]>
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; | |
a0a7d11e | 11 | |
0cfd8f5b | 12 | use Fcntl qw(:DEFAULT :flock); |
787b66eb DM |
13 | use File::Copy; |
14 | use File::Path qw(make_path remove_tree); | |
a0a7d11e TL |
15 | use IO::File; |
16 | use JSON; | |
17 | use POSIX qw(strftime EINTR); | |
18 | ||
c982dfee | 19 | use PVE::HA::FenceConfig; |
7d33cb12 | 20 | use PVE::HA::Groups; |
f5a14b93 | 21 | |
17b5cf98 | 22 | my $watchdog_timeout = 60; |
0bba8f60 | 23 | |
787b66eb DM |
24 | # Status directory layout |
25 | # | |
26 | # configuration | |
27 | # | |
8456bde2 DM |
28 | # $testdir/cmdlist Command list for simulation |
29 | # $testdir/hardware_status Hardware description (number of nodes, ...) | |
30 | # $testdir/manager_status CRM status (start with {}) | |
31 | # $testdir/service_config Service configuration | |
eea0c609 | 32 | # $testdir/static_service_stats Static service usage information (cpu, memory) |
abc920b4 | 33 | # $testdir/groups HA groups configuration |
8456bde2 | 34 | # $testdir/service_status_<node> Service status |
ed408b44 | 35 | # $testdir/datacenter.cfg Datacenter wide HA configuration |
3c36cbca | 36 | |
9329c1e2 DM |
37 | # |
38 | # runtime status for simulation system | |
39 | # | |
40 | # $testdir/status/cluster_locks Cluster locks | |
41 | # $testdir/status/hardware_status Hardware status (power/network on/off) | |
eea0c609 | 42 | # $testdir/status/static_service_stats Static service usage information (cpu, memory) |
9329c1e2 | 43 | # $testdir/status/watchdog_status Watchdog status |
787b66eb DM |
44 | # |
45 | # runtime status | |
9329c1e2 | 46 | # |
8456bde2 DM |
47 | # $testdir/status/lrm_status_<node> LRM status |
48 | # $testdir/status/manager_status CRM status | |
abc920b4 | 49 | # $testdir/status/crm_commands CRM command queue |
8456bde2 DM |
50 | # $testdir/status/service_config Service configuration |
51 | # $testdir/status/service_status_<node> Service status | |
abc920b4 | 52 | # $testdir/status/groups HA groups configuration |
c4a221bc DM |
53 | |
54 | sub read_lrm_status { | |
55 | my ($self, $node) = @_; | |
56 | ||
57 | my $filename = "$self->{statusdir}/lrm_status_$node"; | |
58 | ||
289e4784 | 59 | return PVE::HA::Tools::read_json_from_file($filename, {}); |
c4a221bc DM |
60 | } |
61 | ||
62 | sub write_lrm_status { | |
63 | my ($self, $node, $status_obj) = @_; | |
64 | ||
65 | my $filename = "$self->{statusdir}/lrm_status_$node"; | |
66 | ||
289e4784 | 67 | PVE::HA::Tools::write_json_to_file($filename, $status_obj); |
c4a221bc | 68 | } |
787b66eb | 69 | |
8b3f9144 | 70 | sub read_hardware_status_nolock { |
0cfd8f5b DM |
71 | my ($self) = @_; |
72 | ||
8b3f9144 | 73 | my $filename = "$self->{statusdir}/hardware_status"; |
0cfd8f5b DM |
74 | |
75 | my $raw = PVE::Tools::file_get_contents($filename); | |
76 | my $cstatus = decode_json($raw); | |
77 | ||
78 | return $cstatus; | |
79 | } | |
80 | ||
8b3f9144 | 81 | sub write_hardware_status_nolock { |
0cfd8f5b DM |
82 | my ($self, $cstatus) = @_; |
83 | ||
8b3f9144 | 84 | my $filename = "$self->{statusdir}/hardware_status"; |
0cfd8f5b DM |
85 | |
86 | PVE::Tools::file_set_contents($filename, encode_json($cstatus)); | |
87 | }; | |
88 | ||
95360669 DM |
89 | sub read_service_config { |
90 | my ($self) = @_; | |
91 | ||
92 | my $filename = "$self->{statusdir}/service_config"; | |
289e4784 | 93 | my $conf = PVE::HA::Tools::read_json_from_file($filename); |
95360669 DM |
94 | |
95 | foreach my $sid (keys %$conf) { | |
96 | my $d = $conf->{$sid}; | |
8456bde2 DM |
97 | |
98 | die "service '$sid' without assigned node!" if !$d->{node}; | |
99 | ||
ec368d74 | 100 | if ($sid =~ m/^(vm|ct|fa):(\d+)$/) { |
