]>
Commit | Line | Data |
---|---|---|
714a4016 DM |
1 | package PVE::HA::Env::PVE2; |
2 | ||
3 | use strict; | |
4 | use warnings; | |
76737af5 DM |
5 | use POSIX qw(:errno_h :fcntl_h); |
6 | use IO::File; | |
115805fd | 7 | use IO::Socket::UNIX; |
5db695c3 | 8 | use JSON; |
714a4016 DM |
9 | |
10 | use PVE::SafeSyslog; | |
11 | use PVE::Tools; | |
119656b9 | 12 | use PVE::Cluster qw(cfs_register_file cfs_read_file cfs_write_file cfs_lock_file); |
ef39a1ca | 13 | use PVE::DataCenterConfig; |
022e4e79 DM |
14 | use PVE::INotify; |
15 | use PVE::RPCEnvironment; | |
714a4016 | 16 | |
a89ff919 | 17 | use PVE::HA::Tools ':exit_codes'; |
714a4016 | 18 | use PVE::HA::Env; |
ce216792 | 19 | use PVE::HA::Config; |
c982dfee | 20 | use PVE::HA::FenceConfig; |
9e5ea8f7 DM |
21 | use PVE::HA::Resources; |
22 | use PVE::HA::Resources::PVEVM; | |
23 | use PVE::HA::Resources::PVECT; | |
714a4016 | 24 | |
9e5ea8f7 DM |
25 | PVE::HA::Resources::PVEVM->register(); |
26 | PVE::HA::Resources::PVECT->register(); | |
27 | ||
28 | PVE::HA::Resources->init(); | |
022e4e79 | 29 | |
007fcc8b DM |
30 | my $lockdir = "/etc/pve/priv/lock"; |
31 | ||
714a4016 DM |
32 | sub new { |
33 | my ($this, $nodename) = @_; | |
34 | ||
35 | die "missing nodename" if !$nodename; | |
36 | ||
37 | my $class = ref($this) || $this; | |
38 | ||
39 | my $self = bless {}, $class; | |
40 | ||
41 | $self->{nodename} = $nodename; | |
42 | ||
43 | return $self; | |
44 | } | |
45 | ||
46 | sub nodename { | |
47 | my ($self) = @_; | |
48 | ||
49 | return $self->{nodename}; | |
50 | } | |
51 | ||
dd9c0c9d TL |
52 | sub hardware { |
53 | my ($self) = @_; | |
54 | ||
55 | die "hardware is for testing and simulation only"; | |
56 | } | |
57 | ||
714a4016 DM |
58 | sub read_manager_status { |
59 | my ($self) = @_; | |
714a4016 | 60 | |
139a9b90 | 61 | return PVE::HA::Config::read_manager_status(); |
714a4016 DM |
62 | } |
63 | ||
64 | sub write_manager_status { | |
65 | my ($self, $status_obj) = @_; | |
63f6a08c | 66 | |
139a9b90 | 67 | PVE::HA::Config::write_manager_status($status_obj); |
714a4016 DM |
68 | } |
69 | ||
c4a221bc DM |
70 | sub read_lrm_status { |
71 | my ($self, $node) = @_; | |
72 | ||
73 | $node = $self->{nodename} if !defined($node); | |
74 | ||
139a9b90 | 75 | return PVE::HA::Config::read_lrm_status($node); |
c4a221bc DM |
76 | } |
77 | ||
78 | sub write_lrm_status { | |
79 | my ($self, $status_obj) = @_; | |
80 | ||
6cbcb5f7 | 81 | my $node = $self->{nodename}; |
63f6a08c | 82 | |
139a9b90 DM |
83 | PVE::HA::Config::write_lrm_status($node, $status_obj); |
84 | } | |
c4a221bc | 85 | |
cde77779 | 86 | sub is_node_shutdown { |
d42219a3 TL |
87 | my ($self) = @_; |
88 | ||
cde77779 | 89 | my $shutdown = 0; |
f65f41b9 | 90 | my $reboot = 0; |
d42219a3 TL |
91 | |
92 | my $code = sub { | |
93 | my $line = shift; | |
94 | ||
61ae38eb | 95 | # ensure we match the full unit name by matching /^JOB_ID UNIT / |
f65f41b9 TL |
96 | # see: man systemd.special |
97 | $shutdown = 1 if ($line =~ m/^\d+\s+shutdown\.target\s+/); | |
98 | $reboot = 1 if ($line =~ m/^\d+\s+reboot\.target\s+/); | |
