-Workaround the known place where 2.3 can leak memory, put notes in on how

to produce the memleak for future fix instead of workaround


git-svn-id: https://svn.code.sf.net/p/xcat/code/xcat-core/trunk@4112 8638fb3e-16cb-4fca-ae20-7b5d299a9bcd

perl-xCAT/xCAT/NodeRange.pm

@@ -11,6 +11,25 @@ our @EXPORT_OK = qw(extnoderange abbreviate_noderange);
 
 my $missingnodes=[];
 my $nodelist; #=xCAT::Table->new('nodelist',-create =>1);
+my $grptab;
+#TODO: MEMLEAK note
+# I've moved grptab up here to avoid calling 'new' on it on every noderange
+# Something is wrong in the Table object such that it leaks
+# a few kilobytes of memory, even if nodelist member is not created
+# To reproduce the mem leak, move 'my $grptab' to the place where it is used
+# then call 'getAllNodesAttribs' a few thousand times on some table
+# No one noticed before 2.3 because the lifetime of processes doing noderange 
+# expansion was short (seconds)
+# In 2.3, the problem has been 'solved' for most contexts in that the DB worker
+# reuses Table objects rather than ever destroying them
+# The exception is when the DB worker process itself wants to expand
+# a noderange, which only ever happens from getAllNodesAttribs
+# in this case, we change NodeRange to reuse the same Table object
+# even if not relying upon DB worker to figure it out for noderange
+# This may be a good idea anyway, regardless of memory leak
+# It remains a good way to induce the memleak to correctly fix it 
+# rather than hiding from the problem
+
 #my $nodeprefix = "node";
 my @allnodeset;
 my $retaincache=0;
@@ -153,7 +172,9 @@ sub expandatom { #TODO: implement table selection as an atom (nodetype.os==rhels
      }
 
     # Try to match groups?
-        my $grptab = xCAT::Table->new('nodegroup'); #TODO: build cache once per noderange and use it instead of repeats
+        unless ($grptab) {
+           $grptab = xCAT::Table->new('nodegroup'); #TODO: build cache once per noderange and use it instead of repeats
+        }
         my @grplist;
         if ($grptab) { 
             @grplist = @{$grptab->getAllEntries()};

perl-xCAT/xCAT/Table.pm

@@ -1,5 +1,13 @@
 # IBM(c) 2007 EPL license http://www.eclipse.org/legal/epl-v10.html
 #TODO:
+#MEMLEAK fix
+# see NodeRange.pm for notes about how to produce a memory leak
+# xCAT as it stands at this moment shouldn't leak anymore due to what is 
+# described there, but that only hides from the real problem and the leak will
+# likely crop up if future architecture changes happen
+# in summary, a created Table object without benefit of db worker thread
+# to abstract its existance will consume a few kilobytes of memory
+# that never gets reused
 # just enough notes to remind me of the design that I think would allow for
 #   -cache to persist so long as '_build_cache' calls concurrently stack (for NodeRange interpretation mainly) (done)
 #   -Allow plugins to define a staleness threshold for getNodesAttribs freshness (complicated enough to postpone...)