]>
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; |
9e5ea8f7 DM |
18 | use PVE::HA::Resources; |
19 | use PVE::HA::Resources::PVEVM; | |
20 | use PVE::HA::Resources::PVECT; | |
714a4016 | 21 | |
9e5ea8f7 DM |
22 | PVE::HA::Resources::PVEVM->register(); |
23 | PVE::HA::Resources::PVECT->register(); | |
24 | ||
25 | PVE::HA::Resources->init(); | |
022e4e79 | 26 | |
007fcc8b DM |
27 | my $lockdir = "/etc/pve/priv/lock"; |
28 | ||
714a4016 DM |
29 | sub new { |
30 | my ($this, $nodename) = @_; | |
31 | ||
32 | die "missing nodename" if !$nodename; | |
33 | ||
34 | my $class = ref($this) || $this; | |
35 | ||
36 | my $self = bless {}, $class; | |
37 | ||
38 | $self->{nodename} = $nodename; | |
39 | ||
40 | return $self; | |
41 | } | |
42 | ||
43 | sub nodename { | |
44 | my ($self) = @_; | |
45 | ||
46 | return $self->{nodename}; | |
47 | } | |
48 | ||
dd9c0c9d TL |
49 | sub hardware { |
50 | my ($self) = @_; | |
51 | ||
52 | die "hardware is for testing and simulation only"; | |
53 | } | |
54 | ||
714a4016 DM |
55 | sub read_manager_status { |
56 | my ($self) = @_; | |
714a4016 | 57 | |
139a9b90 | 58 | return PVE::HA::Config::read_manager_status(); |
714a4016 DM |
59 | } |
60 | ||
61 | sub write_manager_status { | |
62 | my ($self, $status_obj) = @_; | |
63f6a08c | 63 | |
139a9b90 | 64 | PVE::HA::Config::write_manager_status($status_obj); |
714a4016 DM |
65 | } |
66 | ||
c4a221bc DM |
67 | sub read_lrm_status { |
68 | my ($self, $node) = @_; | |
69 | ||
70 | $node = $self->{nodename} if !defined($node); | |
71 | ||
139a9b90 | 72 | return PVE::HA::Config::read_lrm_status($node); |
c4a221bc DM |
73 | } |
74 | ||
75 | sub write_lrm_status { | |
76 | my ($self, $status_obj) = @_; | |
77 | ||
6cbcb5f7 | 78 | my $node = $self->{nodename}; |
63f6a08c | 79 | |
139a9b90 DM |
80 | PVE::HA::Config::write_lrm_status($node, $status_obj); |
81 | } | |
c4a221bc | 82 | |
cde77779 | 83 | sub is_node_shutdown { |
d42219a3 TL |
84 | my ($self) = @_; |
85 | ||
cde77779 | 86 | my $shutdown = 0; |
d42219a3 TL |
87 | |
88 | my $code = sub { | |
89 | my $line = shift; | |
90 | ||
cde77779 | 91 | $shutdown = 1 if ($line =~ m/shutdown\.target/); |
d42219a3 TL |
92 | }; |
93 | ||
94 | my $cmd = ['/bin/systemctl', 'list-jobs']; | |
95 | eval { PVE::Tools::run_command($cmd, outfunc => $code, noerr => 1); }; | |
96 | ||
cde77779 | 97 | return $shutdown; |
d42219a3 TL |
98 | } |
99 | ||
139a9b90 DM |
100 | sub queue_crm_commands { |
101 | my ($self, $cmd) = @_; | |
c4a221bc | 102 | |
139a9b90 DM |
103 | return PVE::HA::Config::queue_crm_commands($cmd); |
104 | } | |
105 | ||
106 | sub read_crm_commands { | |
107 | my ($self) = @_; | |
108 | ||
109 | return PVE::HA::Config::read_crm_commands(); | |
c4a221bc DM |
110 | } |
111 | ||
b83b4ae8 | 112 | sub service_config_exists { |
714a4016 | 113 | my ($self) = @_; |
63f6a08c | 114 | |
b83b4ae8 DM |
115 | return PVE::HA::Config::resources_config_exists(); |
116 | } | |
714a4016 | 117 | |
b83b4ae8 DM |
118 | sub read_service_config { |
119 | my ($self) = @_; | |
ce216792 | 120 | |
b83b4ae8 | 121 | my $res = PVE::HA::Config::read_resources_config(); |
63f6a08c | 122 | |
ce216792 DM |
123 | my $vmlist = PVE::Cluster::get_vmlist(); |
