]>
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; |
714a4016 DM |
8 | |
9 | use PVE::SafeSyslog; | |
10 | use PVE::Tools; | |
119656b9 | 11 | use PVE::Cluster qw(cfs_register_file cfs_read_file cfs_write_file cfs_lock_file); |
022e4e79 DM |
12 | use PVE::INotify; |
13 | use PVE::RPCEnvironment; | |
714a4016 | 14 | |
a89ff919 | 15 | use PVE::HA::Tools ':exit_codes'; |
714a4016 | 16 | use PVE::HA::Env; |
ce216792 | 17 | use PVE::HA::Config; |
714a4016 | 18 | |
022e4e79 | 19 | |
007fcc8b DM |
20 | my $lockdir = "/etc/pve/priv/lock"; |
21 | ||
714a4016 DM |
22 | sub new { |
23 | my ($this, $nodename) = @_; | |
24 | ||
25 | die "missing nodename" if !$nodename; | |
26 | ||
27 | my $class = ref($this) || $this; | |
28 | ||
29 | my $self = bless {}, $class; | |
30 | ||
31 | $self->{nodename} = $nodename; | |
32 | ||
33 | return $self; | |
34 | } | |
35 | ||
36 | sub nodename { | |
37 | my ($self) = @_; | |
38 | ||
39 | return $self->{nodename}; | |
40 | } | |
41 | ||
42 | sub read_manager_status { | |
43 | my ($self) = @_; | |
714a4016 | 44 | |
139a9b90 | 45 | return PVE::HA::Config::read_manager_status(); |
714a4016 DM |
46 | } |
47 | ||
48 | sub write_manager_status { | |
49 | my ($self, $status_obj) = @_; | |
63f6a08c | 50 | |
139a9b90 | 51 | PVE::HA::Config::write_manager_status($status_obj); |
714a4016 DM |
52 | } |
53 | ||
c4a221bc DM |
54 | sub read_lrm_status { |
55 | my ($self, $node) = @_; | |
56 | ||
57 | $node = $self->{nodename} if !defined($node); | |
58 | ||
139a9b90 | 59 | return PVE::HA::Config::read_lrm_status($node); |
c4a221bc DM |
60 | } |
61 | ||
62 | sub write_lrm_status { | |
63 | my ($self, $status_obj) = @_; | |
64 | ||
6cbcb5f7 | 65 | my $node = $self->{nodename}; |
63f6a08c | 66 | |
139a9b90 DM |
67 | PVE::HA::Config::write_lrm_status($node, $status_obj); |
68 | } | |
c4a221bc | 69 | |
cde77779 | 70 | sub is_node_shutdown { |
d42219a3 TL |
71 | my ($self) = @_; |
72 | ||
cde77779 | 73 | my $shutdown = 0; |
d42219a3 TL |
74 | |
75 | my $code = sub { | |
76 | my $line = shift; | |
77 | ||
cde77779 | 78 | $shutdown = 1 if ($line =~ m/shutdown\.target/); |
d42219a3 TL |
79 | }; |
80 | ||
81 | my $cmd = ['/bin/systemctl', 'list-jobs']; | |
82 | eval { PVE::Tools::run_command($cmd, outfunc => $code, noerr => 1); }; | |
83 | ||
cde77779 | 84 | return $shutdown; |
d42219a3 TL |
85 | } |
86 | ||
139a9b90 DM |
87 | sub queue_crm_commands { |
88 | my ($self, $cmd) = @_; | |
c4a221bc | 89 | |
139a9b90 DM |
90 | return PVE::HA::Config::queue_crm_commands($cmd); |
91 | } | |
92 | ||
93 | sub read_crm_commands { | |
94 | my ($self) = @_; | |
95 | ||
96 | return PVE::HA::Config::read_crm_commands(); | |
c4a221bc DM |
97 | } |
98 | ||
b83b4ae8 | 99 | sub service_config_exists { |
714a4016 | 100 | my ($self) = @_; |
63f6a08c | 101 | |
b83b4ae8 DM |
102 | return PVE::HA::Config::resources_config_exists(); |
103 | } | |
714a4016 | 104 | |
b83b4ae8 DM |
105 | sub read_service_config { |
106 | my ($self) = @_; | |
ce216792 | 107 | |
b83b4ae8 | 108 | my $res = PVE::HA::Config::read_resources_config(); |
63f6a08c | 109 | |
ce216792 DM |
110 | my $vmlist = PVE::Cluster::get_vmlist(); |