b026c8c9 DM |
101 | $d->{type} = $1; |
102 | $d->{name} = $2; | |
95360669 DM |
103 | } else { |
104 | die "implement me"; | |
105 | } | |
106 | $d->{state} = 'disabled' if !$d->{state}; | |
bb07bd2c | 107 | $d->{state} = 'started' if $d->{state} eq 'enabled'; # backward compatibility |
ec368d74 TL |
108 | $d->{max_restart} = 1 if !defined($d->{max_restart}); |
109 | $d->{max_relocate} = 1 if !defined($d->{max_relocate}); | |
95360669 DM |
110 | } |
111 | ||
112 | return $conf; | |
113 | } | |
114 | ||
76b83c72 FE |
115 | sub update_service_config { |
116 | my ($self, $sid, $param) = @_; | |
117 | ||
118 | my $conf = $self->read_service_config(); | |
119 | ||
120 | my $sconf = $conf->{$sid} || die "no such resource '$sid'\n"; | |
121 | ||
122 | foreach my $k (%$param) { | |
123 | $sconf->{$k} = $param->{$k}; | |
124 | } | |
125 | ||
126 | $self->write_service_config($conf); | |
127 | } | |
128 | ||
79e0e005 DM |
129 | sub write_service_config { |
130 | my ($self, $conf) = @_; | |
131 | ||
95360669 DM |
132 | $self->{service_config} = $conf; |
133 | ||
79e0e005 DM |
134 | my $filename = "$self->{statusdir}/service_config"; |
135 | return PVE::HA::Tools::write_json_to_file($filename, $conf); | |
c982dfee TL |
136 | } |
137 | ||
138 | sub read_fence_config { | |
139 | my ($self) = @_; | |
140 | ||
141 | my $raw = undef; | |
142 | ||
143 | my $filename = "$self->{statusdir}/fence.cfg"; | |
144 | if (-e $filename) { | |
145 | $raw = PVE::Tools::file_get_contents($filename); | |
146 | } | |
147 | ||
148 | return PVE::HA::FenceConfig::parse_config($filename, $raw); | |
149 | } | |
150 | ||
151 | sub exec_fence_agent { | |
152 | my ($self, $agent, $node, @param) = @_; | |
153 | ||
154 | # let all agent succeed and behave the same for now | |
155 | $self->sim_hardware_cmd("power $node off", $agent); | |
156 | ||
157 | return 0; # EXIT_SUCCESS | |
158 | } | |
79e0e005 | 159 | |
e5f43426 TL |
160 | sub set_service_state { |
161 | my ($self, $sid, $state) = @_; | |
162 | ||
163 | my $conf = $self->read_service_config(); | |
164 | die "no such service '$sid'" if !$conf->{$sid}; | |
165 | ||
166 | $conf->{$sid}->{state} = $state; | |
167 | ||
168 | $self->write_service_config($conf); | |
169 | ||
170 | return $conf; | |
171 | } | |
172 | ||
27ccc95c TL |
173 | sub add_service { |
174 | my ($self, $sid, $opts) = @_; | |
175 | ||
176 | my $conf = $self->read_service_config(); | |
177 | die "resource ID '$sid' already defined\n" if $conf->{$sid}; | |
178 | ||
179 | $conf->{$sid} = $opts; | |
1323ef6e | 180 | $conf->{$sid}->@{qw(type name)} = split(/:/, $sid); |
27ccc95c TL |
181 | |
182 | $self->write_service_config($conf); | |
183 | ||
184 | return $conf; | |
185 | } | |
186 | ||
187 | sub delete_service { | |
188 | my ($self, $sid) = @_; | |
189 | ||
190 | my $conf = $self->read_service_config(); | |
191 | ||
192 | die "no such service '$sid'" if !$conf->{$sid}; | |
193 | ||
194 | delete $conf->{$sid}; | |
195 | ||
196 | $self->write_service_config($conf); | |
197 | ||
198 | return $conf; | |
199 | } | |
200 | ||
8456bde2 | 201 | sub change_service_location { |
6da27e23 | 202 | my ($self, $sid, $current_node, $new_node) = @_; |
8456bde2 DM |
203 | |
204 | my $conf = $self->read_service_config(); | |
205 | ||
206 | die "no such service '$sid'\n" if !$conf->{$sid}; | |
207 | ||
289e4784 | 208 | die "current_node for '$sid' does not match ($current_node != $conf->{$sid}->{node})\n" |
6da27e23 | 209 | if $current_node ne $conf->{$sid}->{node}; |
289e4784 | 210 | |
6da27e23 | 211 | $conf->{$sid}->{node} = $new_node; |
8456bde2 DM |
212 | |
213 | $self->write_service_config($conf); | |
214 | } | |