d42219a3 TL |
99 | }; |
100 | ||
61ae38eb | 101 | my $cmd = ['/bin/systemctl', '--full', 'list-jobs']; |
d42219a3 TL |
102 | eval { PVE::Tools::run_command($cmd, outfunc => $code, noerr => 1); }; |
103 | ||
f65f41b9 | 104 | return ($shutdown, $reboot); |
d42219a3 TL |
105 | } |
106 | ||
139a9b90 DM |
107 | sub queue_crm_commands { |
108 | my ($self, $cmd) = @_; | |
c4a221bc | 109 | |
139a9b90 DM |
110 | return PVE::HA::Config::queue_crm_commands($cmd); |
111 | } | |
112 | ||
113 | sub read_crm_commands { | |
114 | my ($self) = @_; | |
115 | ||
116 | return PVE::HA::Config::read_crm_commands(); | |
c4a221bc DM |
117 | } |
118 | ||
b83b4ae8 DM |
119 | sub read_service_config { |
120 | my ($self) = @_; | |
ce216792 | 121 | |
85f6e9ca | 122 | return PVE::HA::Config::read_and_check_resources_config(); |
714a4016 DM |
123 | } |
124 | ||
76b83c72 FE |
125 | sub update_service_config { |
126 | my ($self, $sid, $param) = @_; | |
127 | ||
128 | return PVE::HA::Config::update_resources_config($sid, $param); | |
129 | } | |
130 | ||
0087839a FG |
131 | sub parse_sid { |
132 | my ($self, $sid) = @_; | |
133 | ||
134 | return PVE::HA::Config::parse_sid($sid); | |
135 | } | |
136 | ||
c982dfee TL |
137 | sub read_fence_config { |
138 | my ($self) = @_; | |
139 | ||
140 | return PVE::HA::Config::read_fence_config(); | |
141 | } | |
142 | ||
143 | sub fencing_mode { | |
144 | my ($self) = @_; | |
145 | ||
146 | my $datacenterconfig = cfs_read_file('datacenter.cfg'); | |
147 | ||
148 | return 'watchdog' if !$datacenterconfig->{fencing}; | |
149 | ||
150 | return $datacenterconfig->{fencing}; | |
151 | } | |
152 | ||
153 | sub exec_fence_agent { | |
154 | my ($self, $agent, $node, @param) = @_; | |
155 | ||
156 | # setup execution environment | |
157 | $ENV{'PATH'} = '/sbin:/bin:/usr/sbin:/usr/bin'; | |
158 | ||
159 | my $cmd = "$agent " . PVE::HA::FenceConfig::gen_arg_str(@param); | |
160 | ||
161 | exec($cmd); | |
162 | exit -1; | |
163 | } | |
164 | ||
9da84a0d TL |
165 | # this is only allowed by the master to recover a _fenced_ service |
166 | sub steal_service { | |
6da27e23 | 167 | my ($self, $sid, $current_node, $new_node) = @_; |
8456bde2 | 168 | |
0354cbe9 | 169 | my (undef, $type, $name) = PVE::HA::Config::parse_sid($sid); |
6da27e23 | 170 | |
303a08aa TL |
171 | if(my $plugin = PVE::HA::Resources->lookup($type)) { |
172 | my $old = $plugin->config_file($name, $current_node); | |
173 | my $new = $plugin->config_file($name, $new_node); | |
6da27e23 DM |
174 | rename($old, $new) || |
175 | die "rename '$old' to '$new' failed - $!\n"; | |
176 | } else { | |
177 | die "implement me"; | |
178 | } | |
48f2144b FE |
179 | |
180 | # Necessary for (at least) static usage plugin to always be able to read service config from new | |
181 | # node right away. | |
182 | $self->cluster_state_update(); | |
8456bde2 DM |
183 | } |
184 | ||
abc920b4 DM |
185 | sub read_group_config { |
186 | my ($self) = @_; | |
187 | ||
139a9b90 | 188 | return PVE::HA::Config::read_group_config(); |
3b996922 DM |
189 | } |
190 | ||
714a4016 DM |
