]>
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; | |
abc920b4 | 11 | use PVE::Cluster qw(cfs_register_file cfs_read_file cfs_lock_file); |
022e4e79 DM |
12 | use PVE::INotify; |
13 | use PVE::RPCEnvironment; | |
714a4016 DM |
14 | |
15 | use PVE::HA::Tools; | |
16 | use PVE::HA::Env; | |
ce216792 | 17 | use PVE::HA::Config; |
714a4016 | 18 | |
022e4e79 DM |
19 | use PVE::QemuServer; |
20 | use PVE::API2::Qemu; | |
21 | ||
007fcc8b DM |
22 | my $lockdir = "/etc/pve/priv/lock"; |
23 | ||
ce216792 | 24 | my $manager_status_filename = "/etc/pve/ha/manager_status"; |
7a19642e DM |
25 | my $ha_groups_config = "/etc/pve/ha/groups.cfg"; |
26 | my $ha_resources_config = "/etc/pve/ha/resources.cfg"; | |
abc920b4 | 27 | |
6cbcb5f7 DM |
28 | #cfs_register_file($ha_groups_config, |
29 | # sub { PVE::HA::Groups->parse_config(@_); }, | |
30 | # sub { PVE::HA::Groups->write_config(@_); }); | |
714a4016 DM |
31 | |
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 $self->{nodename}; | |
50 | } | |
51 | ||
52 | sub read_manager_status { | |
53 | my ($self) = @_; | |
54 | ||
55 | my $filename = $manager_status_filename; | |
56 | ||
57 | return PVE::HA::Tools::read_json_from_file($filename, {}); | |
58 | } | |
59 | ||
60 | sub write_manager_status { | |
61 | my ($self, $status_obj) = @_; | |
62 | ||
63 | my $filename = $manager_status_filename; | |
64 | ||
65 | PVE::HA::Tools::write_json_to_file($filename, $status_obj); | |
66 | } | |
67 | ||
c4a221bc DM |
68 | sub read_lrm_status { |
69 | my ($self, $node) = @_; | |
70 | ||
71 | $node = $self->{nodename} if !defined($node); | |
72 | ||
73 | my $filename = "/etc/pve/nodes/$node/lrm_status"; | |
74 | ||
75 | return PVE::HA::Tools::read_json_from_file($filename, {}); | |
76 | } | |
77 | ||
78 | sub write_lrm_status { | |
79 | my ($self, $status_obj) = @_; | |
80 | ||
6cbcb5f7 | 81 | my $node = $self->{nodename}; |
c4a221bc DM |
82 | |
83 | my $filename = "/etc/pve/nodes/$node/lrm_status"; | |
84 | ||
85 | PVE::HA::Tools::write_json_to_file($filename, $status_obj); | |
86 | } | |
87 | ||
714a4016 DM |
88 | sub manager_status_exists { |
89 | my ($self) = @_; | |
90 | ||
91 | return -f $manager_status_filename ? 1 : 0; | |
92 | } | |
93 | ||
94 | sub read_service_config { | |
95 | my ($self) = @_; | |
96 | ||
ce216792 DM |
97 | # fixme: use cfs_read_file |
98 | ||
99 | my $raw = ''; | |
100 | ||
101 | $raw = PVE::Tools::file_get_contents($ha_resources_config) | |
102 | if -f $ha_resources_config; | |
103 | ||
104 | my $res = PVE::HA::Config::parse_resources_config($ha_resources_config, $raw); | |
105 | ||
106 | my $vmlist = PVE::Cluster::get_vmlist(); | |
107 | my $conf = {}; | |
108 | ||
109 | foreach my $sid (keys %{$res->{ids}}) { | |
110 | my $d = $res->{ids}->{$sid}; | |
7a19642e | 111 | $d->{state} = 'enabled' if !defined($d->{state}); |
ce216792 DM |
112 | if ($d->{type} eq 'pvevm') { |
113 | if (my $vmd = $vmlist->{ids}->{$d->{name}}) { | |
114 | if (!$vmd) { | |
115 | warn "no such VM '$d->{name}'\n"; | |
116 | } else { | |
117 | $d->{node} = $vmd->{node}; | |
118 | $conf->{$sid} = $d; | |
119 | } | |
120 | } else { | |
121 | if (defined($d->{node})) { | |
122 | $conf->{$sid} = $d; | |
123 | } else { | |
124 | warn "service '$sid' without node\n"; | |
125 | } | |
126 | } | |
127 | } | |
128 | } | |
129 | ||
130 | return $conf; | |
714a4016 DM |
131 | } |
132 | ||
8456bde2 DM |
133 | sub change_service_location { |
134 | my ($self, $sid, $node) = @_; | |
135 | ||
136 | die "implement me"; | |
137 | } | |
138 | ||
abc920b4 DM |
