]> git.proxmox.com Git - pve-ha-manager.git/blob - src/PVE/HA/NodeStatus.pm
delete node from HA stack when deleted from cluster
[pve-ha-manager.git] / src / PVE / HA / NodeStatus.pm
1 package PVE::HA::NodeStatus;
2
3 use strict;
4 use warnings;
5
6 use Data::Dumper;
7
8 sub new {
9 my ($this, $haenv, $status) = @_;
10
11 my $class = ref($this) || $this;
12
13 my $self = bless {
14 haenv => $haenv,
15 status => $status,
16 last_online => {},
17 }, $class;
18
19 return $self;
20 }
21
22 # possible node state:
23 my $valid_node_states = {
24 online => "node online and member of quorate partition",
25 unknown => "not member of quorate partition, but possibly still running",
26 fence => "node needs to be fenced",
27 gone => "node vanished from cluster members list, possibly deleted"
28 };
29
30 sub get_node_state {
31 my ($self, $node) = @_;
32
33 $self->{status}->{$node} = 'unknown'
34 if !$self->{status}->{$node};
35
36 return $self->{status}->{$node};
37 }
38
39 sub node_is_online {
40 my ($self, $node) = @_;
41
42 return $self->get_node_state($node) eq 'online';
43 }
44
45 sub node_is_offline_delayed {
46 my ($self, $node, $delay) = @_;
47
48 my $haenv = $self->{haenv};
49
50 return undef if $self->get_node_state($node) eq 'online';
51
52 my $last_online = $self->{last_online}->{$node};
53
54 my $ctime = $haenv->get_time();
55
56 if (!defined($last_online)) {
57 $self->{last_online}->{$node} = $ctime;
58 return undef;
59 }
60
61 return ($ctime - $last_online) >= $delay;
62 }
63
64 sub list_nodes {
65 my ($self) = @_;
66
67 return [sort keys %{$self->{status}}];
68 }
69
70 sub list_online_nodes {
71 my ($self) = @_;
72
73 my $res = [];
74
75 foreach my $node (sort keys %{$self->{status}}) {
76 next if $self->{status}->{$node} ne 'online';
77 push @$res, $node;
78 }
79
80 return $res;
81 }
82
83 my $delete_node = sub {
84 my ($self, $node) = @_;
85
86 return undef if $self->get_node_state($node) ne 'gone';
87
88 my $haenv = $self->{haenv};
89
90 delete $self->{last_online}->{$node};
91 delete $self->{status}->{$node};
92
93 $haenv->log('notice', "deleting gone node '$node', not a cluster member".
94 " anymore.");
95 };
96
97 my $set_node_state = sub {
98 my ($self, $node, $state) = @_;
99
100 my $haenv = $self->{haenv};
101
102 die "unknown node state '$state'\n"
103 if !defined($valid_node_states->{$state});
104
105 my $last_state = $self->get_node_state($node);
106
107 return if $state eq $last_state;
108
109 $self->{status}->{$node} = $state;
110
111 $haenv->log('info', "node '$node': state changed from " .
112 "'$last_state' => '$state'\n");
113 };
114
115 sub update {
116 my ($self, $node_info) = @_;
117
118 my $haenv = $self->{haenv};
119
120 foreach my $node (sort keys %$node_info) {
121 my $d = $node_info->{$node};
122 next if !$d->{online};
123
124 # record last time the node was online (required to implement fence delay)
125 $self->{last_online}->{$node} = $haenv->get_time();
126
127 my $state = $self->get_node_state($node);
128
129 if ($state eq 'online') {
130 # &$set_node_state($self, $node, 'online');
131 } elsif ($state eq 'unknown' || $state eq 'gone') {
132 &$set_node_state($self, $node, 'online');
133 } elsif ($state eq 'fence') {
134 # do nothing, wait until fenced
135 } else {
136 die "detected unknown node state '$state";
137 }
138 }
139
140 foreach my $node (keys %{$self->{status}}) {
141 my $d = $node_info->{$node};
142 next if $d && $d->{online};
143
144 my $state = $self->get_node_state($node);
145
146 # node is not inside quorate partition, possibly not active
147
148 if ($state eq 'online') {
149 &$set_node_state($self, $node, 'unknown');
150 } elsif ($state eq 'unknown') {
151
152 # node isn't in the member list anymore, deleted from the cluster?
153 &$set_node_state($self, $node, 'gone') if(!defined($d));
154
155 } elsif ($state eq 'fence') {
156 # do nothing, wait until fenced
157 } elsif($state eq 'gone') {
158 if($self->node_is_offline_delayed($node, 3600)) {
159 &$delete_node($self, $node);
160 }
161 } else {
162 die "detected unknown node state '$state";
163 }
164
165 }
166 }
167
168 # start fencing
169 sub fence_node {
170 my ($self, $node) = @_;
171
172 my $haenv = $self->{haenv};
173
174 my $state = $self->get_node_state($node);
175
176 if ($state ne 'fence') {
177 &$set_node_state($self, $node, 'fence');
178 }
179
180 my $success = $haenv->get_ha_agent_lock($node);
181
182 if ($success) {
183 $haenv->log("info", "fencing: acknowleged - got agent lock for node '$node'");
184 &$set_node_state($self, $node, 'unknown');
185 }
186
187 return $success;
188 }
189
190 1;