215 | ||
cde11324 TL |
216 | sub service_has_lock { |
217 | my ($self, $sid) = @_; | |
218 | ||
219 | my $conf = $self->read_service_config(); | |
220 | ||
221 | die "no such service '$sid'\n" if !$conf->{$sid}; | |
222 | ||
223 | return $conf->{$sid}->{lock}; | |
224 | } | |
225 | ||
226 | sub lock_service { | |
227 | my ($self, $sid, $lock) = @_; | |
228 | ||
229 | my $conf = $self->read_service_config(); | |
230 | ||
231 | die "no such service '$sid'\n" if !$conf->{$sid}; | |
232 | ||
233 | $conf->{$sid}->{lock} = $lock || 'backup'; | |
234 | ||
235 | $self->write_service_config($conf); | |
236 | ||
237 | return $conf; | |
238 | } | |
239 | ||
240 | sub unlock_service { | |
241 | my ($self, $sid, $lock) = @_; | |
242 | ||
243 | my $conf = $self->read_service_config(); | |
244 | ||
245 | die "no such service '$sid'\n" if !$conf->{$sid}; | |
246 | ||
247 | if (!defined($conf->{$sid}->{lock})) { | |
cde11324 TL |
248 | return undef; |
249 | } | |
250 | ||
251 | if (defined($lock) && $conf->{$sid}->{lock} ne $lock) { | |
252 | warn "found lock '$conf->{$sid}->{lock}' trying to remove '$lock' lock\n"; | |
253 | return undef; | |
254 | } | |
255 | ||
256 | my $removed_lock = delete $conf->{$sid}->{lock}; | |
257 | ||
258 | $self->write_service_config($conf); | |
259 | ||
260 | return $removed_lock; | |
261 | } | |
262 | ||
b70aa69e | 263 | sub queue_crm_commands_nolock { |
3b996922 DM |
264 | my ($self, $cmd) = @_; |
265 | ||
266 | chomp $cmd; | |
267 | ||
b70aa69e DM |
268 | my $data = ''; |
269 | my $filename = "$self->{statusdir}/crm_commands"; | |
270 | if (-f $filename) { | |
271 | $data = PVE::Tools::file_get_contents($filename); | |
272 | } | |
273 | $data .= "$cmd\n"; | |
274 | PVE::Tools::file_set_contents($filename, $data); | |
275 | ||
276 | return undef; | |
277 | } | |
278 | ||
279 | sub queue_crm_commands { | |
280 | my ($self, $cmd) = @_; | |
281 | ||
282 | my $code = sub { $self->queue_crm_commands_nolock($cmd); }; | |
289e4784 | 283 | |
3b996922 DM |
284 | $self->global_lock($code); |
285 | ||
286 | return undef; | |
287 | } | |
288 | ||
289 | sub read_crm_commands { | |
290 | my ($self) = @_; | |
291 | ||
292 | my $code = sub { | |
293 | my $data = ''; | |
294 | ||
295 | my $filename = "$self->{statusdir}/crm_commands"; | |
296 | if (-f $filename) { | |
297 | $data = PVE::Tools::file_get_contents($filename); | |
298 | } | |
299 | PVE::Tools::file_set_contents($filename, ''); | |
300 | ||
301 | return $data; | |
302 | }; | |
289e4784 | 303 | |
3b996922 DM |
304 | return $self->global_lock($code); |
305 | } | |
306 | ||
abc920b4 DM |
307 | sub read_group_config { |
308 | my ($self) = @_; | |
309 | ||
310 | my $filename = "$self->{statusdir}/groups"; | |
311 | my $raw = ''; | |
312 | $raw = PVE::Tools::file_get_contents($filename) if -f $filename; | |
313 | ||
7d33cb12 | 314 | return PVE::HA::Groups->parse_config($filename, $raw); |
abc920b4 DM |
315 | } |
316 | ||
c4a221bc | 317 | sub read_service_status { |
8456bde2 | 318 | my ($self, $node) = @_; |
c4a221bc | 319 | |
8456bde2 | 320 | my $filename = "$self->{statusdir}/service_status_$node"; |
289e4784 | 321 | return PVE::HA::Tools::read_json_from_file($filename); |
c4a221bc DM |
322 | } |
323 | ||
324 | sub write_service_status { | |
8456bde2 DM |
325 | my ($self, $node, $data) = @_; |
326 | ||
327 | my $filename = "$self->{statusdir}/service_status_$node"; | |
328 | my $res = PVE::HA::Tools::write_json_to_file($filename, $data); | |
329 | ||
330 | # fixme: add test if a service runs on two nodes!!! | |
c4a221bc | 331 | |
8456bde2 | 332 | return $res; |
289e4784 | 333 | } |
c4a221bc | 334 | |
eea0c609 FE |
