1 package PVE
::ReplicationState
;
10 use PVE
::CalendarEvent
;
12 use PVE
::GuestHelpers
;
13 use PVE
::ReplicationConfig
;
15 # Note: regression tests can overwrite $state_path for testing
16 our $state_path = "/var/lib/pve-manager/pve-replication-state.json";
17 our $state_lock = "/var/lib/pve-manager/pve-replication-state.lck";
18 our $replicate_logdir = "/var/log/pve/replicate";
20 # regression tests should overwrite this
21 sub job_logfile_name
{
24 return "${replicate_logdir}/$jobid";
27 # Note: We use PVE::Tools::file_set_contents to write state file atomically,
28 # so read_state() always returns an consistent copy (even when not locked).
32 return {} if ! -e
$state_path;
34 my $raw = PVE
::Tools
::file_get_contents
($state_path);
36 return {} if $raw eq '';
39 if ($raw =~ m/^({.*})$/) {
40 return decode_json
($1);
43 die "invalid json data in '$state_path'\n";
46 sub extract_job_state
{
47 my ($stateobj, $jobcfg) = @_;
49 my $plugin = PVE
::ReplicationConfig-
>lookup($jobcfg->{type
});
51 my $vmid = $jobcfg->{guest
};
52 my $tid = $plugin->get_unique_target_id($jobcfg);
53 my $state = $stateobj->{$vmid}->{$tid};
55 $state = {} if !$state;
57 $state->{last_iteration
} //= 0;
58 $state->{last_try
} //= 0; # last sync start time
59 $state->{last_sync
} //= 0; # last successful sync start time
60 $state->{fail_count
} //= 0;
65 sub extract_vmid_tranfer_state
{
66 my ($stateobj, $vmid, $old_target, $new_target) = @_;
68 my $oldid = PVE
::ReplicationConfig
::Cluster-
>get_unique_target_id({ target
=> $old_target });
69 my $newid = PVE
::ReplicationConfig
::Cluster-
>get_unique_target_id({ target
=> $new_target });
71 if (defined(my $vmstate = $stateobj->{$vmid})) {
72 $vmstate->{$newid} = delete($vmstate->{$oldid}) if defined($vmstate->{$oldid});
82 my $stateobj = read_state
();
83 return extract_job_state
($stateobj, $jobcfg);
86 # update state for a single job
87 # pass $state = undef to delete the job state completely
89 my ($jobcfg, $state) = @_;
91 my $plugin = PVE
::ReplicationConfig-
>lookup($jobcfg->{type
});
93 my $vmid = $jobcfg->{guest
};
94 my $tid = $plugin->get_unique_target_id($jobcfg);
98 my $stateobj = read_state
();
99 # Note: tuple ($vmid, $tid) is unique
100 if (defined($state)) {
101 $stateobj->{$vmid}->{$tid} = $state;
103 delete $stateobj->{$vmid}->{$tid};
104 delete $stateobj->{$vmid} if !%{$stateobj->{$vmid}};
106 PVE
::Tools
::file_set_contents
($state_path, encode_json
($stateobj));
110 PVE
::Tools
::lock_file
($state_lock, 10, $update);
114 # make sure we have guest_migration_lock during update
115 PVE
::GuestHelpers
::guest_migration_lock
($vmid, undef, $code);
118 # update all job states related to a specific $vmid
119 sub write_vmid_job_states
{
120 my ($vmid_state, $vmid) = @_;
123 my $stateobj = read_state
();
124 $stateobj->{$vmid} = $vmid_state;
125 PVE
::Tools
::file_set_contents
($state_path, encode_json
($stateobj));
129 PVE
::Tools
::lock_file
($state_lock, 10, $update);
133 # make sure we have guest_migration_lock during update
134 PVE
::GuestHelpers
::guest_migration_lock
($vmid, undef, $code);
137 sub record_job_start
{
138 my ($jobcfg, $state, $start_time, $iteration) = @_;
141 $state->{ptime
} = PVE
::ProcFSTools
::read_proc_starttime
($state->{pid
});
142 $state->{last_node
} = PVE
::INotify
::nodename
();
143 $state->{last_try
} = $start_time;
144 $state->{last_iteration
} = $iteration;
145 $state->{storeid_list
} //= [];
147 write_job_state
($jobcfg, $state);
151 my ($jobcfg, $state, $start_time, $duration, $err) = @_;
153 $state->{duration
} = $duration;
154 delete $state->{pid
};
155 delete $state->{ptime
};
159 $state->{fail_count
}++;
160 $state->{error
} = "$err";
162 $state->{last_sync
} = $start_time;
163 $state->{fail_count
} = 0;
164 delete $state->{error
};
166 write_job_state
($jobcfg, $state);
169 sub replication_snapshot_name
{
170 my ($jobid, $last_sync) = @_;
172 my $prefix = "__replicate_${jobid}_";
173 my $snapname = "${prefix}${last_sync}__";
175 wantarray ?
