Do not use crm_node --partition to discover resources

During the promote action, we used to call crm_node --partition to
find other existing nodes and compare LSNs to check if the current node to
promote is the best one. If some node in the partition does not host one of
the clone, we were failing the promotion because we couldn't check its LSN.

The call of crm_node has been replaced by some computation on
environment variables given during notify to find out which clones
on what nodes are actually taking part of the PgSQL cluster. The node
list is then passed to the promote action using a variable stored in a
file localy (outside of the CIB to not break the transition).

This should fix gh issue #9 for old stacks not supporting privte
attributes.

script/pgsqlms

@@ -18,6 +18,7 @@ use POSIX qw(locale_h);
 use File::Spec;
 use File::Temp;
 use Data::Dumper;
+use Storable qw(lock_store lock_retrieve);
 
 use FindBin;
 use lib "$FindBin::RealBin/../lib/";
@@ -202,12 +203,59 @@ sub _rm_stamp {
     return $rc;
 }
 
+# set variable local to the node, non visible to other nodes.
+sub _set_local_var {
+    my ( $name, $val ) = @_;
+    my $storage        = "$HA_RSCTMP/$ENV{'OCF_RESOURCE_INSTANCE'}.bin";
+    my $all            = {};
+    my $rc;
+
+    $all = lock_retrieve($storage) if -r $storage;
+
+    $all->{$name} = $val;
+
+    lock_store( $all, $storage )
+        or die "Can't store data in '$storage'!\n";
+}
+
+# get variable local to the node.
+# Returns the local variable value on success
+# Returns '' on failure
+sub _get_local_var {
+    my ( $name ) = @_;
+    my $storage  = "$HA_RSCTMP/$ENV{'OCF_RESOURCE_INSTANCE'}.bin";
+    my $all      = {};
+    my $rc;
+
+    $all = lock_retrieve($storage) if -r $storage;
+
+    return $all->{$name} if exists $all->{$name};
+
+    return '';
+
+    lock_store( $all, $storage )
+        or die "Can't store data in '$storage'!\n";
+}
+
+# remove variable local to the node.
+sub _rm_local_var {
+    my ( $name ) = @_;
+    my $storage  = "$HA_RSCTMP/$ENV{'OCF_RESOURCE_INSTANCE'}.bin";
+    my $all      = {};
+    my $rc;
+
+    $all = lock_retrieve($storage) if -r $storage;
+
+    delete $all->{$name};
+}
+
 # asynchronous environment cleanup to run during monitor action.
 sub _env_cleanup {
     return unless $OCF_ACTION eq 'monitor';
 
     _delete_attr( 'lsn_location' ) if _get_attr( 'lsn_location' );
     _rm_stamp( 'recover_master' );
+    _rm_local_var( 'nodes' );
 }
 
 # Run the given command as the "system_user" given as parameter.
@@ -683,7 +731,8 @@ sub _confirm_role {
             # The instance is a primary.
             ocf_log( 'debug', "_confirm_role: instance $OCF_RESOURCE_INSTANCE is a primary");
             # Check lsn diff with current slaves if any
-            _check_locations() if $OCF_ACTION eq 'monitor';
+            _check_locations() if $OCF_ACTION eq 'monitor'
+                and not ocf_is_probe();
             return $OCF_RUNNING_MASTER;
         }
 
@@ -1434,9 +1483,9 @@ sub pgsql_promote {
         ocf_log( 'debug',
             'pgsql_promote: checking if current node is the best candidate for promotion');
 
-        # Exclude nodes that are known to be unavailable (not in the current
-        # partition) using the "crm_node" command
-        my @partition_nodes = split /\s/ => qx{ $CRM_NODE --partition };
+        # Only query node that are available as set in local var 'nodes' during
+        # pre-promote action
+        my @active_nodes = split /\s/ => _get_local_var( 'nodes' );
         my $node_to_promote = '';
         my $max_lsn;
         my $node_lsn;
@@ -1469,7 +1518,7 @@ sub pgsql_promote {
             $max_lsn, $max_lsn_dec );
 
         # Now we compare with the other available nodes.
-        foreach my $node ( @partition_nodes ) {
+        foreach my $node ( @active_nodes ) {
             # We exclude the current node from the check.
             next if $node eq $nodename;
 
@@ -1617,6 +1666,7 @@ sub pgsql_notify_pre_promote {
     my @rs;
     my $rc;
     my $node_lsn;
+    my %active_nodes;
 
     ocf_log( 'info', sprintf
         'pgsql_notify: promoting instance on node "%s"',
@@ -1670,6 +1720,25 @@ sub pgsql_notify_pre_promote {
     ocf_log( 'warning', sprintf
         'pgsql_notify: could not set the current node LSN' )
         if $? != 0 ;
+
+    # If this node is the future master, keep track of the slaves that
+    # received the same notification to compare our LSN with them during
+    # promotion
+    if ( $OCF_NOTIFY_ENV{'promote'}[0]{'uname'} eq $nodename ) {
+        my $attr_nodes;
+
+        # build the list of active nodes:
+        #   master + slave + start - stop
+        $active_nodes{ $_->{'uname'} }++ foreach @{ $OCF_NOTIFY_ENV{'master'} };
+        $active_nodes{ $_->{'uname'} }++ foreach @{ $OCF_NOTIFY_ENV{'slave'} };
+        $active_nodes{ $_->{'uname'} }++ foreach @{ $OCF_NOTIFY_ENV{'start'} };
+        $active_nodes{ $_->{'uname'} }-- foreach @{ $OCF_NOTIFY_ENV{'stop'} };
+
+        $attr_nodes = join " "
+            => grep { $active_nodes{$_} > 0 } keys %active_nodes;
+
+        _set_local_var( 'nodes', $attr_nodes );
+    }
 }
 
 # This action is called after a promote action.