3 # Cluster Resource Manager
14 # Server can have several state:
17 wait_for_quorum
=> "cluster is not quorate, waiting",
18 master
=> "quorate, and we got the ha_manager lock",
19 lost_manager_lock
=> "we lost the ha_manager lock (watchgog active)",
20 slave
=> "quorate, but we do not own the ha_manager lock",
24 my ($this, $haenv) = @_;
26 my $class = ref($this) || $this;
31 status
=> { state => 'startup' },
34 $self->set_local_status({ state => 'wait_for_quorum' });
39 sub shutdown_request
{
42 syslog
('info' , "server received shutdown request")
43 if !$self->{shutdown_request
};
45 $self->{shutdown_request
} = 1;
48 sub get_local_status
{
51 return $self->{status
};
54 sub set_local_status
{
55 my ($self, $new) = @_;
57 die "invalid state '$new->{state}'" if !$valid_states->{$new->{state}};
59 my $haenv = $self->{haenv
};
61 my $old = $self->{status
};
63 # important: only update if if really changed
64 return if $old->{state} eq $new->{state};
66 $haenv->log('info', "status change $old->{state} => $new->{state}");
68 $new->{state_change_time
} = $haenv->get_time();
70 $self->{status
} = $new;
72 # fixme: do not use extra class
73 if ($new->{state} eq 'master') {
74 $self->{manager
} = PVE
::HA
::Manager-
>new($haenv);
76 if ($self->{manager
}) {
77 # fixme: what should we do here?
78 $self->{manager
}->cleanup();
79 $self->{manager
} = undef;
84 sub get_protected_ha_manager_lock
{
87 my $haenv = $self->{haenv
};
90 my $starttime = $haenv->get_time();
94 if ($haenv->get_ha_manager_lock()) {
95 if ($self->{ha_manager_wd
}) {
96 $haenv->watchdog_update($self->{ha_manager_wd
});
98 my $wfh = $haenv->watchdog_open();
99 $self->{ha_manager_wd
} = $wfh;
104 last if ++$count > 5; # try max 5 time
106 my $delay = $haenv->get_time() - $starttime;
107 last if $delay > 5; # for max 5 seconds
115 sub check_pending_fencing
{
116 my ($manager_status, $node) = @_;
118 my $ss = $manager_status->{service_status
};
120 return 1 if PVE
::HA
::Tools
::count_fenced_services
($ss, $node);
125 sub do_one_iteration
{
128 my $haenv = $self->{haenv
};
130 my $status = $self->get_local_status();
131 my $state = $status->{state};
133 my $manager_status = $haenv->read_manager_status();
134 my $pending_fencing = check_pending_fencing
($manager_status, $haenv->nodename());
136 # do state changes first
138 if ($state eq 'wait_for_quorum') {
140 if (!$pending_fencing && $haenv->quorate() &&
141 $haenv->service_config_exists()) {
142 if ($self->get_protected_ha_manager_lock()) {
143 $self->set_local_status({ state => 'master' });
145 $self->set_local_status({ state => 'slave' });
149 } elsif ($state eq 'slave') {
151 if (!$pending_fencing && $haenv->quorate() &&
152 $haenv->service_config_exists()) {
153 if ($self->get_protected_ha_manager_lock()) {
154 $self->set_local_status({ state => 'master' });
157 $self->set_local_status({ state => 'wait_for_quorum' });
160 } elsif ($state eq 'lost_manager_lock') {
162 if (!$pending_fencing && $haenv->quorate()) {
163 if ($self->get_protected_ha_manager_lock()) {
164 $self->set_local_status({ state => 'master' });
168 } elsif ($state eq 'master') {
170 if (!$self->get_protected_ha_manager_lock()) {
171 $self->set_local_status({ state => 'lost_manager_lock'});
175 $status = $self->get_local_status();
176 $state = $status->{state};
180 if ($state eq 'wait_for_quorum') {
182 return 0 if $self->{shutdown_request
};
186 } elsif ($state eq 'master') {
188 my $manager = $self->{manager
};
190 die "no manager" if !defined($manager);
192 my $startime = $haenv->get_time();
198 # do work (max_time seconds)
200 # fixme: set alert timer
202 if ($self->{shutdown_request
}) {
204 if ($self->{ha_manager_wd
}) {
205 $haenv->watchdog_close($self->{ha_manager_wd
});
206 delete $self->{ha_manager_wd
};
216 $haenv->log('err', "got unexpected error - $err");
219 return 0 if $shutdown;
221 $haenv->sleep_until($startime + $max_time);
223 } elsif ($state eq 'lost_manager_lock') {
225 if ($self->{ha_manager_wd
}) {
226 $haenv->watchdog_close($self->{ha_manager_wd
});
227 delete $self->{ha_manager_wd
};
230 return 0 if $self->{shutdown_request
};
232 $self->set_local_status({ state => 'wait_for_quorum' });
234 } elsif ($state eq 'slave') {
236 return 0 if $self->{shutdown_request
};
238 # wait until we get master
242 die "got unexpected status '$state'\n";