191 | # this should return a hash containing info |
192 | # what nodes are members and online. | |
193 | sub get_node_info { | |
194 | my ($self) = @_; | |
195 | ||
d706ef8b | 196 | my ($node_info, $quorate) = ({}, 0); |
63f6a08c | 197 | |
d706ef8b DM |
198 | my $nodename = $self->{nodename}; |
199 | ||
200 | $quorate = PVE::Cluster::check_cfs_quorum(1) || 0; | |
201 | ||
202 | my $members = PVE::Cluster::get_members(); | |
203 | ||
204 | foreach my $node (keys %$members) { | |
205 | my $d = $members->{$node}; | |
63f6a08c | 206 | $node_info->{$node}->{online} = $d->{online}; |
d706ef8b | 207 | } |
63f6a08c | 208 | |
d706ef8b | 209 | $node_info->{$nodename}->{online} = 1; # local node is always up |
63f6a08c | 210 | |
d706ef8b | 211 | return ($node_info, $quorate); |
714a4016 DM |
212 | } |
213 | ||
214 | sub log { | |
215 | my ($self, $level, $msg) = @_; | |
216 | ||
217 | chomp $msg; | |
218 | ||
219 | syslog($level, $msg); | |
220 | } | |
221 | ||
1b3969b6 TL |
222 | sub sendmail { |
223 | my ($self, $subject, $text) = @_; | |
224 | ||
227c2c74 PA |
225 | # Leave it to postfix to append the correct hostname |
226 | my $mailfrom = 'root'; | |
227 | # /root/.forward makes pvemailforward redirect the | |
228 | # mail to the address configured in the datacenter | |
229 | my $mailto = 'root'; | |
1b3969b6 TL |
230 | |
231 | PVE::Tools::sendmail($mailto, $subject, $text, undef, $mailfrom); | |
232 | } | |
233 | ||
d69a79f3 | 234 | my $last_lock_status_hash = {}; |
007fcc8b DM |
235 | |
236 | sub get_pve_lock { | |
237 | my ($self, $lockid) = @_; | |
714a4016 | 238 | |
007fcc8b | 239 | my $got_lock = 0; |
4d24e7db | 240 | |
4d24e7db DM |
241 | my $filename = "$lockdir/$lockid"; |
242 | ||
d69a79f3 DM |
243 | $last_lock_status_hash->{$lockid} //= { lock_time => 0, got_lock => 0}; |
244 | my $last = $last_lock_status_hash->{$lockid}; | |
007fcc8b DM |
245 | |
246 | my $ctime = time(); | |
d69a79f3 DM |
247 | my $last_lock_time = $last->{lock_time} // 0; |
248 | my $last_got_lock = $last->{got_lock}; | |
4d24e7db | 249 | |
5d2406c9 | 250 | my $retry_timeout = 120; # hardcoded lock lifetime limit from pmxcfs |
63f6a08c | 251 | |
4d24e7db DM |
252 | eval { |
253 | ||
254 | mkdir $lockdir; | |
255 | ||
007fcc8b DM |
256 | # pve cluster filesystem not online |
257 | die "can't create '$lockdir' (pmxcfs not mounted?)\n" if ! -d $lockdir; | |
258 | ||
cb0bac5e | 259 | if (($ctime - $last_lock_time) < $retry_timeout) { |
737abf2f DM |
260 | # try cfs lock update request (utime) |
261 | if (utime(0, $ctime, $filename)) { | |
262 | $got_lock = 1; | |
263 | return; | |
264 | } | |
d69a79f3 DM |
265 | die "cfs lock update failed - $!\n"; |
266 | } | |
007fcc8b | 267 | |
d69a79f3 DM |
268 | if (mkdir $filename) { |
269 | $got_lock = 1; | |
270 | return; | |
007fcc8b | 271 | } |
4d24e7db | 272 | |
d69a79f3 DM |
273 | utime 0, 0, $filename; # cfs unlock request |
274 | die "can't get cfs lock\n"; | |
4d24e7db DM |
275 | }; |
276 | ||
007fcc8b DM |
277 | my $err = $@; |
278 | ||
d69a79f3 | 279 | #$self->log('err', $err) if $err; # for debugging |
63f6a08c | 280 | |
d69a79f3 DM |
