src/PVE/API2/HA/Status.pm

   1 package PVE::API2::HA::Status;
   2
   3 use strict;
   4 use warnings;
   5
   6 use PVE::SafeSyslog;
   7 use PVE::INotify;
   8 use PVE::Cluster;
   9 use PVE::HA::Config;
  10 use PVE::JSONSchema qw(get_standard_option);
  11 use PVE::RPCEnvironment;
  12
  13 use PVE::RESTHandler;
  14
  15 use base qw(PVE::RESTHandler);
  16
  17 my $nodename = PVE::INotify::nodename();
  18
  19 my $timestamp_to_status = sub {
  20     my ($ctime, $timestamp) = @_;
  21
  22     my $tdiff = $ctime - $timestamp;
  23     if ($tdiff > 30) {
  24         return "old timestamp - dead?";
  25     } elsif ($tdiff < -2) {
  26         return "detected time drift!";
  27     } else {
  28         return "active";
  29     }
  30 };
  31
  32 __PACKAGE__->register_method ({
  33     name => 'index',
  34     path => '',
  35     method => 'GET',
  36     permissions => { user => 'all' },
  37     description => "Directory index.",
  38     parameters => {
  39         additionalProperties => 0,
  40         properties => {},
  41     },
  42     returns => {
  43         type => 'array',
  44         items => {
  45             type => "object",
  46             properties => {},
  47         },
  48         links => [ { rel => 'child', href => "{name}" } ],
  49     },
  50     code => sub {
  51         my ($param) = @_;
  52
  53         my $result = [
  54             { name => 'current' },
  55             { name => 'manager_status' },
  56             ];
  57
  58         return $result;
  59     }});
  60
  61 __PACKAGE__->register_method ({
  62     name => 'status',
  63     path => 'current',
  64     method => 'GET',
  65     description => "Get HA manger status.",
  66     permissions => {
  67         check => ['perm', '/', [ 'Sys.Audit' ]],
  68     },
  69     parameters => {
  70         additionalProperties => 0,
  71         properties => {},
  72     },
  73     returns => { type => 'array' },
  74     code => sub {
  75         my ($param) = @_;
  76
  77         my $res = [];
  78
  79         my $quorate = PVE::Cluster::check_cfs_quorum(1);
  80         if ($quorate) {
  81             push @$res, { id => 'quorum', type => 'quorum',
  82                           node => $nodename, status => "OK", quorate => 1 };
  83         } else {
  84             push @$res, { id => 'quorum', type => 'quorum', node => $nodename,
  85                           status => "No quorum on node '$nodename'!", quorate => 0 };
  86         }
  87
  88         my $status = PVE::HA::Config::read_manager_status();
  89
  90         my $service_config = PVE::HA::Config::read_and_check_resources_config();
  91
  92         my $ctime = time();
  93
  94         if (defined($status->{master_node}) && defined($status->{timestamp})) {
  95             my $master = $status->{master_node};
  96             my $status_str = &$timestamp_to_status($ctime, $status->{timestamp});
  97             # mark crm idle if it has no service configured and is not active
  98             if ($quorate && $status_str ne 'active' && !keys %{$service_config}) {
  99                 $status_str = 'idle';
 100             }
 101             my $time_str = localtime($status->{timestamp});
 102             my $status_text = "$master ($status_str, $time_str)";
 103             push @$res, { id => 'master', type => 'master', node => $master,
 104                           status => $status_text, timestamp => $status->{timestamp} };
 105         }
 106
 107         # compute active services for all nodes
 108         my $active_count = {};
 109         foreach my $sid (sort keys %{$status->{service_status}}) {
 110             my $sd = $status->{service_status}->{$sid};
 111             next if !$sd->{node};
 112             $active_count->{$sd->{node}} = 0 if !defined($active_count->{$sd->{node}});
 113             my $req_state = $sd->{state};
 114             next if !defined($req_state);
 115             next if $req_state eq 'stopped';
 116             next if $req_state eq 'freeze';
 117             $active_count->{$sd->{node}}++;
 118         }
 119
 120         foreach my $node (sort keys %{$status->{node_status}}) {
