm:105' => { node => 'node2', group => 'prefer_node2' }, | |
380 | 'vm:106' => { node => 'node3', group => 'prefer_node3' }, | |
79e0e005 DM |
381 | }; |
382 | $self->write_service_config($conf); | |
383 | } | |
787b66eb | 384 | |
853f5867 DM |
385 | if (-f "$testdir/hardware_status") { |
386 | copy("$testdir/hardware_status", "$statusdir/hardware_status") || | |
387 | die "Copy failed: $!\n"; | |
388 | } else { | |
389 | my $cstatus = { | |
390 | node1 => { power => 'off', network => 'off' }, | |
391 | node2 => { power => 'off', network => 'off' }, | |
392 | node3 => { power => 'off', network => 'off' }, | |
393 | }; | |
394 | $self->write_hardware_status_nolock($cstatus); | |
395 | } | |
787b66eb | 396 | |
c982dfee TL |
397 | if (-f "$testdir/fence.cfg") { |
398 | copy("$testdir/fence.cfg", "$statusdir/fence.cfg"); | |
399 | } | |
0cfd8f5b | 400 | |
ed408b44 TL |
401 | if (-f "$testdir/datacenter.cfg") { |
402 | copy("$testdir/datacenter.cfg", "$statusdir/datacenter.cfg"); | |
403 | } | |
404 | ||
8b3f9144 | 405 | my $cstatus = $self->read_hardware_status_nolock(); |
0cfd8f5b DM |
406 | |
407 | foreach my $node (sort keys %$cstatus) { | |
0bba8f60 | 408 | $self->{nodes}->{$node} = {}; |
8456bde2 DM |
409 | |
410 | if (-f "$testdir/service_status_$node") { | |
411 | copy("$testdir/service_status_$node", "$statusdir/service_status_$node"); | |
289e4784 | 412 | } else { |
8456bde2 DM |
413 | $self->write_service_status($node, {}); |
414 | } | |
0cfd8f5b DM |
415 | } |
416 | ||
95360669 DM |
417 | $self->{service_config} = $self->read_service_config(); |
418 | ||
0cfd8f5b DM |
419 | return $self; |
420 | } | |
421 | ||
422 | sub get_time { | |
423 | my ($self) = @_; | |
424 | ||
bf93e2a2 | 425 | die "implement in subclass"; |
0cfd8f5b DM |
426 | } |
427 | ||
428 | sub log { | |
fde8362a | 429 | my ($self, $level, $msg, $id) = @_; |
0cfd8f5b DM |
430 | |
431 | chomp $msg; | |
432 | ||
433 | my $time = $self->get_time(); | |
434 | ||
fde8362a DM |
435 | $id = 'hardware' if !$id; |
436 | ||
0bba8f60 | 437 | printf("%-5s %5d %12s: $msg\n", $level, $time, $id); |
0cfd8f5b DM |
438 | } |
439 | ||
440 | sub statusdir { | |
441 | my ($self, $node) = @_; | |
442 | ||
443 | return $self->{statusdir}; | |
444 | } | |
445 | ||
ed408b44 TL |
446 | sub read_datacenter_conf { |
447 | my ($self, $node) = @_; | |
448 | ||
449 | my $filename = "$self->{statusdir}/datacenter.cfg"; | |
450 | return PVE::HA::Tools::read_json_from_file($filename, {}); | |
451 | } | |
452 | ||
8b3f9144 | 453 | sub global_lock { |
0cfd8f5b DM |
454 | my ($self, $code, @param) = @_; |
455 | ||
8b3f9144 | 456 | my $lockfile = "$self->{statusdir}/hardware.lck"; |
0cfd8f5b DM |
457 | my $fh = IO::File->new(">>$lockfile") || |
458 | die "unable to open '$lockfile'\n"; | |
459 | ||
460 | my $success; | |
461 | for (;;) { | |
462 | $success = flock($fh, LOCK_EX); | |
463 | if ($success || ($! != EINTR)) { | |
464 | last; | |
465 | } | |
466 | if (!$success) { | |
9de9a6ce | 467 | close($fh); |
63f6a08c | 468 | die "can't acquire lock '$lockfile' - $!\n"; |
0cfd8f5b DM |
469 | } |