281 | $last->{got_lock} = $got_lock; |
282 | $last->{lock_time} = $ctime if $got_lock; | |
007fcc8b | 283 | |
d69a79f3 | 284 | if (!!$got_lock != !!$last_got_lock) { |
007fcc8b | 285 | if ($got_lock) { |
63f6a08c | 286 | $self->log('info', "successfully acquired lock '$lockid'"); |
007fcc8b DM |
287 | } else { |
288 | my $msg = "lost lock '$lockid"; | |
63f6a08c | 289 | $msg .= " - $err" if $err; |
007fcc8b DM |
290 | $self->log('err', $msg); |
291 | } | |
292 | } | |
293 | ||
294 | return $got_lock; | |
295 | } | |
296 | ||
297 | sub get_ha_manager_lock { | |
298 | my ($self) = @_; | |
299 | ||
007fcc8b | 300 | return $self->get_pve_lock("ha_manager_lock"); |
714a4016 DM |
301 | } |
302 | ||
de002253 TL |
303 | # release the cluster wide manager lock. |
304 | # when released another CRM may step up and get the lock, thus this should only | |
305 | # get called when shutting down/deactivating the current master | |
306 | sub release_ha_manager_lock { | |
307 | my ($self) = @_; | |
308 | ||
309 | return rmdir("$lockdir/ha_manager_lock"); | |
310 | } | |
311 | ||
714a4016 | 312 | sub get_ha_agent_lock { |
714a4016 | 313 | my ($self, $node) = @_; |
63f6a08c | 314 | |
f5c29173 | 315 | $node = $self->nodename() if !defined($node); |
714a4016 | 316 | |
f5c29173 | 317 | return $self->get_pve_lock("ha_agent_${node}_lock"); |
714a4016 DM |
318 | } |
319 | ||
ff165cd8 TL |
320 | # release the respective node agent lock. |
321 | # this should only get called if the nodes LRM gracefully shuts down with | |
322 | # all services already cleanly stopped! | |
323 | sub release_ha_agent_lock { | |
324 | my ($self) = @_; | |
325 | ||
326 | my $node = $self->nodename(); | |
327 | ||
328 | return rmdir("$lockdir/ha_agent_${node}_lock"); | |
329 | } | |
330 | ||
714a4016 DM |
331 | sub quorate { |
332 | my ($self) = @_; | |
333 | ||
4d24e7db | 334 | my $quorate = 0; |
63f6a08c TL |
335 | eval { |
336 | $quorate = PVE::Cluster::check_cfs_quorum(); | |
4d24e7db | 337 | }; |
63f6a08c | 338 | |
4d24e7db | 339 | return $quorate; |
714a4016 DM |
340 | } |
341 | ||
342 | sub get_time { | |
343 | my ($self) = @_; | |
344 | ||
345 | return time(); | |
346 | } | |
347 | ||
348 | sub sleep { | |
349 | my ($self, $delay) = @_; | |
350 | ||
351 | CORE::sleep($delay); | |
352 | } | |
353 | ||
354 | sub sleep_until { | |
355 | my ($self, $end_time) = @_; | |
356 | ||
357 | for (;;) { | |
358 | my $cur_time = time(); | |
359 | ||
360 | last if $cur_time >= $end_time; | |
361 | ||
362 | $self->sleep(1); | |
363 | } | |
364 | } | |
365 | ||
366 | sub loop_start_hook { | |
367 | my ($self) = @_; | |
368 | ||
369 | $self->{loop_start} = $self->get_time(); | |
3df15380 | 370 | |
714a4016 DM |
371 | } |
372 | ||
373 | sub loop_end_hook { | |
374 | my ($self) = @_; | |
375 | ||
376 | my $delay = $self->get_time() - $self->{loop_start}; | |
63f6a08c | 377 | |
714a4016 DM |
378 | warn "loop take too long ($delay seconds)\n" if $delay > 30; |
379 | } | |
380 | ||
3df15380 TL |
381 | sub cluster_state_update { |
382 | my ($self) = @_; | |
383 | ||
384 | eval { PVE::Cluster::cfs_update(1) }; | |