111 | my $conf = {}; | |
112 | ||
113 | foreach my $sid (keys %{$res->{ids}}) { | |
114 | my $d = $res->{ids}->{$sid}; | |
6ca2edcd | 115 | my (undef, undef, $name) = PVE::HA::Tools::parse_sid($sid); |
7a19642e | 116 | $d->{state} = 'enabled' if !defined($d->{state}); |
ea4443cc TL |
117 | $d->{max_restart} = 1 if !defined($d->{max_restart}); |
118 | $d->{max_relocate} = 1 if !defined($d->{max_relocate}); | |
303a08aa | 119 | if (PVE::HA::Resources->lookup($d->{type})) { |
b47a7a1b | 120 | if (my $vmd = $vmlist->{ids}->{$name}) { |
ce216792 | 121 | if (!$vmd) { |
b47a7a1b | 122 | warn "no such VM '$name'\n"; |
ce216792 DM |
123 | } else { |
124 | $d->{node} = $vmd->{node}; | |
125 | $conf->{$sid} = $d; | |
126 | } | |
127 | } else { | |
128 | if (defined($d->{node})) { | |
129 | $conf->{$sid} = $d; | |
130 | } else { | |
131 | warn "service '$sid' without node\n"; | |
132 | } | |
133 | } | |
134 | } | |
135 | } | |
63f6a08c | 136 | |
ce216792 | 137 | return $conf; |
714a4016 DM |
138 | } |
139 | ||
8456bde2 | 140 | sub change_service_location { |
6da27e23 | 141 | my ($self, $sid, $current_node, $new_node) = @_; |
8456bde2 | 142 | |
6ca2edcd | 143 | my (undef, $type, $name) = PVE::HA::Tools::parse_sid($sid); |
6da27e23 | 144 | |
303a08aa TL |
145 | if(my $plugin = PVE::HA::Resources->lookup($type)) { |
146 | my $old = $plugin->config_file($name, $current_node); | |
147 | my $new = $plugin->config_file($name, $new_node); | |
6da27e23 DM |
148 | rename($old, $new) || |
149 | die "rename '$old' to '$new' failed - $!\n"; | |
150 | } else { | |
151 | die "implement me"; | |
152 | } | |
8456bde2 DM |
153 | } |
154 | ||
abc920b4 DM |
155 | sub read_group_config { |
156 | my ($self) = @_; | |
157 | ||
139a9b90 | 158 | return PVE::HA::Config::read_group_config(); |
3b996922 DM |
159 | } |
160 | ||
714a4016 DM |
161 | # this should return a hash containing info |
162 | # what nodes are members and online. | |
163 | sub get_node_info { | |
164 | my ($self) = @_; | |
165 | ||
d706ef8b | 166 | my ($node_info, $quorate) = ({}, 0); |
63f6a08c | 167 | |
d706ef8b DM |
168 | my $nodename = $self->{nodename}; |
169 | ||
170 | $quorate = PVE::Cluster::check_cfs_quorum(1) || 0; | |
171 | ||
172 | my $members = PVE::Cluster::get_members(); | |
173 | ||
174 | foreach my $node (keys %$members) { | |
175 | my $d = $members->{$node}; | |
63f6a08c | 176 | $node_info->{$node}->{online} = $d->{online}; |
d706ef8b | 177 | } |
63f6a08c | 178 | |
d706ef8b | 179 | $node_info->{$nodename}->{online} = 1; # local node is always up |
63f6a08c | 180 | |
d706ef8b | 181 | return ($node_info, $quorate); |
714a4016 DM |
182 | } |
183 | ||
184 | sub log { | |
185 | my ($self, $level, $msg) = @_; | |
186 | ||
187 | chomp $msg; | |
188 | ||
189 | syslog($level, $msg); | |
190 | } | |
191 | ||
007fcc8b DM |
192 | my $last_lock_status = {}; |
193 | ||
194 | sub get_pve_lock { | |
195 | my ($self, $lockid) = @_; | |
714a4016 | 196 | |
007fcc8b | 197 | my $got_lock = 0; |
4d24e7db | 198 | |
4d24e7db DM |
199 | my $filename = "$lockdir/$lockid"; |
200 | ||
007fcc8b DM |
201 | my $last = $last_lock_status->{$lockid} || 0; |
202 | ||
203 | my $ctime = time(); | |
