do simple fallback if node comes back online from maintenance

[pve-ha-manager.git] / src / PVE / HA / Manager.pm

diff --git a/src/PVE/HA/Manager.pm b/src/PVE/HA/Manager.pm
index 617e3692c7e492a4b640847efd64d3fca6405060..9e46f19d6ff55ef1c2721bd5a8790a1a450c528d 100644 (file)
--- a/src/PVE/HA/Manager.pm
+++ b/src/PVE/HA/Manager.pm
@@ -93,7 +93,7 @@ sub get_node_priority_groups {
 }
 
 sub select_service_node {
-    my ($groups, $online_node_usage, $service_conf, $current_node, $try_next, $tried_nodes) = @_;
+    my ($groups, $online_node_usage, $service_conf, $current_node, $try_next, $tried_nodes, $maintenance_fallback) = @_;
 
     my $group = get_service_group($groups, $online_node_usage, $service_conf);
 
@@ -123,28 +123,31 @@ sub select_service_node {
     } keys %{$pri_groups->{$top_pri}};
 
     my $found;
+    my $found_maintenace_fallback;
     for (my $i = scalar(@nodes) - 1; $i >= 0; $i--) {
        my $node = $nodes[$i];
        if ($node eq $current_node) {
            $found = $i;
-           last;
+       }
+       if (defined($maintenance_fallback) && $node eq $maintenance_fallback) {
+           $found_maintenace_fallback = $i;
        }
     }
 
-    if ($try_next) {
+    if (defined($found_maintenace_fallback)) {
+       return $nodes[$found_maintenace_fallback];
+    }
 
+    if ($try_next) {
        if (defined($found) && ($found < (scalar(@nodes) - 1))) {
            return $nodes[$found + 1];
        } else {
            return $nodes[0];
        }
-
+    } elsif (defined($found)) {
+       return $nodes[$found];
     } else {
-
-       return $nodes[$found] if defined($found);
-
        return $nodes[0];
-
     }
 }
 
@@ -187,6 +190,8 @@ sub recompute_online_node_usage {
                ($state eq 'fence') || ($state eq 'freeze') || ($state eq 'error')) {
                $online_node_usage->{$sd->{node}}++;
            } elsif (($state eq 'migrate') || ($state eq 'relocate')) {
+               # count it for both, source and target as load is put on both
+               $online_node_usage->{$sd->{node}}++;
                $online_node_usage->{$sd->{target}}++;
            } elsif ($state eq 'stopped') {
                # do nothing
@@ -209,6 +214,7 @@ my $change_service_state = sub {
     my $old_state = $sd->{state};
     my $old_node = $sd->{node};
     my $old_failed_nodes = $sd->{failed_nodes};
+    my $old_maintenance_node = $sd->{maintenance_node};
 
     die "no state change" if $old_state eq $new_state; # just to be sure
 
@@ -219,6 +225,7 @@ my $change_service_state = sub {
     $sd->{state} = $new_state;
     $sd->{node} = $old_node;
     $sd->{failed_nodes} = $old_failed_nodes if defined($old_failed_nodes);
+    $sd->{maintenance_node} = $old_maintenance_node if defined($old_maintenance_node);
 
     my $text_state = '';
     foreach my $k (sort keys %params) {
@@ -249,8 +256,9 @@ my $fence_recovery_cleanup = sub {
     # should not happen
     die "unknown resource type '$type'" if !$plugin;
 
-    # locks may block recovery, cleanup those which are safe to remove after fencing
-    my $removable_locks = ['backup', 'mounted'];
+    # locks may block recovery, cleanup those which are safe to remove after fencing,
+    # i.e., after the original node was reset and thus all it's state
+    my $removable_locks = ['backup', 'mounted', 'migrate', 'clone', 'rollback', 'snapshot', 'snapshot-delete', 'suspending', 'suspended'];
     if (my $removed_lock = $plugin->remove_locks($haenv, $id, $removable_locks, $fenced_node)) {
        $haenv->log('warning', "removed leftover lock '$removed_lock' from recovered " .
                    "service '$sid' to allow its start.");
@@ -286,6 +294,7 @@ my $recover_fenced_service = sub {
        &$fence_recovery_cleanup($self, $sid, $fenced_node);
 