385 | if (my $err = $@) { | |
386 | $self->log('warn', "cluster file system update failed - $err"); | |
387 | return 0; | |
388 | } | |
389 | ||
390 | return 1; | |
391 | } | |
392 | ||
76737af5 DM |
393 | my $watchdog_fh; |
394 | ||
714a4016 DM |
395 | sub watchdog_open { |
396 | my ($self) = @_; | |
397 | ||
76737af5 DM |
398 | die "watchdog already open\n" if defined($watchdog_fh); |
399 | ||
115805fd DM |
400 | $watchdog_fh = IO::Socket::UNIX->new( |
401 | Type => SOCK_STREAM(), | |
402 | Peer => "/run/watchdog-mux.sock") || | |
403 | die "unable to open watchdog socket - $!\n"; | |
63f6a08c | 404 | |
76737af5 | 405 | $self->log('info', "watchdog active"); |
714a4016 DM |
406 | } |
407 | ||
408 | sub watchdog_update { | |
409 | my ($self, $wfh) = @_; | |
410 | ||
76737af5 DM |
411 | my $res = $watchdog_fh->syswrite("\0", 1); |
412 | if (!defined($res)) { | |
413 | $self->log('err', "watchdog update failed - $!\n"); | |
414 | return 0; | |
415 | } | |
416 | if ($res != 1) { | |
417 | $self->log('err', "watchdog update failed - write $res bytes\n"); | |
418 | return 0; | |
419 | } | |
420 | ||
421 | return 1; | |
714a4016 DM |
422 | } |
423 | ||
424 | sub watchdog_close { | |
425 | my ($self, $wfh) = @_; | |
426 | ||
76737af5 DM |
427 | $watchdog_fh->syswrite("V", 1); # magic watchdog close |
428 | if (!$watchdog_fh->close()) { | |
429 | $self->log('err', "watchdog close failed - $!"); | |
430 | } else { | |
431 | $watchdog_fh = undef; | |
432 | $self->log('info', "watchdog closed (disabled)"); | |
433 | } | |
714a4016 DM |
434 | } |
435 | ||
a2aae08a TL |
436 | sub after_fork { |
437 | my ($self) = @_; | |
438 | ||
439 | # close inherited inotify FD from parent and reopen our own | |
440 | PVE::INotify::inotify_close(); | |
441 | PVE::INotify::inotify_init(); | |
442 | ||
443 | PVE::Cluster::cfs_update(); | |
444 | } | |
445 | ||
a28fa330 TL |
446 | sub get_max_workers { |
447 | my ($self) = @_; | |
448 | ||
449 | my $datacenterconfig = cfs_read_file('datacenter.cfg'); | |
450 | ||
451 | return $datacenterconfig->{max_workers} || 4; | |
452 | } | |
453 | ||
ed408b44 | 454 | # return cluster wide enforced HA settings |
749d8161 | 455 | sub get_datacenter_settings { |
ed408b44 TL |
456 | my ($self) = @_; |
457 | ||
f3e2a4f2 | 458 | my $datacenterconfig = eval { cfs_read_file('datacenter.cfg') }; |
7c142d68 | 459 | $self->log('err', "unable to get HA settings from datacenter.cfg - $@") if $@; |
ed408b44 | 460 | |
7c142d68 FE |
461 | return { |
462 | ha => $datacenterconfig->{ha} // {}, | |
463 | crs => $datacenterconfig->{crs} // {}, | |
464 | }; | |
ed408b44 TL |
465 | } |
466 | ||
5db695c3 FE |
467 | sub get_static_node_stats { |
468 | my ($self) = @_; | |
469 | ||
470 | my $stats = PVE::Cluster::get_node_kv('static-info'); | |
471 | for my $node (keys $stats->%*) { | |
472 | $stats->{$node} = eval { decode_json($stats->{$node}) }; | |
473 | $self->log('err', "unable to decode static node info for '$node' - $@") if $@; | |
474 | } | |
475 | ||
476 | return $stats; | |
477 | } | |
478 | ||
714a4016 | 479 | 1; |