470 | } | |
9de9a6ce | 471 | |
0cfd8f5b DM |
472 | my $res; |
473 | ||
9de9a6ce | 474 | eval { $res = &$code($fh, @param) }; |
0cfd8f5b | 475 | my $err = $@; |
289e4784 | 476 | |
0cfd8f5b DM |
477 | close($fh); |
478 | ||
479 | die $err if $err; | |
289e4784 | 480 | |
0cfd8f5b DM |
481 | return $res; |
482 | } | |
483 | ||
8b3f9144 DM |
484 | my $compute_node_info = sub { |
485 | my ($self, $cstatus) = @_; | |
486 | ||
487 | my $node_info = {}; | |
488 | ||
489 | my $node_count = 0; | |
490 | my $online_count = 0; | |
491 | ||
492 | foreach my $node (keys %$cstatus) { | |
493 | my $d = $cstatus->{$node}; | |
494 | ||
495 | my $online = ($d->{power} eq 'on' && $d->{network} eq 'on') ? 1 : 0; | |
496 | $node_info->{$node}->{online} = $online; | |
497 | ||
498 | $node_count++; | |
499 | $online_count++ if $online; | |
500 | } | |
501 | ||
502 | my $quorate = ($online_count > int($node_count/2)) ? 1 : 0; | |
289e4784 | 503 | |
8b3f9144 DM |
504 | if (!$quorate) { |
505 | foreach my $node (keys %$cstatus) { | |
506 | my $d = $cstatus->{$node}; | |
507 | $node_info->{$node}->{online} = 0; | |
508 | } | |
509 | } | |
510 | ||
511 | return ($node_info, $quorate); | |
512 | }; | |
513 | ||
514 | sub get_node_info { | |
515 | my ($self) = @_; | |
516 | ||
5516f102 TL |
517 | my $cstatus = $self->read_hardware_status_nolock(); |
518 | my ($node_info, $quorate) = &$compute_node_info($self, $cstatus); | |
8b3f9144 DM |
519 | |
520 | return ($node_info, $quorate); | |
521 | } | |
522 | ||
ba2a45cd TL |
523 | # helper for Sim/ only |
524 | sub get_cfs_state { | |
525 | my ($self, $node, $state) = @_; | |
526 | ||
527 | # TODO: ensure nolock is OK when adding this to RTSim | |
528 | my $cstatus = $self->read_hardware_status_nolock(); | |
529 | my $res = $cstatus->{$node}->{cfs}->{$state}; | |
530 | ||
531 | # we assume default true if not defined | |
532 | return !defined($res) || $res; | |
533 | } | |
534 | ||
8b3f9144 | 535 | # simulate hardware commands |
0cfd8f5b DM |
536 | # power <node> <on|off> |
537 | # network <node> <on|off> | |
ba2a45cd | 538 | # cfs <node> <rw|update> <work|fail> |
e08a0717 TL |
539 | # reboot <node> |
540 | # shutdown <node> | |
541 | # restart-lrm <node> | |
667670b2 | 542 | # service <sid> <started|disabled|stopped|ignored> |
e08a0717 | 543 | # service <sid> <migrate|relocate> <target> |
21caf0db | 544 | # service <sid> stop <timeout> |
e08a0717 | 545 | # service <sid> lock/unlock [lockname] |
0cfd8f5b | 546 | |
8b3f9144 | 547 | sub sim_hardware_cmd { |
fde8362a | 548 | my ($self, $cmdstr, $logid) = @_; |
0cfd8f5b | 549 | |
e08a0717 TL |
550 | my $code = sub { |
551 | my ($lock_fh) = @_; | |
552 | ||
553 | my $cstatus = $self->read_hardware_status_nolock(); | |
554 | ||
b94b4785 | 555 | my ($cmd, $objid, $action, $param) = split(/\s+/, $cmdstr); |
e08a0717 TL |
556 | |
557 | die "sim_hardware_cmd: no node or service for command specified" | |
558 | if !$objid; | |
559 | ||
560 | my ($node, $sid, $d); | |
561 | ||
562 | if ($cmd eq 'service') { | |
563 | $sid = PVE::HA::Tools::pve_verify_ha_resource_id($objid); | |