335 | sub read_static_service_stats { |
336 | my ($self) = @_; | |
337 | ||
338 | my $filename = "$self->{statusdir}/static_service_stats"; | |
49b0ccc7 TL |
339 | my $stats = eval { PVE::HA::Tools::read_json_from_file($filename) }; |
340 | $self->log('error', "loading static service stats failed - $@") if $@; | |
eea0c609 FE |
341 | |
342 | return $stats; | |
343 | } | |
344 | ||
abc920b4 DM |
345 | my $default_group_config = <<__EOD; |
346 | group: prefer_node1 | |
347 | nodes node1 | |
e941bdc5 | 348 | nofailback 1 |
abc920b4 DM |
349 | |
350 | group: prefer_node2 | |
351 | nodes node2 | |
e941bdc5 | 352 | nofailback 1 |
abc920b4 DM |
353 | |
354 | group: prefer_node3 | |
7a294ad4 | 355 | nodes node3 |
e941bdc5 | 356 | nofailback 1 |
abc920b4 DM |
357 | __EOD |
358 | ||
0cfd8f5b DM |
359 | sub new { |
360 | my ($this, $testdir) = @_; | |
361 | ||
362 | die "missing testdir" if !$testdir; | |
363 | ||
ba9e808e TL |
364 | die "testdir '$testdir' does not exist or is not a directory!\n" |
365 | if !-d $testdir; | |
366 | ||
0cfd8f5b DM |
367 | my $class = ref($this) || $this; |
368 | ||
369 | my $self = bless {}, $class; | |
370 | ||
787b66eb DM |
371 | my $statusdir = $self->{statusdir} = "$testdir/status"; |
372 | ||
373 | remove_tree($statusdir); | |
374 | mkdir $statusdir; | |
0cfd8f5b | 375 | |
787b66eb DM |
376 | # copy initial configuartion |
377 | copy("$testdir/manager_status", "$statusdir/manager_status"); # optional | |
79e0e005 | 378 | |
abc920b4 DM |
379 | if (-f "$testdir/groups") { |
380 | copy("$testdir/groups", "$statusdir/groups"); | |
381 | } else { | |
382 | PVE::Tools::file_set_contents("$statusdir/groups", $default_group_config); | |
383 | } | |
384 | ||
79e0e005 DM |
385 | if (-f "$testdir/service_config") { |
386 | copy("$testdir/service_config", "$statusdir/service_config"); | |
387 | } else { | |
388 | my $conf = { | |
eda9314d DM |
389 | 'vm:101' => { node => 'node1', group => 'prefer_node1' }, |
390 | 'vm:102' => { node => 'node2', group => 'prefer_node2' }, | |
391 | 'vm:103' => { node => 'node3', group => 'prefer_node3' }, | |
392 | 'vm:104' => { node => 'node1', group => 'prefer_node1' }, | |
393 | 'vm:105' => { node => 'node2', group => 'prefer_node2' }, | |
394 | 'vm:106' => { node => 'node3', group => 'prefer_node3' }, | |
79e0e005 DM |
395 | }; |
396 | $self->write_service_config($conf); | |
397 | } | |
787b66eb | 398 | |
853f5867 DM |
399 | if (-f "$testdir/hardware_status") { |
400 | copy("$testdir/hardware_status", "$statusdir/hardware_status") || | |
401 | die "Copy failed: $!\n"; | |
402 | } else { | |
403 | my $cstatus = { | |
404 | node1 => { power => 'off', network => 'off' }, | |
405 | node2 => { power => 'off', network => 'off' }, | |
406 | node3 => { power => 'off', network => 'off' }, | |
407 | }; | |
408 | $self->write_hardware_status_nolock($cstatus); | |
409 | } | |
787b66eb | 410 | |
c982dfee TL |
411 | if (-f "$testdir/fence.cfg") { |
412 | copy("$testdir/fence.cfg", "$statusdir/fence.cfg"); | |
413 | } | |
0cfd8f5b | 414 | |
ed408b44 TL |
415 | if (-f "$testdir/datacenter.cfg") { |
416 | copy("$testdir/datacenter.cfg", "$statusdir/datacenter.cfg"); | |
417 | } | |
418 | ||
eea0c609 FE |
419 | if (-f "$testdir/static_service_stats") { |
420 | copy("$testdir/static_service_stats", "$statusdir/static_service_stats"); | |
421 | } | |
422 | ||
8b3f9144 | 423 | my $cstatus = $self->read_hardware_status_nolock(); |
0cfd8f5b DM |
424 | |
425 | foreach my $node (sort keys %$cstatus) { | |
0bba8f60 | 426 | $self->{nodes}->{$node} = {}; |
8456bde2 DM |
427 | |
428 | if (-f "$testdir/service_status_$node") { | |
429 | copy("$testdir/service_status_$node", "$statusdir/service_status_$node"); | |