139 | sub read_group_config { |
140 | my ($self) = @_; | |
141 | ||
ce216792 DM |
142 | # fixme: use cfs_read_file |
143 | ||
144 | my $raw = ''; | |
145 | ||
146 | $raw = PVE::Tools::file_get_contents($ha_groups_config) | |
147 | if -f $ha_groups_config; | |
148 | ||
149 | return PVE::HA::Config::parse_groups_config($ha_groups_config, $raw); | |
abc920b4 DM |
150 | } |
151 | ||
3b996922 DM |
152 | sub queue_crm_commands { |
153 | my ($self, $cmd) = @_; | |
154 | ||
ce216792 DM |
155 | chomp $cmd; |
156 | ||
157 | my $code = sub { | |
158 | my $data = ''; | |
159 | my $filename = "/etc/pve/ha/crm_commands"; | |
160 | if (-f $filename) { | |
161 | $data = PVE::Tools::file_get_contents($filename); | |
162 | } | |
163 | $data .= "$cmd\n"; | |
164 | PVE::Tools::file_set_contents($filename, $data); | |
165 | }; | |
166 | ||
167 | # fixme: do not use cfs_lock_storage (replace with cfs_lock_ha) | |
168 | my $res = PVE::Cluster::cfs_lock_storage("_ha_crm_commands", undef, $code); | |
169 | die $@ if $@; | |
170 | return $res; | |
3b996922 DM |
171 | } |
172 | ||
173 | sub read_crm_commands { | |
174 | my ($self) = @_; | |
175 | ||
ce216792 DM |
176 | my $code = sub { |
177 | my $data = ''; | |
178 | ||
179 | my $filename = "/etc/pve/ha/crm_commands"; | |
180 | if (-f $filename) { | |
181 | $data = PVE::Tools::file_get_contents($filename); | |
182 | PVE::Tools::file_set_contents($filename, ''); | |
183 | } | |
184 | ||
185 | return $data; | |
186 | }; | |
187 | ||
188 | # fixme: do not use cfs_lock_storage (replace with cfs_lock_ha) | |
189 | my $res = PVE::Cluster::cfs_lock_storage("_ha_crm_commands", undef, $code); | |
190 | die $@ if $@; | |
191 | return $res; | |
3b996922 DM |
192 | } |
193 | ||
714a4016 DM |
194 | # this should return a hash containing info |
195 | # what nodes are members and online. | |
196 | sub get_node_info { | |
197 | my ($self) = @_; | |
198 | ||
d706ef8b DM |
199 | my ($node_info, $quorate) = ({}, 0); |
200 | ||
201 | my $nodename = $self->{nodename}; | |
202 | ||
203 | $quorate = PVE::Cluster::check_cfs_quorum(1) || 0; | |
204 | ||
205 | my $members = PVE::Cluster::get_members(); | |
206 | ||
207 | foreach my $node (keys %$members) { | |
208 | my $d = $members->{$node}; | |
209 | $node_info->{$node}->{online} = $d->{online}; | |
210 | } | |
211 | ||
212 | $node_info->{$nodename}->{online} = 1; # local node is always up | |
213 | ||
214 | return ($node_info, $quorate); | |
714a4016 DM |
215 | } |
216 | ||
217 | sub log { | |
218 | my ($self, $level, $msg) = @_; | |
219 | ||
220 | chomp $msg; | |
221 | ||
222 | syslog($level, $msg); | |
223 | } | |
224 | ||
007fcc8b DM |
225 | my $last_lock_status = {}; |
226 | ||
227 | sub get_pve_lock { | |
228 | my ($self, $lockid) = @_; | |
714a4016 | 229 | |
007fcc8b | 230 | my $got_lock = 0; |
4d24e7db | 231 | |
4d24e7db DM |
232 | my $filename = "$lockdir/$lockid"; |
233 | ||
007fcc8b DM |
234 | my $last = $last_lock_status->{$lockid} || 0; |
235 | ||
236 | my $ctime = time(); | |
4d24e7db DM |
237 | |
238 | eval { | |
239 | ||
240 | mkdir $lockdir; | |
241 | ||
007fcc8b DM |
242 | # pve cluster filesystem not online |
243 | die "can't create '$lockdir' (pmxcfs not mounted?)\n" if ! -d $lockdir; | |
244 | ||
245 | if ($last && (($ctime - $last) < 100)) { # fixme: what timeout | |
246 | utime(0, $ctime, $filename) || # cfs lock update request | |
247 | die "cfs lock update failed - $!\n"; | |
248 | } else { | |
249 | ||
250 | # fixme: wait some time? | |
251 | if (!(mkdir $filename)) { | |
252 | utime 0, 0, $filename; # cfs unlock request | |
