#!/usr/local/bin/perl -w
package Tie::Cache;
use strict;
use vars qw(
$VERSION $Debug $STRUCT_SIZE $REF_SIZE
$BEFORE $AFTER $KEY $VALUE $BYTES $DIRTY
);
$VERSION = .17;
$Debug = 0; # set to 1 for summary, 2 for debug output
$STRUCT_SIZE = 240; # per cached elem bytes overhead, approximate
$REF_SIZE = 16;
# NODE ARRAY STRUCT
$KEY = 0;
$VALUE = 1;
$BYTES = 2;
$BEFORE = 3;
$AFTER = 4;
$DIRTY = 5;
=pod
=head1 NAME
Tie::Cache - LRU Cache in Memory
=head1 SYNOPSIS
use Tie::Cache;
tie %cache, 'Tie::Cache', 100, { Debug => 1 };
tie %cache2, 'Tie::Cache', { MaxCount => 100, MaxBytes => 50000 };
tie %cache3, 'Tie::Cache', 100, { Debug => 1 , WriteSync => 0};
# Options ##################################################################
#
# Debug => 0 - DEFAULT, no debugging output
# 1 - prints cache statistics upon destroying
# 2 - prints detailed debugging info
#
# MaxCount => Maximum entries in cache.
#
# MaxBytes => Maximum bytes taken in memory for cache based on approximate
# size of total cache structure in memory
#
# There is approximately 240 bytes used per key/value pair in the cache for
# the cache data structures, so a cache of 5000 entries would take
# at approximately 1.2M plus the size of the data being cached.
#
# MaxSize => Maximum size of each cache entry. Larger entries are not cached.
# This helps prevent much of the cache being flushed when
# you set an exceptionally large entry. Defaults to MaxBytes/10
#
# WriteSync => 1 - DEFAULT, write() when data is dirtied for
# TRUE CACHE (see below)
# 0 - write() dirty data as late as possible, when leaving
# cache, or when cache is being DESTROY'd
#
############################################################################
# cache supports normal tied hash functions
$cache{1} = 2; # STORE
print "$cache{1}\n"; # FETCH
# FIRSTKEY, NEXTKEY
while(($k, $v) = each %cache) { print "$k: $v\n"; }
delete $cache{1}; # DELETE
%cache = (); # CLEAR
=head1 DESCRIPTION
This module implements a least recently used (LRU) cache in memory
through a tie interface. Any time data is stored in the tied hash,
that key/value pair has an entry time associated with it, and
as the cache fills up, those members of the cache that are
the oldest are removed to make room for new entries.
So, the cache only "remembers" the last written entries, up to the
size of the cache. This can be especially useful if you access
great amounts of data, but only access a minority of the data a
majority of the time.
The implementation is a hash, for quick lookups,
overlaying a doubly linked list for quick insertion and deletion.
On a WinNT PII 300, writes to the hash were done at a rate
3100 per second, and reads from the hash at 6300 per second.
Work has been done to optimize refreshing cache entries that are
frequently read from, code like $cache{entry}, which moves the
entry to the end of the linked list internally.
=cut Documentation continues at the end of the module.
sub TIEHASH {
my($class, $max_count, $options) = @_;
if(ref($max_count)) {
$options = $max_count;
$max_count = $options->{MaxCount};
}
unless($max_count || $options->{MaxBytes}) {
die('you must specify cache size with either MaxBytes or MaxCount');
}
my $sync = exists($options->{WriteSync}) ? $options->{WriteSync} : 1;
my $self = bless
{
# how many items to cache
max_count=> $max_count,
# max bytes to cache
max_bytes => $options->{MaxBytes},
# max size (in bytes) of an individual cache entry
max_size => $options->{MaxSize} || ($options->{MaxBytes} ? (int($options->{MaxBytes}/10) + 1) : 0),
# class track, so know if overridden subs should be used
'class' => $class,
'subclass' => $class ne 'Tie::Cache' ? 1 : 0,
# current sizes
count=>0,
bytes=>0,
# inner structures
head=>0,
tail=>0,
nodes=>{},
'keys'=>[],
# statistics
hit => 0,
miss => 0,
# config
sync => $sync,
dbg => $options->{Debug} || $Debug
}, $class;
if (($self->{max_bytes} && ! $self->{max_size})) {
die("MaxSize must be defined when MaxBytes is");
}
if($self->{max_bytes} and $self->{max_bytes} < 1000) {