289e4784 | 430 | } else { |
8456bde2 DM |
431 | $self->write_service_status($node, {}); |
432 | } | |
0cfd8f5b DM |
433 | } |
434 | ||
95360669 DM |
435 | $self->{service_config} = $self->read_service_config(); |
436 | ||
0cfd8f5b DM |
437 | return $self; |
438 | } | |
439 | ||
440 | sub get_time { | |
441 | my ($self) = @_; | |
442 | ||
bf93e2a2 | 443 | die "implement in subclass"; |
0cfd8f5b DM |
444 | } |
445 | ||
446 | sub log { | |
fde8362a | 447 | my ($self, $level, $msg, $id) = @_; |
0cfd8f5b DM |
448 | |
449 | chomp $msg; | |
450 | ||
451 | my $time = $self->get_time(); | |
452 | ||
fde8362a DM |
453 | $id = 'hardware' if !$id; |
454 | ||
0bba8f60 | 455 | printf("%-5s %5d %12s: $msg\n", $level, $time, $id); |
0cfd8f5b DM |
456 | } |
457 | ||
458 | sub statusdir { | |
459 | my ($self, $node) = @_; | |
460 | ||
461 | return $self->{statusdir}; | |
462 | } | |
463 | ||
ed408b44 TL |
464 | sub read_datacenter_conf { |
465 | my ($self, $node) = @_; | |
466 | ||
467 | my $filename = "$self->{statusdir}/datacenter.cfg"; | |
468 | return PVE::HA::Tools::read_json_from_file($filename, {}); | |
469 | } | |
470 | ||
8b3f9144 | 471 | sub global_lock { |
0cfd8f5b DM |
472 | my ($self, $code, @param) = @_; |
473 | ||
8b3f9144 | 474 | my $lockfile = "$self->{statusdir}/hardware.lck"; |
0cfd8f5b DM |
475 | my $fh = IO::File->new(">>$lockfile") || |
476 | die "unable to open '$lockfile'\n"; | |
477 | ||
478 | my $success; | |
479 | for (;;) { | |
480 | $success = flock($fh, LOCK_EX); | |
481 | if ($success || ($! != EINTR)) { | |
482 | last; | |
483 | } | |
484 | if (!$success) { | |
9de9a6ce | 485 | close($fh); |
63f6a08c | 486 | die "can't acquire lock '$lockfile' - $!\n"; |
0cfd8f5b DM |
487 | } |
488 | } | |
9de9a6ce | 489 | |
0cfd8f5b DM |
490 | my $res; |
491 | ||
9de9a6ce | 492 | eval { $res = &$code($fh, @param) }; |
0cfd8f5b | 493 | my $err = $@; |
289e4784 | 494 | |
0cfd8f5b DM |
495 | close($fh); |
496 | ||
497 | die $err if $err; | |
289e4784 | 498 | |
0cfd8f5b DM |
499 | return $res; |
500 | } | |
501 | ||
8b3f9144 DM |
502 | my $compute_node_info = sub { |
503 | my ($self, $cstatus) = @_; | |
504 | ||
505 | my $node_info = {}; | |
506 | ||
507 | my $node_count = 0; | |
508 | my $online_count = 0; | |
509 | ||
510 | foreach my $node (keys %$cstatus) { | |
511 | my $d = $cstatus->{$node}; | |
512 | ||
513 | my $online = ($d->{power} eq 'on' && $d->{network} eq 'on') ? 1 : 0; | |
514 | $node_info->{$node}->{online} = $online; | |
515 | ||
516 | $node_count++; | |
517 | $online_count++ if $online; | |
518 | } | |
519 | ||
520 | my $quorate = ($online_count > int($node_count/2)) ? 1 : 0; | |
289e4784 | 521 | |
8b3f9144 DM |
522 | if (!$quorate) { |
523 | foreach my $node (keys %$cstatus) { | |
524 | my $d = $cstatus->{$node}; | |
525 | $node_info->{$node}->{online} = 0; | |
526 | } | |
527 | } | |
528 | ||
529 | return ($node_info, $quorate); | |
530 | }; | |
531 | ||
532 | sub get_node_info { | |
533 | my ($self) = @_; | |
534 | ||
5516f102 TL |
535 | my $cstatus = $self->read_hardware_status_nolock(); |
536 | my ($node_info, $quorate) = &$compute_node_info($self, $cstatus); | |
8b3f9144 DM |
537 | |
538 | return ($node_info, $quorate); | |
539 | } | |
540 | ||
ba2a45cd TL |
541 | # helper for Sim/ only |
542 | sub get_cfs_state { | |
543 | my ($self, $node, $state) = @_; | |
544 | ||
545 | # TODO: ensure nolock is OK when adding this to RTSim | |
546 | my $cstatus = $self->read_hardware_status_nolock(); | |
547 | my $res = $cstatus->{$node}->{cfs}->{$state}; | |
548 | ||
549 | # we assume default true if not defined | |
550 | return !defined($res) || $res; | |