124 | my $conf = {}; | |
125 | ||
126 | foreach my $sid (keys %{$res->{ids}}) { | |
127 | my $d = $res->{ids}->{$sid}; | |
6ca2edcd | 128 | my (undef, undef, $name) = PVE::HA::Tools::parse_sid($sid); |
7a19642e | 129 | $d->{state} = 'enabled' if !defined($d->{state}); |
ea4443cc TL |
130 | $d->{max_restart} = 1 if !defined($d->{max_restart}); |
131 | $d->{max_relocate} = 1 if !defined($d->{max_relocate}); | |
303a08aa | 132 | if (PVE::HA::Resources->lookup($d->{type})) { |
b47a7a1b | 133 | if (my $vmd = $vmlist->{ids}->{$name}) { |
ce216792 | 134 | if (!$vmd) { |
b47a7a1b | 135 | warn "no such VM '$name'\n"; |
ce216792 DM |
136 | } else { |
137 | $d->{node} = $vmd->{node}; | |
138 | $conf->{$sid} = $d; | |
139 | } | |
140 | } else { | |
141 | if (defined($d->{node})) { | |
142 | $conf->{$sid} = $d; | |
143 | } else { | |
144 | warn "service '$sid' without node\n"; | |
145 | } | |
146 | } | |
147 | } | |
148 | } | |
63f6a08c | 149 | |
ce216792 | 150 | return $conf; |
714a4016 DM |
151 | } |
152 | ||
8456bde2 | 153 | sub change_service_location { |
6da27e23 | 154 | my ($self, $sid, $current_node, $new_node) = @_; |
8456bde2 | 155 | |
6ca2edcd | 156 | my (undef, $type, $name) = PVE::HA::Tools::parse_sid($sid); |
6da27e23 | 157 | |
303a08aa TL |
158 | if(my $plugin = PVE::HA::Resources->lookup($type)) { |
159 | my $old = $plugin->config_file($name, $current_node); | |
160 | my $new = $plugin->config_file($name, $new_node); | |
6da27e23 DM |
161 | rename($old, $new) || |
162 | die "rename '$old' to '$new' failed - $!\n"; | |
163 | } else { | |
164 | die "implement me"; | |
165 | } | |
8456bde2 DM |
166 | } |
167 | ||
abc920b4 DM |
168 | sub read_group_config { |
169 | my ($self) = @_; | |
170 | ||
139a9b90 | 171 | return PVE::HA::Config::read_group_config(); |
3b996922 DM |
172 | } |
173 | ||
714a4016 DM |
174 | # this should return a hash containing info |
175 | # what nodes are members and online. | |
176 | sub get_node_info { | |
177 | my ($self) = @_; | |
178 | ||
d706ef8b | 179 | my ($node_info, $quorate) = ({}, 0); |
63f6a08c | 180 | |
d706ef8b DM |
181 | my $nodename = $self->{nodename}; |
182 | ||
183 | $quorate = PVE::Cluster::check_cfs_quorum(1) || 0; | |
184 | ||
185 | my $members = PVE::Cluster::get_members(); | |
186 | ||
187 | foreach my $node (keys %$members) { | |
188 | my $d = $members->{$node}; | |
63f6a08c | 189 | $node_info->{$node}->{online} = $d->{online}; |
d706ef8b | 190 | } |
63f6a08c | 191 | |
d706ef8b | 192 | $node_info->{$nodename}->{online} = 1; # local node is always up |
63f6a08c | 193 | |
d706ef8b | 194 | return ($node_info, $quorate); |
714a4016 DM |
195 | } |
196 | ||
197 | sub log { | |
198 | my ($self, $level, $msg) = @_; | |
199 | ||
200 | chomp $msg; | |
201 | ||
202 | syslog($level, $msg); | |
203 | } | |
204 | ||
007fcc8b DM |
205 | my $last_lock_status = {}; |
206 | ||
207 | sub get_pve_lock { | |
208 | my ($self, $lockid) = @_; | |
714a4016 | 209 | |
007fcc8b | 210 | my $got_lock = 0; |
4d24e7db | 211 | |
4d24e7db DM |
212 | my $filename = "$lockdir/$lockid"; |
213 | ||
007fcc8b DM |
214 | my $last = $last_lock_status->{$lockid} || 0; |
215 | ||
216 | my $ctime = time(); | |
4d24e7db | 217 | |