4d24e7db | 204 | |
75aca181 DM |
205 | my $retry = 0; |
206 | my $retry_timeout = 100; # fixme: what timeout | |
63f6a08c | 207 | |
4d24e7db DM |
208 | eval { |
209 | ||
210 | mkdir $lockdir; | |
211 | ||
007fcc8b DM |
212 | # pve cluster filesystem not online |
213 | die "can't create '$lockdir' (pmxcfs not mounted?)\n" if ! -d $lockdir; | |
214 | ||
75aca181 DM |
215 | if ($last && (($ctime - $last) < $retry_timeout)) { |
216 | # send cfs lock update request (utime) | |
217 | if (!utime(0, $ctime, $filename)) { | |
218 | $retry = 1; | |
007fcc8b | 219 | die "cfs lock update failed - $!\n"; |
75aca181 | 220 | } |
007fcc8b DM |
221 | } else { |
222 | ||
223 | # fixme: wait some time? | |
224 | if (!(mkdir $filename)) { | |
225 | utime 0, 0, $filename; # cfs unlock request | |
226 | die "can't get cfs lock\n"; | |
227 | } | |
228 | } | |
4d24e7db | 229 | |
007fcc8b | 230 | $got_lock = 1; |
4d24e7db DM |
231 | }; |
232 | ||
007fcc8b DM |
233 | my $err = $@; |
234 | ||
75aca181 DM |
235 | if ($retry) { |
236 | # $self->log('err', $err) if $err; # for debugging | |
237 | return 0; | |
238 | } | |
63f6a08c | 239 | |
007fcc8b DM |
240 | $last_lock_status->{$lockid} = $got_lock ? $ctime : 0; |
241 | ||
17e90af6 | 242 | if (!!$got_lock != !!$last) { |
007fcc8b | 243 | if ($got_lock) { |
63f6a08c | 244 | $self->log('info', "successfully acquired lock '$lockid'"); |
007fcc8b DM |
245 | } else { |
246 | my $msg = "lost lock '$lockid"; | |
63f6a08c | 247 | $msg .= " - $err" if $err; |
007fcc8b DM |
248 | $self->log('err', $msg); |
249 | } | |
75aca181 DM |
250 | } else { |
251 | # $self->log('err', $err) if $err; # for debugging | |
007fcc8b DM |
252 | } |
253 | ||
254 | return $got_lock; | |
255 | } | |
256 | ||
257 | sub get_ha_manager_lock { | |
258 | my ($self) = @_; | |
259 | ||
007fcc8b | 260 | return $self->get_pve_lock("ha_manager_lock"); |
714a4016 DM |
261 | } |
262 | ||
263 | sub get_ha_agent_lock { | |
714a4016 | 264 | my ($self, $node) = @_; |
63f6a08c | 265 | |
f5c29173 | 266 | $node = $self->nodename() if !defined($node); |
714a4016 | 267 | |
f5c29173 | 268 | return $self->get_pve_lock("ha_agent_${node}_lock"); |
714a4016 DM |
269 | } |
270 | ||
271 | sub quorate { | |
272 | my ($self) = @_; | |
273 | ||
4d24e7db | 274 | my $quorate = 0; |
63f6a08c TL |
275 | eval { |
276 | $quorate = PVE::Cluster::check_cfs_quorum(); | |
4d24e7db | 277 | }; |
63f6a08c | 278 | |
4d24e7db | 279 | return $quorate; |
714a4016 DM |
280 | } |
281 | ||
282 | sub get_time { | |
283 | my ($self) = @_; | |
284 | ||
285 | return time(); | |
286 | } | |
287 | ||
288 | sub sleep { | |
289 | my ($self, $delay) = @_; | |
290 | ||
291 | CORE::sleep($delay); | |
292 | } | |
293 | ||
294 | sub sleep_until { | |
295 | my ($self, $end_time) = @_; | |
296 | ||
297 | for (;;) { | |
298 | my $cur_time = time(); | |
299 | ||
300 | last if $cur_time >= $end_time; | |
301 | ||
302 | $self->sleep(1); | |
303 | } | |
304 | } | |
305 | ||
306 | sub loop_start_hook { | |
307 | my ($self) = @_; | |
308 | ||
4d24e7db | 309 | PVE::Cluster::cfs_update(); |
63f6a08c | 310 | |
714a4016 DM |
311 | $self->{loop_start} = $self->get_time(); |
312 | } | |
