src/PVE/HA/Env/PVE2.pm

   1 package PVE::HA::Env::PVE2;
   2
   3 use strict;
   4 use warnings;
   5 use POSIX qw(:errno_h :fcntl_h);
   6 use IO::File;
   7 use IO::Socket::UNIX;
   8
   9 use PVE::SafeSyslog;
  10 use PVE::Tools;
  11 use PVE::Cluster qw(cfs_register_file cfs_read_file cfs_write_file cfs_lock_file);
  12 use PVE::INotify;
  13 use PVE::RPCEnvironment;
  14
  15 use PVE::HA::Tools ':exit_codes';
  16 use PVE::HA::Env;
  17 use PVE::HA::Config;
  18 use PVE::HA::Resources;
  19 use PVE::HA::Resources::PVEVM;
  20 use PVE::HA::Resources::PVECT;
  21
  22 PVE::HA::Resources::PVEVM->register();
  23 PVE::HA::Resources::PVECT->register();
  24
  25 PVE::HA::Resources->init();
  26
  27 my $lockdir = "/etc/pve/priv/lock";
  28
  29 sub new {
  30     my ($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 $self->{nodename};
  47 }
  48
  49 sub hardware {
  50     my ($self) = @_;
  51
  52     die "hardware is for testing and simulation only";
  53 }
  54
  55 sub read_manager_status {
  56     my ($self) = @_;
  57
  58     return PVE::HA::Config::read_manager_status();
  59 }
  60
  61 sub write_manager_status {
  62     my ($self, $status_obj) = @_;
  63
  64     PVE::HA::Config::write_manager_status($status_obj);
  65 }
  66
  67 sub read_lrm_status {
  68     my ($self, $node) = @_;
  69
  70     $node = $self->{nodename} if !defined($node);
  71
  72     return PVE::HA::Config::read_lrm_status($node);
  73 }
  74
  75 sub write_lrm_status {
  76     my ($self, $status_obj) = @_;
  77
  78     my $node = $self->{nodename};
  79
  80     PVE::HA::Config::write_lrm_status($node, $status_obj);
  81 }
  82
  83 sub is_node_shutdown {
  84     my ($self) = @_;
  85
  86     my $shutdown = 0;
  87
  88     my $code = sub {
  89         my $line = shift;
  90
  91         $shutdown = 1 if ($line =~ m/shutdown\.target/);
  92     };
  93
  94     my $cmd = ['/bin/systemctl', 'list-jobs'];
  95     eval { PVE::Tools::run_command($cmd, outfunc => $code, noerr => 1); };
  96
  97     return $shutdown;
  98 }
  99
 100 sub queue_crm_commands {
 101     my ($self, $cmd) = @_;
 102
 103     return PVE::HA::Config::queue_crm_commands($cmd);
 104 }
 105
 106 sub read_crm_commands {
 107     my ($self) = @_;
 108
 109     return PVE::HA::Config::read_crm_commands();
 110 }
 111
 112 sub read_service_config {
 113     my ($self) = @_;
 114
 115     my $res = PVE::HA::Config::read_resources_config();
 116
 117     my $vmlist = PVE::Cluster::get_vmlist();
 118     my $conf = {};
 119
 120     foreach my $sid (keys %{$res->{ids}}) {
 121         my $d = $res->{ids}->{$sid};
 122         my (undef, undef, $name) = PVE::HA::Tools::parse_sid($sid);
 123         $d->{state} = 'enabled' if !defined($d->{state});
 124         $d->{max_restart} = 1 if !defined($d->{max_restart});
 125         $d->{max_relocate} = 1 if !defined($d->{max_relocate});
 126         if (PVE::HA::Resources->lookup($d->{type})) {
 127             if (my $vmd = $vmlist->{ids}->{$name}) {
 128                 if (!$vmd) {
 129                     warn "no such VM '$name'\n";
 130                 } else {
 131                     $d->{node} = $vmd->{node};
 132                     $conf->{$sid} = $d;
 133                 }
 134             } else {
 135                 if (defined($d->{node})) {
 136                     $conf->{$sid} = $d;
 137                 } else {
 138                     warn "service '$sid' without node\n";
 139                 }
 140             }
 141         }
 142     }
 143
 144     return $conf;
 145 }
 146
 147 sub change_service_location {
 148     my ($self, $sid, $current_node, $new_node) = @_;
 149
 150     my (undef, $type, $name) = PVE::HA::Tools::parse_sid($sid);
 151
 152     if(my $plugin = PVE::HA::Resources->lookup($type)) {
 153         my $old = $plugin->config_file($name, $current_node);
 154         my $new = $plugin->config_file($name, $new_node);
 155         rename($old, $new) ||
 156             die "rename '$old' to '$new' failed - $!\n";
 157     } else {
 158         die "implement me";
 159     }
 160 }
 161
 162 sub read_group_config {
 163     my ($self) = @_;
 164
 165     return PVE::HA::Config::read_group_config();
 166 }
 167
 168 # this should return a hash containing info
 169 # what nodes are members and online.