die("cannot set MaxBytes to under 1000, each raw entry takes $STRUCT_SIZE bytes alone");
}
if($self->{max_size} && $self->{max_size} < 3) {
die("cannot set MaxSize to under 3 bytes, assuming error in config");
}
$self;
}
# override to write data leaving cache
sub write { undef; }
# commented this section out for speed
# my($self, $key, $value) = @_;
# 1;
#}
# override to get data if not in cache, should return $value
# associated with $key
sub read { undef; }
# commented this section out for speed
# my($self, $key) = @_;
# undef;
#}
sub FETCH {
my($self, $key) = @_;
my $node = $self->{nodes}{$key};
if($node) {
# refresh node's entry
$self->{hit}++; # if $self->{dbg};
# we used to call delete then insert, but we streamlined code
if(my $after = $node->[$AFTER]) {
$self->{dbg} > 1 and $self->print("update() node $node to tail of list");
# reconnect the nodes
my $before = $after->[$BEFORE] = $node->[$BEFORE];
if($before) {
$before->[$AFTER] = $after;
} else {
$self->{head} = $after;
}
# place at the end
$self->{tail}[$AFTER] = $node;
$node->[$BEFORE] = $self->{tail};
$node->[$AFTER] = undef;
$self->{tail} = $node; # always true after this
} else {
# if there is nothing after node, then we are at the end already
# so don't do anything to move the nodes around
die("this node is the tail, so something's wrong")
unless($self->{tail} eq $node);
}
$self->print("FETCH [$key, $node->[$VALUE]]") if ($self->{dbg} > 1);
$node->[$VALUE];
} else {
# we have a cache miss here
$self->{miss}++; # if $self->{dbg};
# its fine to always insert a node, even when we have an undef,
# because even if we aren't a sub-class, we should assume use
# that would then set the entry. This model works well with
# sub-classing and reads() that might want to return undef as
# a valid value.
my $value;
if ($self->{subclass}) {
$self->print("read() for key $key") if $self->{dbg} > 1;
$value = $self->read($key);
}
if(defined $value) {
my $length;
if($self->{max_size}) {
# check max size of entry, that it not exceed max size
$length = &_get_data_length(\$key, \$value);
if($length > $self->{max_size}) {
$self->print("direct read() [$key, $value]") if ($self->{dbg} > 1);
return $value;
}
}
# if we get here, we should insert the new node
$node = &create_node($self, \$key, \$value, $length);
&insert($self, $node);
$value;
} else {
undef;
}
}
}
sub STORE {
my($self, $key, $value) = @_;
my $node;
$self->print("STORE [$key,$value]") if ($self->{dbg} > 1);
# do not cache undefined values
defined($value) || return(undef);
# check max size of entry, that it not exceed max size
my $length;
if($self->{max_size}) {
$length = &_get_data_length(\$key, \$value);
if($length > $self->{max_size}) {
if ($self->{subclass}) {
$self->print("direct write() [$key, $value]") if ($self->{dbg} > 1);
$self->write($key, $value);
}
return $value;
}
}
# do we have node already ?
if($self->{nodes}{$key}) {
$node = &delete($self, $key);
# $node = &delete($self, $key);
# $node->[$VALUE] = $value;
# $node->[$BYTES] = $length || &_get_data_length(\$key, \$value);
}
# insert new node
$node = &create_node($self, \$key, \$value, $length);
# $node ||= &create_node($self, \$key, \$value, $length);
&insert($self, $node);
# if the data is sync'd call write now, otherwise defer the data
# writing, but mark it dirty so it can be cleanup up at the end
if ($self->{subclass}) {
if($self->{sync}) {
$self->print("sync write() [$key, $value]") if $self->{dbg} > 1;
$self->write($key, $value);
} else {
$node->[$DIRTY] = 1;
}
}
$value;
}
sub DELETE {
my($self, $key) = @_;
$self->print("DELETE $key") if ($self->{dbg} > 1);
my $node = $self->delete($key);
$node ? $node->[$VALUE] : undef;
}
sub CLEAR {
my($self) = @_;
$self->print("CLEAR CACHE") if ($self->{dbg} > 1);
if($self->{subclass}) {
my $flushed = $self->flush();
$self->print("FLUSH COUNT $flushed") if ($self->{dbg} > 1);
}
my $node;
while($node = $self->{head}) {
$self->delete($self->{head}[$KEY]);
}
1;
}
sub EXISTS {
my($self, $key) = @_;
exists $self->{nodes}{$key};