313 | ||
314 | sub loop_end_hook { | |
315 | my ($self) = @_; | |
316 | ||
317 | my $delay = $self->get_time() - $self->{loop_start}; | |
63f6a08c | 318 | |
714a4016 DM |
319 | warn "loop take too long ($delay seconds)\n" if $delay > 30; |
320 | } | |
321 | ||
76737af5 DM |
322 | my $watchdog_fh; |
323 | ||
714a4016 DM |
324 | sub watchdog_open { |
325 | my ($self) = @_; | |
326 | ||
76737af5 DM |
327 | die "watchdog already open\n" if defined($watchdog_fh); |
328 | ||
115805fd DM |
329 | $watchdog_fh = IO::Socket::UNIX->new( |
330 | Type => SOCK_STREAM(), | |
331 | Peer => "/run/watchdog-mux.sock") || | |
332 | die "unable to open watchdog socket - $!\n"; | |
63f6a08c | 333 | |
76737af5 | 334 | $self->log('info', "watchdog active"); |
714a4016 DM |
335 | } |
336 | ||
337 | sub watchdog_update { | |
338 | my ($self, $wfh) = @_; | |
339 | ||
76737af5 DM |
340 | my $res = $watchdog_fh->syswrite("\0", 1); |
341 | if (!defined($res)) { | |
342 | $self->log('err', "watchdog update failed - $!\n"); | |
343 | return 0; | |
344 | } | |
345 | if ($res != 1) { | |
346 | $self->log('err', "watchdog update failed - write $res bytes\n"); | |
347 | return 0; | |
348 | } | |
349 | ||
350 | return 1; | |
714a4016 DM |
351 | } |
352 | ||
353 | sub watchdog_close { | |
354 | my ($self, $wfh) = @_; | |
355 | ||
76737af5 DM |
356 | $watchdog_fh->syswrite("V", 1); # magic watchdog close |
357 | if (!$watchdog_fh->close()) { | |
358 | $self->log('err', "watchdog close failed - $!"); | |
359 | } else { | |
360 | $watchdog_fh = undef; | |
361 | $self->log('info', "watchdog closed (disabled)"); | |
362 | } | |
714a4016 DM |
363 | } |
364 | ||
022e4e79 DM |
365 | sub upid_wait { |
366 | my ($self, $upid) = @_; | |
367 | ||
368 | my $task = PVE::Tools::upid_decode($upid); | |
369 | ||
370 | CORE::sleep(1); | |
371 | while (PVE::ProcFSTools::check_process_running($task->{pid}, $task->{pstart})) { | |
372 | $self->log('debug', "Task still active, waiting"); | |
373 | CORE::sleep(1); | |
374 | } | |
375 | } | |
376 | ||
0d1d32fb DM |
377 | sub can_fork { |
378 | my ($self) = @_; | |
379 | ||
380 | return 1; | |
381 | } | |
382 | ||
c4a221bc | 383 | sub exec_resource_agent { |
6dbf93a0 | 384 | my ($self, $sid, $service_config, $cmd, @params) = @_; |
c4a221bc | 385 | |
022e4e79 | 386 | # setup execution environment |
63f6a08c | 387 | |
022e4e79 DM |
388 | $ENV{'PATH'} = '/sbin:/bin:/usr/sbin:/usr/bin'; |
389 | ||
390 | PVE::INotify::inotify_close(); | |
63f6a08c | 391 | |
022e4e79 DM |
392 | PVE::INotify::inotify_init(); |
393 | ||
394 | PVE::Cluster::cfs_update(); | |
63f6a08c | 395 | |
022e4e79 DM |
396 | my $nodename = $self->{nodename}; |
397 | ||
6ca2edcd | 398 | my (undef, $service_type, $service_name) = PVE::HA::Tools::parse_sid($sid); |
b47a7a1b | 399 | |
303a08aa | 400 | my $plugin = PVE::HA::Resources->lookup($service_type); |
85316fd0 TL |
401 | if (!$plugin) { |
402 | $self->log('err', "service type '$service_type' not implemented"); | |
403 | return EUNKNOWN_SERVICE_TYPE; | |
404 | } | |
303a08aa | 405 | |
85316fd0 TL |
406 | if ($service_config->{node} ne $nodename) { |
407 | $self->log('err', "service '$sid' not on this node"); | |