564 | } else { | |
565 | $node = $objid; | |
566 | $d = $self->{nodes}->{$node} || | |
567 | die "sim_hardware_cmd: no such node '$node'\n"; | |
568 | } | |
569 | ||
570 | $self->log('info', "execute $cmdstr", $logid); | |
571 | ||
572 | if ($cmd eq 'power') { | |
573 | die "sim_hardware_cmd: unknown action '$action'\n" | |
574 | if $action !~ m/^(on|off)$/; | |
575 | ||
576 | if ($cstatus->{$node}->{power} ne $action) { | |
577 | if ($action eq 'on') { | |
578 | ||
579 | $d->{crm} = $self->crm_control('start', $d, $lock_fh) if !defined($d->{crm}); | |
580 | $d->{lrm} = $self->lrm_control('start', $d, $lock_fh) if !defined($d->{lrm}); | |
581 | $d->{lrm_restart} = undef; | |
ba2a45cd | 582 | $cstatus->{$node}->{cfs} = {}; |
e08a0717 TL |
583 | |
584 | } else { | |
585 | ||
586 | if ($d->{crm}) { | |
587 | $d->{crm_env}->log('info', "killed by poweroff"); | |
588 | $self->crm_control('stop', $d, $lock_fh); | |
589 | $d->{crm} = undef; | |
590 | } | |
591 | if ($d->{lrm}) { | |
592 | $d->{lrm_env}->log('info', "killed by poweroff"); | |
593 | $self->lrm_control('stop', $d, $lock_fh); | |
594 | $d->{lrm} = undef; | |
595 | $d->{lrm_restart} = undef; | |
596 | } | |
597 | ||
598 | $self->watchdog_reset_nolock($node); | |
599 | $self->write_service_status($node, {}); | |
600 | } | |
601 | } | |
602 | ||
603 | $cstatus->{$node}->{power} = $action; | |
604 | $cstatus->{$node}->{network} = $action; | |
605 | $cstatus->{$node}->{shutdown} = undef; | |
606 | ||
607 | $self->write_hardware_status_nolock($cstatus); | |
608 | ||
609 | } elsif ($cmd eq 'network') { | |
610 | die "sim_hardware_cmd: unknown network action '$action'" | |
611 | if $action !~ m/^(on|off)$/; | |
612 | $cstatus->{$node}->{network} = $action; | |
613 | ||
614 | $self->write_hardware_status_nolock($cstatus); | |
615 | ||
ba2a45cd TL |
616 | } elsif ($cmd eq 'cfs') { |
617 | die "sim_hardware_cmd: unknown cfs action '$action' for node '$node'" | |
618 | if $action !~ m/^(rw|update)$/; | |
b94b4785 FE |
619 | die "sim_hardware_cmd: unknown cfs command '$param' for '$action' on node '$node'" |
620 | if $param !~ m/^(work|fail)$/; | |
ba2a45cd | 621 | |
b94b4785 | 622 | $cstatus->{$node}->{cfs}->{$action} = $param eq 'work'; |
ba2a45cd TL |
623 | $self->write_hardware_status_nolock($cstatus); |
624 | ||
e08a0717 TL |
625 | } elsif ($cmd eq 'reboot' || $cmd eq 'shutdown') { |
626 | $cstatus->{$node}->{shutdown} = $cmd; | |
627 | ||
628 | $self->write_hardware_status_nolock($cstatus); | |
629 | ||
630 | $self->lrm_control('shutdown', $d, $lock_fh) if defined($d->{lrm}); | |
631 | } elsif ($cmd eq 'restart-lrm') { | |
632 | if ($d->{lrm}) { | |
633 | $d->{lrm_restart} = 1; | |
634 | $self->lrm_control('shutdown', $d, $lock_fh); | |
635 | } | |
636 | } elsif ($cmd eq 'crm') { | |
637 | ||
638 | if ($action eq 'stop') { | |
639 | if ($d->{crm}) { | |
640 | $d->{crm_stop} = 1; | |
641 | $self->crm_control('shutdown', $d, $lock_fh); | |
642 | } | |
643 | } elsif ($action eq 'start') { | |
644 | $d->{crm} = $self->crm_control('start', $d, $lock_fh) if !defined($d->{crm}); | |