 170 sub get_node_info {
 171     my ($self) = @_;
 172
 173     my ($node_info, $quorate) = ({}, 0);
 174
 175     my $nodename = $self->{nodename};
 176
 177     $quorate = PVE::Cluster::check_cfs_quorum(1) || 0;
 178
 179     my $members = PVE::Cluster::get_members();
 180
 181     foreach my $node (keys %$members) {
 182         my $d = $members->{$node};
 183         $node_info->{$node}->{online} = $d->{online};
 184     }
 185
 186     $node_info->{$nodename}->{online} = 1; # local node is always up
 187
 188     return ($node_info, $quorate);
 189 }
 190
 191 sub log {
 192     my ($self, $level, $msg) = @_;
 193
 194     chomp $msg;
 195
 196     syslog($level, $msg);
 197 }
 198
 199 my $last_lock_status = {};
 200
 201 sub get_pve_lock {
 202     my ($self, $lockid) = @_;
 203
 204     my $got_lock = 0;
 205
 206     my $filename = "$lockdir/$lockid";
 207
 208     my $last = $last_lock_status->{$lockid} || 0;
 209
 210     my $ctime = time();
 211
 212     my $retry = 0;
 213     my $retry_timeout = 100; # fixme: what timeout
 214
 215     eval {
 216
 217         mkdir $lockdir;
 218
 219         # pve cluster filesystem not online
 220         die "can't create '$lockdir' (pmxcfs not mounted?)\n" if ! -d $lockdir;
 221
 222         if ($last && (($ctime - $last) < $retry_timeout)) {
 223              # send cfs lock update request (utime)
 224             if (!utime(0, $ctime, $filename))  {
 225                 $retry = 1;
 226                 die "cfs lock update failed - $!\n";
 227             }
 228         } else {
 229
 230             # fixme: wait some time?
 231             if (!(mkdir $filename)) {
 232                 utime 0, 0, $filename; # cfs unlock request
 233                 die "can't get cfs lock\n";
 234             }
 235         }
 236
 237         $got_lock = 1;
 238     };
 239
 240     my $err = $@;
 241
 242     if ($retry) {
 243         # $self->log('err', $err) if $err; # for debugging
 244         return 0;
 245     }
 246
 247     $last_lock_status->{$lockid} = $got_lock ? $ctime : 0;
 248
 249     if (!!$got_lock != !!$last) {
 250         if ($got_lock) {
 251             $self->log('info', "successfully acquired lock '$lockid'");
 252         } else {
 253             my $msg = "lost lock '$lockid";
 254             $msg .= " - $err" if $err;
 255             $self->log('err', $msg);
 256         }
 257     } else {
 258         # $self->log('err', $err) if $err; # for debugging
 259     }
 260
 261     return $got_lock;
 262 }
 263
 264 sub get_ha_manager_lock {
 265     my ($self) = @_;
 266
 267     return $self->get_pve_lock("ha_manager_lock");
 268 }
 269
 270 # release the cluster wide manager lock.