253 | die "can't get cfs lock\n"; | |
254 | } | |
255 | } | |
4d24e7db | 256 | |
007fcc8b | 257 | $got_lock = 1; |
4d24e7db DM |
258 | }; |
259 | ||
007fcc8b DM |
260 | my $err = $@; |
261 | ||
262 | $last_lock_status->{$lockid} = $got_lock ? $ctime : 0; | |
263 | ||
17e90af6 | 264 | if (!!$got_lock != !!$last) { |
007fcc8b DM |
265 | if ($got_lock) { |
266 | $self->log('info', "successfully aquired lock '$lockid'"); | |
267 | } else { | |
268 | my $msg = "lost lock '$lockid"; | |
269 | $msg .= " - $err" if $err; | |
270 | $self->log('err', $msg); | |
271 | } | |
272 | } | |
273 | ||
274 | return $got_lock; | |
275 | } | |
276 | ||
277 | sub get_ha_manager_lock { | |
278 | my ($self) = @_; | |
279 | ||
007fcc8b | 280 | return $self->get_pve_lock("ha_manager_lock"); |
714a4016 DM |
281 | } |
282 | ||
283 | sub get_ha_agent_lock { | |
284 | my ($self) = @_; | |
007fcc8b DM |
285 | |
286 | my $node = $self->nodename(); | |
714a4016 | 287 | |
007fcc8b | 288 | return $self->get_pve_lock("ha_agent_${node}_lock"); |
714a4016 DM |
289 | } |
290 | ||
291 | sub test_ha_agent_lock { | |
292 | my ($self, $node) = @_; | |
007fcc8b DM |
293 | |
294 | my $lockid = "ha_agent_${node}_lock"; | |
295 | my $filename = "$lockdir/$lockid"; | |
296 | my $res = $self->get_pve_lock($lockid); | |
297 | rmdir $filename if $res; # cfs unlock | |
714a4016 | 298 | |
007fcc8b | 299 | return $res; |
714a4016 DM |
300 | } |
301 | ||
302 | sub quorate { | |
303 | my ($self) = @_; | |
304 | ||
4d24e7db DM |
305 | my $quorate = 0; |
306 | eval { | |
307 | $quorate = PVE::Cluster::check_cfs_quorum(); | |
308 | }; | |
309 | ||
310 | return $quorate; | |
714a4016 DM |
311 | } |
312 | ||
313 | sub get_time { | |
314 | my ($self) = @_; | |
315 | ||
316 | return time(); | |
317 | } | |
318 | ||
319 | sub sleep { | |
320 | my ($self, $delay) = @_; | |
321 | ||
322 | CORE::sleep($delay); | |
323 | } | |
324 | ||
325 | sub sleep_until { | |
326 | my ($self, $end_time) = @_; | |
327 | ||
328 | for (;;) { | |
329 | my $cur_time = time(); | |
330 | ||
331 | last if $cur_time >= $end_time; | |
332 | ||
333 | $self->sleep(1); | |
334 | } | |
335 | } | |
336 | ||
337 | sub loop_start_hook { | |
338 | my ($self) = @_; | |
339 | ||
4d24e7db DM |
340 | PVE::Cluster::cfs_update(); |
341 | ||
714a4016 DM |
342 | $self->{loop_start} = $self->get_time(); |
343 | } | |
344 | ||
345 | sub loop_end_hook { | |
346 | my ($self) = @_; | |
347 | ||
348 | my $delay = $self->get_time() - $self->{loop_start}; | |
349 | ||
350 | warn "loop take too long ($delay seconds)\n" if $delay > 30; | |
351 | } | |
352 | ||
76737af5 DM |
353 | my $watchdog_fh; |
354 | ||
714a4016 DM |
355 | sub watchdog_open { |
356 | my ($self) = @_; | |
357 | ||
76737af5 DM |
358 | die "watchdog already open\n" if defined($watchdog_fh); |
359 | ||
115805fd DM |
360 | $watchdog_fh = IO::Socket::UNIX->new( |
361 | Type => SOCK_STREAM(), | |
362 | Peer => "/run/watchdog-mux.sock") || | |
363 | die "unable to open watchdog socket - $!\n"; | |
364 | ||
76737af5 | 365 | $self->log('info', "watchdog active"); |
714a4016 DM |
366 | } |
367 | ||
368 | sub watchdog_update { | |
369 | my ($self, $wfh) = @_; | |
370 | ||
76737af5 DM |
371 | my $res = $watchdog_fh->syswrite("\0", 1); |
372 | if (!defined($res)) { | |
373 | $self->log('err', "watchdog update failed - $!\n"); | |
374 | return 0; | |
375 | } | |
376 | if ($res != 1) { | |
377 | $self->log('err', "watchdog update failed - write $res bytes\n"); | |
378 | return 0; | |
379 | } | |
380 | ||