 121             my $lrm_status = PVE::HA::Config::read_lrm_status($node);
 122             my $id = "lrm:$node";
 123             if (!$lrm_status->{timestamp}) {
 124                 push @$res, { id => $id, type => 'lrm',  node => $node,
 125                               status => "$node (unable to read lrm status)"};
 126             } else {
 127                 my $status_str = &$timestamp_to_status($ctime, $lrm_status->{timestamp});
 128                 if ($status_str eq 'active') {
 129                     my $lrm_mode = $lrm_status->{mode} || 'active';
 130                     my $lrm_state = $lrm_status->{state} || 'unknown';
 131                     if ($lrm_mode ne 'active') {
 132                         $status_str = "$lrm_mode mode";
 133                     } else {
 134                         if ($lrm_state eq 'wait_for_agent_lock' && !$active_count->{$node}) {
 135                             $status_str = 'idle';
 136                         } else {
 137                             $status_str = $lrm_state;
 138                         }
 139                     }
 140                 }
 141
 142                 my $time_str = localtime($lrm_status->{timestamp});
 143                 my $status_text = "$node ($status_str, $time_str)";
 144                 push @$res, { id => $id, type => 'lrm',  node => $node,
 145                               status => $status_text, timestamp => $lrm_status->{timestamp} };
 146             }
 147         }
 148
 149         my $add_service = sub {
 150             my ($sid, $sc, $ss) = @_;
 151
 152             my $data = { id => "service:$sid", type => 'service', sid => $sid };
 153
 154             if ($ss) {
 155                 $data->{node} = $ss->{node};
 156                 $data->{crm_state} = $ss->{state};
 157             } else {
 158                 $data->{node} = $sc->{node};
 159             }
 160
 161             $data->{state} = PVE::HA::Tools::get_verbose_service_state($ss, $sc);
 162             $data->{status} = "$sid ($data->{node}, $data->{state})"; # backward compatibility
 163
 164             # also return common resource attributes
 165             if (defined($sc)) {
 166                 $data->{request_state} = $sc->{state};
 167                 foreach my $key (qw(group max_restart max_relocate comment)) {
 168                     $data->{$key} = $sc->{$key} if defined($sc->{$key});
 169                 }
 170             }
 171
 172             push @$res, $data;
 173         };
 174
 175         foreach my $sid (sort keys %{$status->{service_status}}) {
 176             my $sc = $service_config->{$sid};
 177             my $ss = $status->{service_status}->{$sid};
 178             $add_service->($sid, $sc, $ss);
 179         }
 180
 181         # show also service which aren't yet processed by the CRM
 182         foreach my $sid (sort keys %$service_config) {
 183             next if $status->{service_status}->{$sid};
 184             my $sc = $service_config->{$sid};
 185             $add_service->($sid, $sc);
 186         }
 187
 188         return $res;
 189     }});
 190
 191 __PACKAGE__->register_method ({
 192     name => 'manager_status',
 193     path => 'manager_status',
 194     method => 'GET',
 195     description => "Get full HA manger status, including LRM status.",
 196     permissions => {
 197         check => ['perm', '/', [ 'Sys.Audit' ]],
 198     },
 199     parameters => {
 200         additionalProperties => 0,
 201         properties => {},
 202     },
 203     returns => { type => 'object' },
 204     code => sub {
 205         my ($param) = @_;
 206
 207         my $status = PVE::HA::Config::read_manager_status();
 208
 209         my $data = { manager_status => $status };
 210
 211         $data->{quorum} = {
 212             node => $nodename,
 213             quorate => PVE::Cluster::check_cfs_quorum(1),
 214         };
 215
 216         foreach my $node (sort keys %{$status->{node_status}}) {
 217             my $lrm_status = PVE::HA::Config::read_lrm_status($node);
 218             $data->{lrm_status}->{$node} = $lrm_status;
 219         }
 220
 221         return $data;
 222     }});
 223
 224 1;