645 | } else { | |
646 | die "sim_hardware_cmd: unknown action '$action'"; | |
647 | } | |
648 | ||
649 | } elsif ($cmd eq 'service') { | |
667670b2 TL |
650 | if ($action eq 'started' || $action eq 'disabled' || |
651 | $action eq 'stopped' || $action eq 'ignored') { | |
e08a0717 TL |
652 | |
653 | $self->set_service_state($sid, $action); | |
654 | ||
655 | } elsif ($action eq 'migrate' || $action eq 'relocate') { | |
656 | ||
657 | die "sim_hardware_cmd: missing target node for '$action' command" | |
b94b4785 | 658 | if !$param; |
e08a0717 | 659 | |
b94b4785 | 660 | $self->queue_crm_commands_nolock("$action $sid $param"); |
e08a0717 | 661 | |
21caf0db FE |
662 | } elsif ($action eq 'stop') { |
663 | ||
664 | die "sim_hardware_cmd: missing timeout for '$action' command" | |
665 | if !defined($param); | |
666 | ||
667 | $self->queue_crm_commands_nolock("$action $sid $param"); | |
668 | ||
e08a0717 TL |
669 | } elsif ($action eq 'add') { |
670 | ||
b94b4785 | 671 | $self->add_service($sid, {state => 'started', node => $param}); |
e08a0717 TL |
672 | |
673 | } elsif ($action eq 'delete') { | |
674 | ||
675 | $self->delete_service($sid); | |
676 | ||
677 | } elsif ($action eq 'lock') { | |
678 | ||
b94b4785 | 679 | $self->lock_service($sid, $param); |
e08a0717 TL |
680 | |
681 | } elsif ($action eq 'unlock') { | |
682 | ||
b94b4785 | 683 | $self->unlock_service($sid, $param); |
e08a0717 TL |
684 | |
685 | } else { | |
686 | die "sim_hardware_cmd: unknown service action '$action' " . | |
687 | "- not implemented\n" | |
688 | } | |
689 | } else { | |
690 | die "sim_hardware_cmd: unknown command '$cmdstr'\n"; | |
691 | } | |
692 | ||
693 | return $cstatus; | |
694 | }; | |
695 | ||
696 | return $self->global_lock($code); | |
697 | } | |
698 | ||
699 | # for controlling the resource manager services | |
700 | sub crm_control { | |
701 | my ($self, $action, $data, $lock_fh) = @_; | |
702 | ||
703 | die "implement in subclass"; | |
704 | } | |
705 | ||
706 | sub lrm_control { | |
707 | my ($self, $action, $data, $lock_fh) = @_; | |
708 | ||
bf93e2a2 | 709 | die "implement in subclass"; |
0cfd8f5b DM |
710 | } |
711 | ||
712 | sub run { | |
713 | my ($self) = @_; | |
714 | ||
bf93e2a2 | 715 | die "implement in subclass"; |
0cfd8f5b | 716 | } |
9329c1e2 DM |
717 | |
718 | my $modify_watchog = sub { | |
719 | my ($self, $code) = @_; | |
720 | ||
721 | my $update_cmd = sub { | |
722 | ||
723 | my $filename = "$self->{statusdir}/watchdog_status"; | |
289e4784 | 724 | |
9329c1e2 DM |
725 | my ($res, $wdstatus); |
726 | ||
727 | if (-f $filename) { | |
728 | my $raw = PVE::Tools::file_get_contents($filename); | |
729 | $wdstatus = decode_json($raw); | |
730 | } else { | |
731 | $wdstatus = {}; | |
732 | } | |
289e4784 | 733 | |
9329c1e2 DM |
734 | ($wdstatus, $res) = &$code($wdstatus); |
735 | ||
736 | PVE::Tools::file_set_contents($filename, encode_json($wdstatus)); | |
737 | ||
738 | return $res; | |
739 | }; | |
740 | ||
741 | return $self->global_lock($update_cmd); | |
742 | }; | |
743 | ||
0590c6a7 DM |
744 | sub watchdog_reset_nolock { |