381 | return 1; | |
714a4016 DM |
382 | } |
383 | ||
384 | sub watchdog_close { | |
385 | my ($self, $wfh) = @_; | |
386 | ||
76737af5 DM |
387 | $watchdog_fh->syswrite("V", 1); # magic watchdog close |
388 | if (!$watchdog_fh->close()) { | |
389 | $self->log('err', "watchdog close failed - $!"); | |
390 | } else { | |
391 | $watchdog_fh = undef; | |
392 | $self->log('info', "watchdog closed (disabled)"); | |
393 | } | |
714a4016 DM |
394 | } |
395 | ||
022e4e79 DM |
396 | sub upid_wait { |
397 | my ($self, $upid) = @_; | |
398 | ||
399 | my $task = PVE::Tools::upid_decode($upid); | |
400 | ||
401 | CORE::sleep(1); | |
402 | while (PVE::ProcFSTools::check_process_running($task->{pid}, $task->{pstart})) { | |
403 | $self->log('debug', "Task still active, waiting"); | |
404 | CORE::sleep(1); | |
405 | } | |
406 | } | |
407 | ||
c4a221bc | 408 | sub exec_resource_agent { |
6dbf93a0 | 409 | my ($self, $sid, $service_config, $cmd, @params) = @_; |
c4a221bc | 410 | |
022e4e79 DM |
411 | # setup execution environment |
412 | ||
413 | $ENV{'PATH'} = '/sbin:/bin:/usr/sbin:/usr/bin'; | |
414 | ||
415 | PVE::INotify::inotify_close(); | |
416 | ||
417 | PVE::INotify::inotify_init(); | |
418 | ||
419 | PVE::Cluster::cfs_update(); | |
420 | ||
421 | my $nodename = $self->{nodename}; | |
422 | ||
423 | # fixme: return valid_exit code (instead of using die) ? | |
424 | ||
425 | my $service_type = $service_config->{type}; | |
426 | ||
427 | die "service type '$service_type'not implemented" if $service_type ne 'pvevm'; | |
428 | ||
429 | my $vmid = $service_config->{name}; | |
430 | ||
431 | my $running = PVE::QemuServer::check_running($vmid, 1); | |
432 | ||
433 | if ($cmd eq 'started') { | |
434 | ||
435 | # fixme: return valid_exit code | |
436 | die "service '$sid' not on this node" if $service_config->{node} ne $nodename; | |
437 | ||
438 | # fixme: count failures | |
439 | ||
440 | return 0 if $running; | |
441 | ||
442 | $self->log("info", "starting service $sid"); | |
443 | ||
444 | my $upid = PVE::API2::Qemu->vm_start({node => $nodename, vmid => $vmid}); | |
445 | $self->upid_wait($upid); | |
446 | ||
447 | $running = PVE::QemuServer::check_running($vmid, 1); | |
448 | ||
449 | if ($running) { | |
450 | $self->log("info", "service status $sid started"); | |
451 | return 0; | |
452 | } else { | |
453 | $self->log("info", "unable to start service $sid"); | |
454 | return 1; | |
455 | } | |
456 | ||
457 | } elsif ($cmd eq 'request_stop' || $cmd eq 'stopped') { | |
458 | ||
459 | # fixme: return valid_exit code | |
460 | die "service '$sid' not on this node" if $service_config->{node} ne $nodename; | |
461 | ||
462 | return 0 if !$running; | |
463 | ||
464 | $self->log("info", "stopping service $sid"); | |
465 | ||
466 | my $timeout = 60; # fixme: make this configurable | |
467 | ||
468 | my $param = { | |
469 | node => $nodename, | |
470 | vmid => $vmid, | |
471 | timeout => $timeout, | |
472 | forceStop => 1, | |
473 | }; | |
474 | ||
475 | my $upid = PVE::API2::Qemu->vm_shutdown($param); | |
476 | $self->upid_wait($upid); | |
477 | ||
478 | $running = PVE::QemuServer::check_running($vmid, 1); | |
479 | ||
480 | if (!$running) { | |
481 | $self->log("info", "service status $sid stopped"); | |
482 | return 0; | |
483 | } else { | |
484 | return 1; | |
485 | } | |
486 | ||
487 | } elsif ($cmd eq 'migrate' || $cmd eq 'relocate') { | |
488 | ||
489 | # implement me | |
490 | ||
491 | } | |
492 | ||
493 | die "implement me (cmd '$cmd')"; | |
c4a221bc DM |
494 | } |
495 | ||
714a4016 | 496 | 1; |