75aca181 DM |
218 | my $retry = 0; |
219 | my $retry_timeout = 100; # fixme: what timeout | |
63f6a08c | 220 | |
4d24e7db DM |
221 | eval { |
222 | ||
223 | mkdir $lockdir; | |
224 | ||
007fcc8b DM |
225 | # pve cluster filesystem not online |
226 | die "can't create '$lockdir' (pmxcfs not mounted?)\n" if ! -d $lockdir; | |
227 | ||
75aca181 DM |
228 | if ($last && (($ctime - $last) < $retry_timeout)) { |
229 | # send cfs lock update request (utime) | |
230 | if (!utime(0, $ctime, $filename)) { | |
231 | $retry = 1; | |
007fcc8b | 232 | die "cfs lock update failed - $!\n"; |
75aca181 | 233 | } |
007fcc8b DM |
234 | } else { |
235 | ||
236 | # fixme: wait some time? | |
237 | if (!(mkdir $filename)) { | |
238 | utime 0, 0, $filename; # cfs unlock request | |
239 | die "can't get cfs lock\n"; | |
240 | } | |
241 | } | |
4d24e7db | 242 | |
007fcc8b | 243 | $got_lock = 1; |
4d24e7db DM |
244 | }; |
245 | ||
007fcc8b DM |
246 | my $err = $@; |
247 | ||
75aca181 DM |
248 | if ($retry) { |
249 | # $self->log('err', $err) if $err; # for debugging | |
250 | return 0; | |
251 | } | |
63f6a08c | 252 | |
007fcc8b DM |
253 | $last_lock_status->{$lockid} = $got_lock ? $ctime : 0; |
254 | ||
17e90af6 | 255 | if (!!$got_lock != !!$last) { |
007fcc8b | 256 | if ($got_lock) { |
63f6a08c | 257 | $self->log('info', "successfully acquired lock '$lockid'"); |
007fcc8b DM |
258 | } else { |
259 | my $msg = "lost lock '$lockid"; | |
63f6a08c | 260 | $msg .= " - $err" if $err; |
007fcc8b DM |
261 | $self->log('err', $msg); |
262 | } | |
75aca181 DM |
263 | } else { |
264 | # $self->log('err', $err) if $err; # for debugging | |
007fcc8b DM |
265 | } |
266 | ||
267 | return $got_lock; | |
268 | } | |
269 | ||
270 | sub get_ha_manager_lock { | |
271 | my ($self) = @_; | |
272 | ||
007fcc8b | 273 | return $self->get_pve_lock("ha_manager_lock"); |
714a4016 DM |
274 | } |
275 | ||
de002253 TL |
276 | # release the cluster wide manager lock. |
277 | # when released another CRM may step up and get the lock, thus this should only | |
278 | # get called when shutting down/deactivating the current master | |
279 | sub release_ha_manager_lock { | |
280 | my ($self) = @_; | |
281 | ||
282 | return rmdir("$lockdir/ha_manager_lock"); | |
283 | } | |
284 | ||
714a4016 | 285 | sub get_ha_agent_lock { |
714a4016 | 286 | my ($self, $node) = @_; |
63f6a08c | 287 | |
f5c29173 | 288 | $node = $self->nodename() if !defined($node); |
714a4016 | 289 | |
f5c29173 | 290 | return $self->get_pve_lock("ha_agent_${node}_lock"); |
714a4016 DM |
291 | } |
292 | ||
ff165cd8 TL |
293 | # release the respective node agent lock. |
294 | # this should only get called if the nodes LRM gracefully shuts down with | |
295 | # all services already cleanly stopped! | |
296 | sub release_ha_agent_lock { | |
297 | my ($self) = @_; | |
298 | ||
299 | my $node = $self->nodename(); | |
300 | ||
301 | return rmdir("$lockdir/ha_agent_${node}_lock"); | |
302 | } | |
303 | ||
714a4016 DM |
304 | sub quorate { |
305 | my ($self) = @_; | |
306 | ||
4d24e7db | 307 | my $quorate = 0; |
63f6a08c TL |
308 | eval { |
309 | $quorate = PVE::Cluster::check_cfs_quorum(); | |
4d24e7db | 310 | }; |
63f6a08c | 311 | |