 271 # when released another CRM may step up and get the lock, thus this should only
 272 # get called when shutting down/deactivating the current master
 273 sub release_ha_manager_lock {
 274     my ($self) = @_;
 275
 276     return rmdir("$lockdir/ha_manager_lock");
 277 }
 278
 279 sub get_ha_agent_lock {
 280     my ($self, $node) = @_;
 281
 282     $node = $self->nodename() if !defined($node);
 283
 284     return $self->get_pve_lock("ha_agent_${node}_lock");
 285 }
 286
 287 # release the respective node agent lock.
 288 # this should only get called if the nodes LRM gracefully shuts down with
 289 # all services already cleanly stopped!
 290 sub release_ha_agent_lock {
 291     my ($self) = @_;
 292
 293     my $node = $self->nodename();
 294
 295     return rmdir("$lockdir/ha_agent_${node}_lock");
 296 }
 297
 298 sub quorate {
 299     my ($self) = @_;
 300
 301     my $quorate = 0;
 302     eval {
 303         $quorate = PVE::Cluster::check_cfs_quorum();
 304     };
 305
 306     return $quorate;
 307 }
 308
 309 sub get_time {
 310     my ($self) = @_;
 311
 312     return time();
 313 }
 314
 315 sub sleep {
 316     my ($self, $delay) = @_;
 317
 318     CORE::sleep($delay);
 319 }
 320
 321 sub sleep_until {
 322    my ($self, $end_time) = @_;
 323
 324    for (;;) {
 325        my $cur_time = time();
 326
 327        last if $cur_time >= $end_time;
 328
 329        $self->sleep(1);
 330    }
 331 }
 332
 333 sub loop_start_hook {
 334     my ($self) = @_;
 335
 336     PVE::Cluster::cfs_update();
 337
 338     $self->{loop_start} = $self->get_time();
 339 }
 340
 341 sub loop_end_hook {
 342     my ($self) = @_;
 343
 344     my $delay = $self->get_time() - $self->{loop_start};
 345
 346     warn "loop take too long ($delay seconds)\n" if $delay > 30;
 347 }
 348
 349 my $watchdog_fh;
 350
 351 sub watchdog_open {
 352     my ($self) = @_;
 353
 354     die "watchdog already open\n" if defined($watchdog_fh);
 355
 356     $watchdog_fh = IO::Socket::UNIX->new(
 357         Type => SOCK_STREAM(),
 358         Peer => "/run/watchdog-mux.sock") ||
 359         die "unable to open watchdog socket - $!\n";
 360
 361     $self->log('info', "watchdog active");
 362 }
 363
 364 sub watchdog_update {
 365     my ($self, $wfh) = @_;
 366
 367     my $res = $watchdog_fh->syswrite("\0", 1);
 368     if (!defined($res)) {
 369         $self->log('err', "watchdog update failed - $!\n");
 370         return 0;
 371     }
 372     if ($res != 1) {
 373         $self->log('err', "watchdog update failed - write $res bytes\n");
 374         return 0;
 375     }
 376
 377     return 1;
 378 }
 379
 380 sub watchdog_close {
 381     my ($self, $wfh) = @_;
 382
 383     $watchdog_fh->syswrite("V", 1); # magic watchdog close
 384     if (!$watchdog_fh->close()) {
 385         $self->log('err', "watchdog close failed - $!");
 386     } else {
 387         $watchdog_fh = undef;
 388         $self->log('info', "watchdog closed (disabled)");
 389     }
 390 }
 391
 392 sub after_fork {
 393     my ($self) = @_;
 394
 395     # close inherited inotify FD from parent and reopen our own
 396     PVE::INotify::inotify_close();
 397     PVE::INotify::inotify_init();
 398
 399     PVE::Cluster::cfs_update();
 400 }
 401
 402 sub get_max_workers {
 403     my ($self) = @_;
 404
 405     my $datacenterconfig = cfs_read_file('datacenter.cfg');
 406
 407     return $datacenterconfig->{max_workers} || 4;
 408 }
 409
 410 1;