408 | return EWRONG_NODE; | |
409 | } | |
022e4e79 | 410 | |
b47a7a1b | 411 | my $vmid = $service_name; |
022e4e79 | 412 | |
303a08aa | 413 | my $running = $plugin->check_running($vmid); |
022e4e79 | 414 | |
303a08aa | 415 | if ($cmd eq 'started') { |
022e4e79 | 416 | |
a89ff919 | 417 | return SUCCESS if $running; |
022e4e79 DM |
418 | |
419 | $self->log("info", "starting service $sid"); | |
420 | ||
b865e4cb | 421 | $plugin->start($self, $vmid); |
022e4e79 | 422 | |
303a08aa | 423 | $running = $plugin->check_running($vmid); |
022e4e79 DM |
424 | |
425 | if ($running) { | |
426 | $self->log("info", "service status $sid started"); | |
a89ff919 | 427 | return SUCCESS; |
022e4e79 | 428 | } else { |
4877c031 | 429 | $self->log("warning", "unable to start service $sid"); |
a89ff919 | 430 | return ERROR; |
022e4e79 DM |
431 | } |
432 | ||
433 | } elsif ($cmd eq 'request_stop' || $cmd eq 'stopped') { | |
434 | ||
a89ff919 | 435 | return SUCCESS if !$running; |
022e4e79 DM |
436 | |
437 | $self->log("info", "stopping service $sid"); | |
438 | ||
b865e4cb | 439 | $plugin->shutdown($self, $vmid); |
022e4e79 | 440 | |
303a08aa | 441 | $running = $plugin->check_running($vmid); |
022e4e79 DM |
442 | |
443 | if (!$running) { | |
444 | $self->log("info", "service status $sid stopped"); | |
a89ff919 | 445 | return SUCCESS; |
022e4e79 | 446 | } else { |
33f01524 | 447 | $self->log("info", "unable to stop stop service $sid (still running)"); |
a89ff919 | 448 | return ERROR; |
022e4e79 DM |
449 | } |
450 | ||
451 | } elsif ($cmd eq 'migrate' || $cmd eq 'relocate') { | |
452 | ||
6da27e23 | 453 | my $target = $params[0]; |
85316fd0 TL |
454 | if (!defined($target)) { |
455 | die "$cmd '$sid' failed - missing target\n" if !defined($target); | |
456 | return EINVALID_PARAMETER; | |
457 | } | |
e319b50c | 458 | |
6da27e23 DM |
459 | if ($service_config->{node} eq $target) { |
460 | # already there | |
a89ff919 | 461 | return SUCCESS; |
303a08aa | 462 | } |
e319b50c | 463 | |
c0255b2c TL |
464 | my $online = ($cmd eq 'migrate') ? 1 : 0; |
465 | ||
303a08aa | 466 | my $oldconfig = $plugin->config_file($vmid, $nodename); |
e319b50c | 467 | |
c0255b2c | 468 | $plugin->migrate($self, $vmid, $target, $online); |
e319b50c | 469 | |
303a08aa TL |
470 | # something went wrong if old config file is still there |
471 | if (-f $oldconfig) { | |
472 | $self->log("err", "service $sid not moved (migration error)"); | |
a89ff919 | 473 | return ERROR; |
6da27e23 | 474 | } |
303a08aa | 475 | |
a89ff919 | 476 | return SUCCESS; |
303a08aa | 477 | |
a2881965 TL |
478 | } elsif ($cmd eq 'error') { |
479 | ||
63f6a08c | 480 | if ($running) { |
a2881965 TL |
481 | $self->log("err", "service $sid is in an error state while running"); |
482 | } else { | |
4877c031 | 483 | $self->log("warning", "service $sid is not running and in an error state"); |
a2881965 | 484 | } |
a89ff919 | 485 | return SUCCESS; # error always succeeds |
a2881965 | 486 | |
022e4e79 DM |
487 | } |
488 | ||
85316fd0 TL |
489 | $self->log("err", "implement me (cmd '$cmd')"); |
490 | return EUNKNOWN_COMMAND; | |
c4a221bc DM |
491 | } |
492 | ||
714a4016 | 493 | 1; |