745 | my ($self, $node) = @_; | |
746 | ||
747 | my $filename = "$self->{statusdir}/watchdog_status"; | |
748 | ||
749 | if (-f $filename) { | |
750 | my $raw = PVE::Tools::file_get_contents($filename); | |
751 | my $wdstatus = decode_json($raw); | |
752 | ||
753 | foreach my $id (keys %$wdstatus) { | |
754 | delete $wdstatus->{$id} if $wdstatus->{$id}->{node} eq $node; | |
755 | } | |
289e4784 | 756 | |
0590c6a7 DM |
757 | PVE::Tools::file_set_contents($filename, encode_json($wdstatus)); |
758 | } | |
759 | } | |
760 | ||
9329c1e2 DM |
761 | sub watchdog_check { |
762 | my ($self, $node) = @_; | |
763 | ||
764 | my $code = sub { | |
765 | my ($wdstatus) = @_; | |
766 | ||
767 | my $res = 1; | |
768 | ||
769 | foreach my $wfh (keys %$wdstatus) { | |
770 | my $wd = $wdstatus->{$wfh}; | |
771 | next if $wd->{node} ne $node; | |
772 | ||
773 | my $ctime = $self->get_time(); | |
774 | my $tdiff = $ctime - $wd->{update_time}; | |
775 | ||
0bba8f60 | 776 | if ($tdiff > $watchdog_timeout) { # expired |
9329c1e2 DM |
777 | $res = 0; |
778 | delete $wdstatus->{$wfh}; | |
779 | } | |
780 | } | |
289e4784 | 781 | |
9329c1e2 DM |
782 | return ($wdstatus, $res); |
783 | }; | |
784 | ||
785 | return &$modify_watchog($self, $code); | |
786 | } | |
787 | ||
788 | my $wdcounter = 0; | |
789 | ||
790 | sub watchdog_open { | |
791 | my ($self, $node) = @_; | |
792 | ||
793 | my $code = sub { | |
794 | my ($wdstatus) = @_; | |
795 | ||
796 | ++$wdcounter; | |
797 | ||
798 | my $id = "WD:$node:$$:$wdcounter"; | |
799 | ||
800 | die "internal error" if defined($wdstatus->{$id}); | |
801 | ||
802 | $wdstatus->{$id} = { | |
803 | node => $node, | |
804 | update_time => $self->get_time(), | |
805 | }; | |
806 | ||
807 | return ($wdstatus, $id); | |
808 | }; | |
809 | ||
810 | return &$modify_watchog($self, $code); | |
811 | } | |
812 | ||
813 | sub watchdog_close { | |
814 | my ($self, $wfh) = @_; | |
815 | ||
816 | my $code = sub { | |
817 | my ($wdstatus) = @_; | |
818 | ||
819 | my $wd = $wdstatus->{$wfh}; | |
820 | die "no such watchdog handle '$wfh'\n" if !defined($wd); | |
821 | ||
822 | my $tdiff = $self->get_time() - $wd->{update_time}; | |
0bba8f60 | 823 | die "watchdog expired" if $tdiff > $watchdog_timeout; |
9329c1e2 DM |
824 | |
825 | delete $wdstatus->{$wfh}; | |
826 | ||
827 | return ($wdstatus); | |
828 | }; | |
829 | ||
830 | return &$modify_watchog($self, $code); | |
831 | } | |
832 | ||
833 | sub watchdog_update { | |
834 | my ($self, $wfh) = @_; | |
835 | ||
836 | my $code = sub { | |
837 | my ($wdstatus) = @_; | |
838 | ||
839 | my $wd = $wdstatus->{$wfh}; | |
840 | ||
841 | die "no such watchdog handle '$wfh'\n" if !defined($wd); | |
842 | ||
843 | my $ctime = $self->get_time(); | |
844 | my $tdiff = $ctime - $wd->{update_time}; | |
845 | ||
0bba8f60 | 846 | die "watchdog expired" if $tdiff > $watchdog_timeout; |
289e4784 | 847 | |
9329c1e2 DM |
848 | $wd->{update_time} = $ctime; |
849 | ||
850 | return ($wdstatus); | |
851 | }; | |
852 | ||
853 | return &$modify_watchog($self, $code); | |
854 | } | |
855 | ||
0cfd8f5b | 856 | 1; |