        $haenv->steal_service($sid, $sd->{node}, $recovery_node);
+       $self->{online_node_usage}->{$recovery_node}++;
 
        # $sd *is normally read-only*, fencing is the exception
        $cd->{node} = $sd->{node} = $recovery_node;
@@ -369,15 +378,16 @@ sub manage {
 
     my ($haenv, $ms, $ns, $ss) = ($self->{haenv}, $self->{ms}, $self->{ns}, $self->{ss});
 
-    $ns->update($haenv->get_node_info());
+    my ($node_info) = $haenv->get_node_info();
+    my ($lrm_results, $lrm_modes) = $self->read_lrm_status();
+
+    $ns->update($node_info, $lrm_modes);
 
-    if (!$ns->node_is_online($haenv->nodename())) {
+    if (!$ns->node_is_operational($haenv->nodename())) {
        $haenv->log('info', "master seems offline");
        return;
     }
 
-    my ($lrm_results, $lrm_modes) = $self->read_lrm_status();
-
     my $sc = $haenv->read_service_config();
 
     $self->{groups} = $haenv->read_group_config(); # update
@@ -638,7 +648,13 @@ sub next_state_started {
        if ($ns->node_is_offline_delayed($sd->{node})) {
            &$change_service_state($self, $sid, 'fence');
        }
-       return;
+       if ($ns->get_node_state($sd->{node}) ne 'maintenance') {
+           return;
+       } else {
+           # save current node as fallback for when it comes out of
+           # maintenance
+           $sd->{maintenance_node} = $sd->{node};
+       }
     }
 
     if ($cd->{state} eq 'disabled' || $cd->{state} eq 'stopped') {
@@ -663,7 +679,11 @@ sub next_state_started {
                }
            } elsif ($cmd eq 'stop') {
                my $timeout = shift @{$sd->{cmd}};
-               $haenv->log('info', "$cmd service with timeout '$timeout'");
+               if ($timeout == 0) {
+                   $haenv->log('info', "request immediate service hard-stop for service '$sid'");
+               } else {
+                   $haenv->log('info', "request graceful stop with timeout '$timeout' for service '$sid'");
+               }
                &$change_service_state($self, $sid, 'request_stop', timeout => $timeout);
                $haenv->update_service_config($sid, {'state' => 'stopped'});
            } else {
@@ -726,10 +746,29 @@ sub next_state_started {
                }
            }
 
-           my $node = select_service_node($self->{groups}, $self->{online_node_usage},
-                                          $cd, $sd->{node}, $try_next, $sd->{failed_nodes});
+           my $node = select_service_node(
+               $self->{groups},
+               $self->{online_node_usage},
+               $cd,
+               $sd->{node},
+               $try_next,
+               $sd->{failed_nodes},
+               $sd->{maintenance_node},
+           );
 
            if ($node && ($sd->{node} ne $node)) {
+               $self->{online_node_usage}->{$node}++;
+
+               if (defined(my $fallback = $sd->{maintenance_node})) {
+                   if ($node eq $fallback) {
+                       $haenv->log('info', "moving service '$sid' back to '$fallback', node came back from maintenance.");
+                       delete $sd->{maintenance_node};
+                   } elsif ($sd->{node} ne $fallback) {
+                       $haenv->log('info', "dropping maintenance fallback node '$fallback' for '$sid'");
+                       delete $sd->{maintenance_node};
+                   }
+               }
+
                if ($cd->{type} eq 'vm') {
                    $haenv->log('info', "migrate service '$sid' to node '$node' (running)");
                    &$change_service_state($self, $sid, 'migrate', node => $sd->{node}, target => $node);