551 | } | |
552 | ||
a5d48ae1 TL |
553 | # simulate hardware commands, the following commands are available: |
554 | # power <node> <on|off> | |
555 | # network <node> <on|off> | |
556 | # delay <seconds> | |
1b21e7e6 | 557 | # skip-round <crm|lrm> [<rounds=1>] |
a5d48ae1 TL |
558 | # cfs <node> <rw|update> <work|fail> |
559 | # reboot <node> | |
560 | # shutdown <node> | |
561 | # restart-lrm <node> | |
562 | # service <sid> <started|disabled|stopped|ignored> | |
563 | # service <sid> <migrate|relocate> <target> | |
564 | # service <sid> stop <timeout> | |
565 | # service <sid> lock/unlock [lockname] | |
566 | # service <sid> <add|delete> | |
8b3f9144 | 567 | sub sim_hardware_cmd { |
fde8362a | 568 | my ($self, $cmdstr, $logid) = @_; |
0cfd8f5b | 569 | |
e08a0717 TL |
570 | my $code = sub { |
571 | my ($lock_fh) = @_; | |
572 | ||
573 | my $cstatus = $self->read_hardware_status_nolock(); | |
574 | ||
0e13a6c1 TL |
575 | my ($cmd, $objid, $action, @params) = split(/\s+/, $cmdstr); |
576 | my $param = $params[0]; # for convenience/legacy | |
e08a0717 TL |
577 | |
578 | die "sim_hardware_cmd: no node or service for command specified" | |
579 | if !$objid; | |
580 | ||
581 | my ($node, $sid, $d); | |
582 | ||
583 | if ($cmd eq 'service') { | |
584 | $sid = PVE::HA::Tools::pve_verify_ha_resource_id($objid); | |
585 | } else { | |
586 | $node = $objid; | |
587 | $d = $self->{nodes}->{$node} || | |
588 | die "sim_hardware_cmd: no such node '$node'\n"; | |
589 | } | |
590 | ||
591 | $self->log('info', "execute $cmdstr", $logid); | |
592 | ||
593 | if ($cmd eq 'power') { | |
594 | die "sim_hardware_cmd: unknown action '$action'\n" | |
595 | if $action !~ m/^(on|off)$/; | |
596 | ||
597 | if ($cstatus->{$node}->{power} ne $action) { | |
598 | if ($action eq 'on') { | |
599 | ||
600 | $d->{crm} = $self->crm_control('start', $d, $lock_fh) if !defined($d->{crm}); | |
601 | $d->{lrm} = $self->lrm_control('start', $d, $lock_fh) if !defined($d->{lrm}); | |
602 | $d->{lrm_restart} = undef; | |
ba2a45cd | 603 | $cstatus->{$node}->{cfs} = {}; |
e08a0717 TL |
604 | |
605 | } else { | |
606 | ||
607 | if ($d->{crm}) { | |
608 | $d->{crm_env}->log('info', "killed by poweroff"); | |
609 | $self->crm_control('stop', $d, $lock_fh); | |
610 | $d->{crm} = undef; | |
611 | } | |
612 | if ($d->{lrm}) { | |
613 | $d->{lrm_env}->log('info', "killed by poweroff"); | |
614 | $self->lrm_control('stop', $d, $lock_fh); | |
615 | $d->{lrm} = undef; | |
616 | $d->{lrm_restart} = undef; | |
617 | } | |
618 | ||
619 | $self->watchdog_reset_nolock($node); | |
620 | $self->write_service_status($node, {}); | |
621 | } | |
622 | } | |
623 | ||
624 | $cstatus->{$node}->{power} = $action; | |
625 | $cstatus->{$node}->{network} = $action; | |
626 | $cstatus->{$node}->{shutdown} = undef; | |
627 | ||
628 | $self->write_hardware_status_nolock($cstatus); | |
629 | ||
630 | } elsif ($cmd eq 'network') { | |
631 | die "sim_hardware_cmd: unknown network action '$action'" | |
632 | if $action !~ m/^(on|off)$/; | |
633 | $cstatus->{$node}->{network} = $action; | |
634 | ||
635 | $self->write_hardware_status_nolock($cstatus); | |
636 | ||
ba2a45cd TL |
637 | } elsif ($cmd eq 'cfs') { |
638 | die "sim_hardware_cmd: unknown cfs action '$action' for node '$node'" | |
639 | if $action !~ m/^(rw|update)$/; | |
b94b4785 FE |
640 | die "sim_hardware_cmd: unknown cfs command '$param' for '$action' on node '$node'" |
641 | if $param !~ m/^(work|fail)$/; | |
ba2a45cd | 642 | |
b94b4785 | 643 | $cstatus->{$node}->{cfs}->{$action} = $param eq 'work'; |
ba2a45cd TL |
644 | $self->write_hardware_status_nolock($cstatus); |
645 | ||
e08a0717 TL |
646 | } elsif ($cmd eq 'reboot' || $cmd eq 'shutdown') { |
647 | $cstatus->{$node}->{shutdown} = $cmd; | |
648 | ||
649 | $self->write_hardware_status_nolock($cstatus); | |
650 | ||
651 | $self->lrm_control('shutdown', $d, $lock_fh) if defined($d->{lrm}); | |
652 | } elsif ($cmd eq 'restart-lrm') { | |
653 | if ($d->{lrm}) { | |
654 | $d->{lrm_restart} = 1; | |
655 | $self->lrm_control('shutdown', $d, $lock_fh); | |
656 | } | |
657 | } elsif ($cmd eq 'crm') { | |
658 | ||
659 | if ($action eq 'stop') { | |
660 | if ($d->{crm}) { | |
661 | $d->{crm_stop} = 1; | |
662 | $self->crm_control('shutdown', $d, $lock_fh); | |
663 | } | |
664 | } elsif ($action eq 'start') { | |
665 | $d->{crm} = $self->crm_control('start', $d, $lock_fh) if !defined($d->{crm}); | |
666 | } else { | |
667 | die "sim_hardware_cmd: unknown action '$action'"; | |
668 | } | |
669 | ||
670 | } elsif ($cmd eq 'service') { | |
667670b2 TL |
671 | if ($action eq 'started' || $action eq 'disabled' || |
672 | $action eq 'stopped' || $action eq 'ignored') { | |
e08a0717 TL |
673 | |
674 | $self->set_service_state($sid, $action); | |
675 | ||
676 | } elsif ($action eq 'migrate' || $action eq 'relocate') { | |
677 | ||
678 | die "sim_hardware_cmd: missing target node for '$action' command" | |
b94b4785 | 679 | if !$param; |
e08a0717 | 680 | |
b94b4785 | 681 | $self->queue_crm_commands_nolock("$action $sid $param"); |
e08a0717 | 682 | |
21caf0db FE |
683 | } elsif ($action eq 'stop') { |
684 | ||
685 | die "sim_hardware_cmd: missing timeout for '$action' command" | |
686 | if !defined($param); | |
687 | ||
688 | $self->queue_crm_commands_nolock("$action $sid $param"); | |
689 | ||
e08a0717 TL |
690 | } elsif ($action eq 'add') { |
691 | ||
0e13a6c1 | 692 | $self->add_service($sid, {state => $params[1] || 'started', node => $param}); |
e08a0717 TL |
693 | |
694 | } elsif ($action eq 'delete') { | |
695 | ||
696 | $self->delete_service($sid); | |
697 | ||
698 | } elsif ($action eq 'lock') { | |
699 | ||
b94b4785 | 700 | $self->lock_service($sid, $param); |
e08a0717 TL |
701 | |
702 | } elsif ($action eq 'unlock') { | |
703 | ||
b94b4785 | 704 | $self->unlock_service($sid, $param); |
e08a0717 TL |
705 | |
706 | } else { | |
707 | die "sim_hardware_cmd: unknown service action '$action' " . | |
708 | "- not implemented\n" | |
709 | } | |
710 | } else { | |
711 | die "sim_hardware_cmd: unknown command '$cmdstr'\n"; | |
712 | } | |
713 | ||
714 | return $cstatus; | |
715 | }; | |
716 | ||
717 | return $self->global_lock($code); | |
718 | } | |
719 | ||
720 | # for controlling the resource manager services | |
721 | sub crm_control { | |
722 | my ($self, $action, $data, $lock_fh) = @_; | |
723 | ||
724 | die "implement in subclass"; | |
725 | } | |
726 | ||
727 | sub lrm_control { | |
728 | my ($self, $action, $data, $lock_fh) = @_; | |
729 | ||
bf93e2a2 | 730 | die "implement in subclass"; |
0cfd8f5b DM |
731 | } |
732 | ||
733 | sub run { | |
734 | my ($self) = @_; | |
735 | ||
bf93e2a2 | 736 | die "implement in subclass"; |
0cfd8f5b | 737 | } |
9329c1e2 DM |
738 | |
739 | my $modify_watchog = sub { | |
740 | my ($self, $code) = @_; | |
741 | ||
742 | my $update_cmd = sub { | |
743 | ||
744 | my $filename = "$self->{statusdir}/watchdog_status"; | |
289e4784 | 745 | |
9329c1e2 DM |
746 | my ($res, $wdstatus); |
747 | ||
748 | if (-f $filename) { | |
749 | my $raw = PVE::Tools::file_get_contents($filename); | |
750 | $wdstatus = decode_json($raw); | |
751 | } else { | |
752 | $wdstatus = {}; | |
753 | } | |
289e4784 | 754 | |
9329c1e2 DM |
755 | ($wdstatus, $res) = &$code($wdstatus); |
756 | ||
757 | PVE::Tools::file_set_contents($filename, encode_json($wdstatus)); | |