4d24e7db | 312 | return $quorate; |
714a4016 DM |
313 | } |
314 | ||
315 | sub get_time { | |
316 | my ($self) = @_; | |
317 | ||
318 | return time(); | |
319 | } | |
320 | ||
321 | sub sleep { | |
322 | my ($self, $delay) = @_; | |
323 | ||
324 | CORE::sleep($delay); | |
325 | } | |
326 | ||
327 | sub sleep_until { | |
328 | my ($self, $end_time) = @_; | |
329 | ||
330 | for (;;) { | |
331 | my $cur_time = time(); | |
332 | ||
333 | last if $cur_time >= $end_time; | |
334 | ||
335 | $self->sleep(1); | |
336 | } | |
337 | } | |
338 | ||
339 | sub loop_start_hook { | |
340 | my ($self) = @_; | |
341 | ||
4d24e7db | 342 | PVE::Cluster::cfs_update(); |
63f6a08c | 343 | |
714a4016 DM |
344 | $self->{loop_start} = $self->get_time(); |
345 | } | |
346 | ||
347 | sub loop_end_hook { | |
348 | my ($self) = @_; | |
349 | ||
350 | my $delay = $self->get_time() - $self->{loop_start}; | |
63f6a08c | 351 | |
714a4016 DM |
352 | warn "loop take too long ($delay seconds)\n" if $delay > 30; |
353 | } | |
354 | ||
76737af5 DM |
355 | my $watchdog_fh; |
356 | ||
714a4016 DM |
357 | sub watchdog_open { |
358 | my ($self) = @_; | |
359 | ||
76737af5 DM |
360 | die "watchdog already open\n" if defined($watchdog_fh); |
361 | ||
115805fd DM |
362 | $watchdog_fh = IO::Socket::UNIX->new( |
363 | Type => SOCK_STREAM(), | |
364 | Peer => "/run/watchdog-mux.sock") || | |
365 | die "unable to open watchdog socket - $!\n"; | |
63f6a08c | 366 | |
76737af5 | 367 | $self->log('info', "watchdog active"); |
714a4016 DM |
368 | } |
369 | ||
370 | sub watchdog_update { | |
371 | my ($self, $wfh) = @_; | |
372 | ||
76737af5 DM |
373 | my $res = $watchdog_fh->syswrite("\0", 1); |
374 | if (!defined($res)) { | |
375 | $self->log('err', "watchdog update failed - $!\n"); | |
376 | return 0; | |
377 | } | |
378 | if ($res != 1) { | |
379 | $self->log('err', "watchdog update failed - write $res bytes\n"); | |
380 | return 0; | |
381 | } | |
382 | ||
383 | return 1; | |
714a4016 DM |
384 | } |
385 | ||
386 | sub watchdog_close { | |
387 | my ($self, $wfh) = @_; | |
388 | ||
76737af5 DM |
389 | $watchdog_fh->syswrite("V", 1); # magic watchdog close |
390 | if (!$watchdog_fh->close()) { | |
391 | $self->log('err', "watchdog close failed - $!"); | |
392 | } else { | |
393 | $watchdog_fh = undef; | |
394 | $self->log('info', "watchdog closed (disabled)"); | |
395 | } | |
714a4016 DM |
396 | } |
397 | ||
022e4e79 DM |
398 | sub upid_wait { |
399 | my ($self, $upid) = @_; | |
400 | ||
401 | my $task = PVE::Tools::upid_decode($upid); | |
402 | ||
403 | CORE::sleep(1); | |
404 | while (PVE::ProcFSTools::check_process_running($task->{pid}, $task->{pstart})) { | |
405 | $self->log('debug', "Task still active, waiting"); | |
406 | CORE::sleep(1); | |
407 | } | |
408 | } | |
409 | ||
0d1d32fb DM |
410 | sub can_fork { |
411 | my ($self) = @_; | |
412 | ||
413 | return 1; | |
414 | } | |
415 | ||
a2aae08a TL |
416 | sub after_fork { |
417 | my ($self) = @_; | |
418 | ||
419 | # close inherited inotify FD from parent and reopen our own | |
420 | PVE::INotify::inotify_close(); | |
421 | PVE::INotify::inotify_init(); | |
422 | ||
423 | PVE::Cluster::cfs_update(); | |
424 | } | |
425 | ||
714a4016 | 426 | 1; |