NAME

Tie::DB_FileLock - Locking access to Berkeley DB 1.x

SYNOPSIS

use Tie::DB_FileLock;

[$X =] tie %hash, 'Tie::DB_FileLock', [$file, $flags, $mode, $DB_HASH];
[$X =] tie %hash, 'Tie::DB_FileLock', $file, $flags, $mode, $DB_BTREE;

$X->debug($value);

$status = $X->del($key [, $flags]);
$status = $X->put($key, $value [, $flags]);
$status = $X->get($key, $value [, $flags]);
$status = $X->seq($key, $value, $flags);
$status = $X->sync([$flags]);
$status = $X->fd();

# BTREE only
$count = $X->get_dup($key);
@list  = $X->get_dup($key);
%list  = $X->get_dup($key, 1);
$status = $X->find_dup($key, $value);
$status = $X->del_dup($key, $value);

# DBM Filters
$old_filter = $db->filter_store_key  ( sub { ... } );
$old_filter = $db->filter_store_value( sub { ... } );
$old_filter = $db->filter_fetch_key  ( sub { ... } );
$old_filter = $db->filter_fetch_value( sub { ... } );

[undef $X;]
untie %hash;

DESCRIPTION

Module DB_File allows perl to tie hashes to on-disk dbm files, but fails to provide any method by which the hashes might be locked, providing exclusive access or preventing page-level collisions. Tie::DB_FileLock extends DB_File, providing a locking layer using flock().

Unlike Tie::DB_Lock, Tie::DB_FileLock does not duplicate files to allow concurrent access for readers and writers. Tie::DB_FileLock is therefore suitable for large dbms with relatively short locking periods.

Tie::DB_FileLock is designed as a drop-in replacement for DB_File, requiring minimal code changes. Change all occurrences of "DB_File" to "Tie::DB_FileLock" and all should be well. DB_RECNO is not presently supported by Tie::DB_FileLock.

Arguments to Tie::DB_FileLock are identical as those to DB_File. The dbm is locked for shared access if opened RO, exclusively otherwise. The default, as in DB_File, is read/write/create.

Use of the predefined references $DB_HASH, $DB_BTREE, and $DB_RECNO, is identical as with DB_File. When creating your own, the new call is the same, but the object created is a DB_File::XXX thing and not a Tie::DB_FileLock::XXX thing -- therefore error messages will refer to DB_File::XXX.

LOCKING

The locking autoline presented by 'Programming Perl' is incorrect for multiple simultaneous writers. The problem is that a successful flock() lags the tie() by some amount of time. However the snapshot of the on-disk dbm is that from the time of the tie() and not of the flock(), therefore once the flock() succeeds, the dbm may have changed and therefore must be tie()ed again, loading the latest state of the dbm.

Locking cannot be at the level of methods like FETCH() and STORE() because then statements like $hash{$a}++ are not atomic: that is, a different access could (will) take place between the FETCH($a) and the STORE($a + 1).

Therefore locking must occur at a coarser level and the programmer must not dawdle when locks are active. In the case of loops, an effort need be made to untie() the dbm periodically, permitting other processes their due. At some additional cost, a program may yield access to others by breaking a loop like:

tie(%db, 'Tie::DB_FileLock', "arg1.db");
foreach (1..10000) { accesses; }
untie(%db); 

into:

my $dbobj = tie(%db, 'Tie::DB_FileLock', "arg1.db");
foreach (1..10000) {
  accesses;
  if ($_ % 100 == 0) {
     $dbobj->unlockDB();
     $dbobj->lockDB();
  }
}
untie(%db);

SPEED

The first law of thermodynamics holds true for object methods. Access through DB_FileLock is slower than through DB_File. Working with a 40meg dbm containing 110k records, we find

		Keys	Values	Store
DB_File		30s	35s	55s
DB_FileLock	40s	45s	60s

All values +/- ~3s. YMMV. Previous versions of DB_FileLock had a very inefficient FIRSTKEY() routine.

AUTHOR

John M Vinopal, banshee@resort.com

SEE ALSO

DB_File(3p).