Bug #90134 for BerkeleyDB: memory leak in CDS locking routines

Thu Nov 07 16:06:22 2013 david [...] cobite.com - Ticket created

Subject:	memory leak in CDS locking routines
Date:	Thu, 07 Nov 2013 16:06:00 -0500
To:	bug-BerkeleyDB [...] rt.cpan.org
From:	David Mansfield <david [...] cobite.com>

I've finally tracked down a leak which has bothered me for years! In the package BerkeleyDB::CDS::Lock (in BerkeleyDB.pm), there are two "global" hashes, %Object and %Count. These hashes are keyed by the stringification of the $db object. Entries in these hashes are never deleted, even when the $db object is long gone. This only matters if new $db are being created and destroyed. In my long-running perl daemon, the leak only really becomes significant after months! Some kind of: delete $BerkeleyDB::CDS::Lock::Count{"$db"}; delete $BerkeleyDB::CDS::Lock::Object{"$db"}; is needed, probably when the Count hits 0. -- Thanks, David Mansfield Cobite, INC.

Sat Nov 09 09:34:26 2013 pmqs [...] cpan.org - Correspondence added

Hey David, thanks for the big report. Your analysis is spot on - the Count & Object data structures are not being clean up properly. The code below, from BerkeleyDB.pm, is how I'm going to fix the code if you want to try it out. Will upload an update to CPAN once I smoke test the fix. sub cds_unlock { my $self = shift ; my $db = $self->[0] ; if ($self->[1]) { $self->[1] = 0 ; -- $Count{"$db"} if $Count{"$db"} > 0 ; if ($Count{"$db"} == 0) { $Object{"$db"}->c_close() ; delete $Object{"$db"}; delete $Count{"$db"}; } return 1 ; } return undef ; } Paul

Sat Nov 09 09:34:26 2013 The RT System itself - Status changed from 'new' to 'open'

Sun Nov 10 06:08:47 2013 pmqs [...] cpan.org - Correspondence added

BerkeleyDB 0.54 just uploaded to CPAN. Hope that sorts the issue. Paul