($prefix, $snapname) : $snapname;
180 my $local_node = PVE
::INotify
::nodename
();
184 my $stateobj = read_state
();
186 my $cfg = PVE
::ReplicationConfig-
>new();
188 my $vms = PVE
::Cluster
::get_vmlist
();
190 foreach my $jobid (sort keys %{$cfg->{ids
}}) {
191 my $jobcfg = $cfg->{ids
}->{$jobid};
192 my $vmid = $jobcfg->{guest
};
194 die "internal error - not implemented" if $jobcfg->{type
} ne 'local';
196 # skip non existing vms
197 next if !$vms->{ids
}->{$vmid};
199 # only consider guest on local node
200 next if $vms->{ids
}->{$vmid}->{node
} ne $local_node;
202 if (!$jobcfg->{remove_job
}) {
203 # never sync to local node
204 next if $jobcfg->{target
} eq $local_node;
206 next if $jobcfg->{disable
};
209 my $state = extract_job_state
($stateobj, $jobcfg);
210 $jobcfg->{state} = $state;
211 $jobcfg->{id
} = $jobid;
212 $jobcfg->{vmtype
} = $vms->{ids
}->{$vmid}->{type
};
216 if ($jobcfg->{remove_job
}) {
217 $next_sync = 1; # lowest possible value
218 # todo: consider fail_count? How many retries?
220 if (my $fail_count = $state->{fail_count
}) {
221 if ($fail_count < 3) {
222 $next_sync = $state->{last_try
} + 5*60*$fail_count;
225 my $schedule = $jobcfg->{schedule
} || '*/15';
226 my $calspec = PVE
::CalendarEvent
::parse_calendar_event
($schedule);
227 $next_sync = PVE
::CalendarEvent
::compute_next_event
($calspec, $state->{last_try
}) // 0;
231 $jobcfg->{next_sync
} = $next_sync;
233 $jobs->{$jobid} = $jobcfg;
240 my ($iteration, $start_time) = @_;
242 my $jobs = job_status
();
244 my $sort_func = sub {
245 my $joba = $jobs->{$a};
246 my $jobb = $jobs->{$b};
247 my $sa = $joba->{state};
248 my $sb = $jobb->{state};
249 my $res = $sa->{last_iteration
} cmp $sb->{last_iteration
};
250 return $res if $res != 0;
251 $res = $joba->{next_sync
} <=> $jobb->{next_sync
};
252 return $res if $res != 0;
253 return $joba->{guest
} <=> $jobb->{guest
};
256 foreach my $jobid (sort $sort_func keys %$jobs) {
257 my $jobcfg = $jobs->{$jobid};
258 next if $jobcfg->{state}->{last_iteration
} >= $iteration;
259 if ($jobcfg->{next_sync
} && ($start_time >= $jobcfg->{next_sync
})) {