758 | ||
759 | return $res; | |
760 | }; | |
761 | ||
762 | return $self->global_lock($update_cmd); | |
763 | }; | |
764 | ||
0590c6a7 DM |
765 | sub watchdog_reset_nolock { |
766 | my ($self, $node) = @_; | |
767 | ||
768 | my $filename = "$self->{statusdir}/watchdog_status"; | |
769 | ||
770 | if (-f $filename) { | |
771 | my $raw = PVE::Tools::file_get_contents($filename); | |
772 | my $wdstatus = decode_json($raw); | |
773 | ||
774 | foreach my $id (keys %$wdstatus) { | |
775 | delete $wdstatus->{$id} if $wdstatus->{$id}->{node} eq $node; | |
776 | } | |
289e4784 | 777 | |
0590c6a7 DM |
778 | PVE::Tools::file_set_contents($filename, encode_json($wdstatus)); |
779 | } | |
780 | } | |
781 | ||
9329c1e2 DM |
782 | sub watchdog_check { |
783 | my ($self, $node) = @_; | |
784 | ||
785 | my $code = sub { | |
786 | my ($wdstatus) = @_; | |
787 | ||
788 | my $res = 1; | |
789 | ||
790 | foreach my $wfh (keys %$wdstatus) { | |
791 | my $wd = $wdstatus->{$wfh}; | |
792 | next if $wd->{node} ne $node; | |
793 | ||
794 | my $ctime = $self->get_time(); | |
795 | my $tdiff = $ctime - $wd->{update_time}; | |
796 | ||
0bba8f60 | 797 | if ($tdiff > $watchdog_timeout) { # expired |
9329c1e2 DM |
798 | $res = 0; |
799 | delete $wdstatus->{$wfh}; | |
800 | } | |
801 | } | |
289e4784 | 802 | |
9329c1e2 DM |
803 | return ($wdstatus, $res); |
804 | }; | |
805 | ||
806 | return &$modify_watchog($self, $code); | |
807 | } | |
808 | ||
809 | my $wdcounter = 0; | |
810 | ||
811 | sub watchdog_open { | |
812 | my ($self, $node) = @_; | |
813 | ||
814 | my $code = sub { | |
815 | my ($wdstatus) = @_; | |
816 | ||
817 | ++$wdcounter; | |
818 | ||
819 | my $id = "WD:$node:$$:$wdcounter"; | |
820 | ||
821 | die "internal error" if defined($wdstatus->{$id}); | |
822 | ||
823 | $wdstatus->{$id} = { | |
824 | node => $node, | |
825 | update_time => $self->get_time(), | |
826 | }; | |
827 | ||
828 | return ($wdstatus, $id); | |
829 | }; | |
830 | ||
831 | return &$modify_watchog($self, $code); | |
832 | } | |
833 | ||
834 | sub watchdog_close { | |
835 | my ($self, $wfh) = @_; | |
836 | ||
837 | my $code = sub { | |
838 | my ($wdstatus) = @_; | |
839 | ||
840 | my $wd = $wdstatus->{$wfh}; | |
841 | die "no such watchdog handle '$wfh'\n" if !defined($wd); | |
842 | ||
843 | my $tdiff = $self->get_time() - $wd->{update_time}; | |
0bba8f60 | 844 | die "watchdog expired" if $tdiff > $watchdog_timeout; |
9329c1e2 DM |
845 | |
846 | delete $wdstatus->{$wfh}; | |
847 | ||
848 | return ($wdstatus); | |
849 | }; | |
850 | ||
851 | return &$modify_watchog($self, $code); | |
852 | } | |
853 | ||
854 | sub watchdog_update { | |
855 | my ($self, $wfh) = @_; | |
856 | ||
857 | my $code = sub { | |
858 | my ($wdstatus) = @_; | |
859 | ||
860 | my $wd = $wdstatus->{$wfh}; | |
861 | ||
862 | die "no such watchdog handle '$wfh'\n" if !defined($wd); | |
863 | ||
864 | my $ctime = $self->get_time(); | |
865 | my $tdiff = $ctime - $wd->{update_time}; | |
866 | ||
0bba8f60 | 867 | die "watchdog expired" if $tdiff > $watchdog_timeout; |
289e4784 | 868 | |
9329c1e2 DM |
869 | $wd->{update_time} = $ctime; |
870 | ||
871 | return ($wdstatus); | |
872 | }; | |
873 | ||
874 | return &$modify_watchog($self, $code); | |
875 | } | |
876 | ||
5db695c3 FE |
877 | sub get_static_node_stats { |
878 | my ($self) = @_; | |
879 | ||
880 | my $cstatus = $self->read_hardware_status_nolock(); | |
881 | ||
882 | my $stats = {}; | |
883 | for my $node (keys $cstatus->%*) { | |
884 | $stats->{$node} = { $cstatus->{$node}->%{qw(cpus memory)} }; | |
885 | } | |
886 | ||
887 | return $stats; | |
888 | } | |
889 | ||
0cfd8f5b | 890 | 1; |