retry jobs once per week

lib/App/Netdisco/DB/Result/Admin.pm

@@ -58,14 +58,14 @@ __PACKAGE__->set_primary_key("job");
 
 =head1 RELATIONSHIPS
 
-=head2 device_skips( $backend?, $max_deferrals? )
+=head2 device_skips( $backend?, $max_deferrals?, $retry_after? )
 
 Retuns the set of C<device_skip> entries which apply to this job. They match
 the device IP, current backend, and job action.
 
 You probably want to use the ResultSet method C<skipped> which completes this
-query with a C<backend> host and C<max_deferrals> parameters (or sensible
+query with a C<backend> host, C<max_deferrals>, and C<retry_after> parameters
-defaults).
+(or sensible defaults).
 
 =cut
 
@@ -77,9 +77,13 @@ __PACKAGE__->has_many( device_skips => 'App::Netdisco::DB::Result::DeviceSkip',
       "$args->{foreign_alias}.device"
         => { -ident => "$args->{self_alias}.device" },
       -or => [
-        { "$args->{foreign_alias}.actionset"
+        "$args->{foreign_alias}.actionset"
-            => { '@>' => \"string_to_array($args->{self_alias}.action,'')" } },
+            => { '@>' => \"string_to_array($args->{self_alias}.action,'')" },
-        { "$args->{foreign_alias}.deferrals" => { '>=' => \'?' } },
+        -and => [
+            "$args->{foreign_alias}.deferrals"  => { '>=' => \'?' },
+            "$args->{foreign_alias}.last_defer" =>
+                { '>', \'(LOCALTIMESTAMP - ?::interval)' },
+        ],
       ],
     };
   },

lib/App/Netdisco/DB/ResultSet/Admin.pm

@@ -12,29 +12,26 @@ __PACKAGE__->load_components(qw/
 
 =head1 ADDITIONAL METHODS
 
-=head2 skipped
+=head2 skipped( $backend?, $max_deferrals?, $retry_after? )
 
 Retuns a correlated subquery for the set of C<device_skip> entries that apply
 to some jobs. They match the device IP, current backend, and job action.
 
-Pass the C<backend> FQDN (or the current host will be used as a default), and
+Pass the C<backend> FQDN (or the current host will be used as a default), the
-the C<max_deferrals> (option disabled if 0/undef value is passed).
+C<max_deferrals> (option disabled if 0/undef value is passed), and
+C<retry_after> when devices will be retried once (default 7 days).
 
 =cut
 
 sub skipped {
-  my ($rs, $backend, $max_deferrals) = @_;
+  my ($rs, $backend, $max_deferrals, $retry) = @_;
   $backend ||= (hostfqdn || 'localhost');
   $max_deferrals ||= 10_000_000; # not really 'disabled'
+  $retry ||= '7 days';
 
-  return $rs->correlate('device_skips')->search({
+  return $rs->correlate('device_skips')->search(undef,{
-    -or => [
-      last_defer => undef,
-      last_defer => { '<=', \q{(LOCALTIMESTAMP - INTERVAL '7 days')} },
-    ],
-  },{
     # NOTE: bind param list order is significant
-    bind => [[deferrals => $max_deferrals], [backend => $backend]],
+    bind => [[deferrals => $max_deferrals], [last_defer => $retry], [backend => $backend]],
   });
 }
 

lib/App/Netdisco/JobQueue/PostgreSQL.pm

@@ -44,7 +44,8 @@ sub _getsome {
   my $rs = $jobs->search({
     status => 'queued',
     device => { '-not_in' =>
-      $jobs->skipped($fqdn, setting('workers')->{'max_deferrals'})
+      $jobs->skipped($fqdn, setting('workers')->{'max_deferrals'},
+                            setting('workers')->{'retry_after'})
            ->columns('device')->as_query },
     %$where,
   }, { order_by => 'random()', rows => $num_slots });

lib/App/Netdisco/Manual/Configuration.pod

@@ -1307,6 +1307,7 @@ Value: Settings Tree. Default:
    sleep_time: 1
    min_runtime: 0
    max_deferrals: 10
+   retry_after: '7 days'
 
 Control the activity of the backend daemon with this configuration setting.
 
@@ -1329,6 +1330,9 @@ before the device is no longer polled. The setting and counters are local to
 each poller backend. To reset device counters, restart the backend daemon. To
 disable this feature configure the setting with a value of zero.
 
+C<retry_after> is the time when each job is retried once, even when
+permanently deferred. The default is each job retried once per week.
+
 =head3 C<schedule>
 
 Value: Settings Tree. Default: None.

share/config.yml

@@ -207,6 +207,7 @@ workers:
   sleep_time: 1
   min_runtime: 0
   max_deferrals: 10
+  retry_after: '7 days'
   queue: PostgreSQL
 
 dns: