]> git.proxmox.com Git - pve-ha-manager.git/blob - src/PVE/HA/NodeStatus.pm
coding style cleanup
[pve-ha-manager.git] / src / PVE / HA / NodeStatus.pm
1 package PVE::HA::NodeStatus;
2
3 use strict;
4 use warnings;
5
6 use JSON;
7
8 my $fence_delay = 60;
9
10 sub new {
11 my ($this, $haenv, $status) = @_;
12
13 my $class = ref($this) || $this;
14
15 my $self = bless {
16 haenv => $haenv,
17 status => $status,
18 last_online => {},
19 }, $class;
20
21 return $self;
22 }
23
24 # possible node state:
25 my $valid_node_states = {
26 online => "node online and member of quorate partition",
27 unknown => "not member of quorate partition, but possibly still running",
28 fence => "node needs to be fenced",
29 gone => "node vanished from cluster members list, possibly deleted"
30 };
31
32 sub get_node_state {
33 my ($self, $node) = @_;
34
35 $self->{status}->{$node} = 'unknown'
36 if !$self->{status}->{$node};
37
38 return $self->{status}->{$node};
39 }
40
41 sub node_is_online {
42 my ($self, $node) = @_;
43
44 return $self->get_node_state($node) eq 'online';
45 }
46
47 sub node_is_offline_delayed {
48 my ($self, $node, $delay) = @_;
49
50 $delay = $fence_delay if !defined($delay);
51
52 my $haenv = $self->{haenv};
53
54 return undef if $self->get_node_state($node) eq 'online';
55
56 my $last_online = $self->{last_online}->{$node};
57
58 my $ctime = $haenv->get_time();
59
60 if (!defined($last_online)) {
61 $self->{last_online}->{$node} = $ctime;
62 return undef;
63 }
64
65 return ($ctime - $last_online) >= $delay;
66 }
67
68 sub list_nodes {
69 my ($self) = @_;
70
71 return [sort keys %{$self->{status}}];
72 }
73
74 sub list_online_nodes {
75 my ($self) = @_;
76
77 my $res = [];
78
79 foreach my $node (sort keys %{$self->{status}}) {
80 next if $self->{status}->{$node} ne 'online';
81 push @$res, $node;
82 }
83
84 return $res;
85 }
86
87 my $delete_node = sub {
88 my ($self, $node) = @_;
89
90 return undef if $self->get_node_state($node) ne 'gone';
91
92 my $haenv = $self->{haenv};
93
94 delete $self->{last_online}->{$node};
95 delete $self->{status}->{$node};
96
97 $haenv->log('notice', "deleting gone node '$node', not a cluster member".
98 " anymore.");
99 };
100
101 my $set_node_state = sub {
102 my ($self, $node, $state) = @_;
103
104 my $haenv = $self->{haenv};
105
106 die "unknown node state '$state'\n"
107 if !defined($valid_node_states->{$state});
108
109 my $last_state = $self->get_node_state($node);
110
111 return if $state eq $last_state;
112
113 $self->{status}->{$node} = $state;
114
115 $haenv->log('info', "node '$node': state changed from " .
116 "'$last_state' => '$state'\n");
117 };
118
119 sub update {
120 my ($self, $node_info) = @_;
121
122 my $haenv = $self->{haenv};
123
124 foreach my $node (sort keys %$node_info) {
125 my $d = $node_info->{$node};
126 next if !$d->{online};
127
128 # record last time the node was online (required to implement fence delay)
129 $self->{last_online}->{$node} = $haenv->get_time();
130
131 my $state = $self->get_node_state($node);
132
133 if ($state eq 'online') {
134 # &$set_node_state($self, $node, 'online');
135 } elsif ($state eq 'unknown' || $state eq 'gone') {
136 &$set_node_state($self, $node, 'online');
137 } elsif ($state eq 'fence') {
138 # do nothing, wait until fenced
139 } else {
140 die "detected unknown node state '$state";
141 }
142 }
143
144 foreach my $node (sort keys %{$self->{status}}) {
145 my $d = $node_info->{$node};
146 next if $d && $d->{online};
147
148 my $state = $self->get_node_state($node);
149
150 # node is not inside quorate partition, possibly not active
151
152 if ($state eq 'online') {
153 &$set_node_state($self, $node, 'unknown');
154 } elsif ($state eq 'unknown') {
155
156 # node isn't in the member list anymore, deleted from the cluster?
157 &$set_node_state($self, $node, 'gone') if(!defined($d));
158
159 } elsif ($state eq 'fence') {
160 # do nothing, wait until fenced
161 } elsif($state eq 'gone') {
162 if ($self->node_is_offline_delayed($node, 3600)) {
163 &$delete_node($self, $node);
164 }
165 } else {
166 die "detected unknown node state '$state";
167 }
168
169 }
170 }
171
172 # assembles a commont text for fence emails
173 my $send_fence_state_email = sub {
174 my ($self, $subject_prefix, $subject, $node) = @_;
175
176 my $haenv = $self->{haenv};
177
178 my $mail_text = <<EOF
179 The node '$node' failed and needs manual intervention.
180
181 The PVE HA manager tries to fence it and recover the
182 configured HA resources to a healthy node if possible.
183
184 Current fence status: $subject_prefix
185 $subject
186
187
188 Overall Cluster status:
189 -----------------------
190
191 EOF
192 ;
193 my $mail_subject = $subject_prefix . ': ' . $subject;
194
195 my $status = $haenv->read_manager_status();
196 my $data = { manager_status => $status, node_status => $self->{status} };
197
198 $mail_text .= to_json($data, { pretty => 1, canonical => 1});
199
200 $haenv->sendmail($mail_subject, $mail_text);
201 };
202
203
204 # start fencing
205 sub fence_node {
206 my ($self, $node) = @_;
207
208 my $haenv = $self->{haenv};
209
210 my $state = $self->get_node_state($node);
211
212 if ($state ne 'fence') {
213 &$set_node_state($self, $node, 'fence');
214 my $msg = "Try to fence node '$node'";
215 &$send_fence_state_email($self, 'FENCE', $msg, $node);
216 }
217
218 my $success = $haenv->get_ha_agent_lock($node);
219
220 if ($success) {
221 my $msg = "fencing: acknowledged - got agent lock for node '$node'";
222 $haenv->log("info", $msg);
223 &$set_node_state($self, $node, 'unknown');
224 &$send_fence_state_email($self, 'SUCCEED', $msg, $node);
225 }
226
227 return $success;
228 }
229
230 1;