}
# firstkey / nextkey emulate keys() and each() behavior by
# taking a snapshot of all the nodes at firstkey, and
# iterating through the keys with nextkey
#
# this method therefore will only supports one each() / keys()
# happening during any given time.
#
sub FIRSTKEY {
my($self) = @_;
$self->{'keys'} = [];
my $node = $self->{head};
while($node) {
push(@{$self->{'keys'}}, $node->[$KEY]);
$node = $node->[$AFTER];
}
shift @{$self->{'keys'}};
}
sub NEXTKEY {
my($self, $lastkey) = @_;
shift @{$self->{'keys'}};
}
sub DESTROY {
my($self) = @_;
# if debugging, snapshot cache before clearing
if($self->{dbg}) {
if($self->{hit} || $self->{miss}) {
$self->{hit_ratio} =
sprintf("%4.3f", $self->{hit} / ($self->{hit} + $self->{miss}));
}
$self->print($self->pretty_self());
if($self->{dbg} > 1) {
$self->print($self->pretty_chains());
}
}
$self->print("DESTROYING") if $self->{dbg} > 1;
$self->CLEAR();
1;
}
####PERL##LRU##TIE##CACHE##PERL##LRU##TIE##CACHE##PERL##LRU##TIE##CACHE
## Helper Routines
####PERL##LRU##TIE##CACHE##PERL##LRU##TIE##CACHE##PERL##LRU##TIE##CACHE
# we use scalar_refs for the data for speed
sub create_node {
my($self, $key, $value, $length) = @_;
(defined($$key) && defined($$value))
|| die("need more localized data than $$key and $$value");
# max_size always defined when max_bytes is
if (($self->{max_size})) {
$length = defined $length ? $length : &_get_data_length($key, $value)
} else {
$length = 0;
}
# ORDER SPECIFIC, see top for NODE ARRAY STRUCT
my $node = [ $$key, $$value, $length ];
}
sub _get_data_length {
my($key, $value) = @_;
my $length = 0;
my %refs;
my @data = ($$key, $$value);
while(my $elem = shift @data) {
next if $refs{$elem};
$refs{$elem} = 1;
if(ref $elem && $elem =~ /(SCALAR|HASH|ARRAY)/) {
my $type = $1;
$length += $REF_SIZE; # guess, 16 bytes per ref, probably more
if (($type eq 'SCALAR')) {
$length += length($$elem);
} elsif (($type eq 'HASH')) {
while (my($k,$v) = each %$elem) {
for my $kv($k,$v) {
if ((ref $kv)) {
push(@data, $kv);
} else {
$length += length($kv);
}
}
}
} elsif (($type eq 'ARRAY')) {
for my $val (@$elem){
if ((ref $val)) {
push(@data, $val);
} else {
$length += length($val);
}
}
}
} else {
$length += length($elem);
}
}
$length;
}
sub insert {
my($self, $new_node) = @_;
$new_node->[$AFTER] = 0;
$new_node->[$BEFORE] = $self->{tail};
$self->print("insert() [$new_node->[$KEY], $new_node->[$VALUE]]") if ($self->{dbg} > 1);
$self->{nodes}{$new_node->[$KEY]} = $new_node;
# current sizes
$self->{count}++;
$self->{bytes} += $new_node->[$BYTES] + $STRUCT_SIZE;
if($self->{tail}) {
$self->{tail}[$AFTER] = $new_node;
} else {
$self->{head} = $new_node;
}
$self->{tail} = $new_node;
## if we are too big now, remove head
while(($self->{max_count} && ($self->{count} > $self->{max_count})) ||
($self->{max_bytes} && ($self->{bytes} > $self->{max_bytes})))
{
if($self->{dbg} > 1) {
$self->print("current/max: ".
"bytes ($self->{bytes}/$self->{max_bytes}) ".
"count ($self->{count}/$self->{max_count}) "
);
}
my $old_node = $self->delete($self->{head}[$KEY]);
if ($self->{subclass}) {
if($old_node->[$DIRTY]) {
$self->print("dirty write() [$old_node->[$KEY], $old_node->[$VALUE]]")
if ($self->{dbg} > 1);
$self->write($old_node->[$KEY], $old_node->[$VALUE]);
}
}
# if($self->{dbg} > 1) {
# $self->print("after delete - bytes $self->{bytes}; count $self->{count}");
# }
}
1;
}
sub delete {
my($self, $key) = @_;
my $node = $self->{nodes}{$key} || return;
# return unless $node;
$self->print("delete() [$key, $node->[$VALUE]]") if ($self->{dbg} > 1);
my $before = $node->[$BEFORE];
my $after = $node->[$AFTER];
# my($before, $after) = $node->{before,after};
if($before) {
($before->[$AFTER] = $after);
} else {
$self->{head} = $after;
}
if($after) {
($after->[$BEFORE] = $before);
} else {
$self->{tail} = $before;
}
delete $self->{nodes}{$key};
$self->{bytes} -= ($node->[$BYTES] + $STRUCT_SIZE);
$self->{count}--;
$node;
}
sub flush {
my $self = shift;
$self->print("FLUSH CACHE") if ($self->{dbg} > 1);
my $node = $self->{head};
my $flush_count = 0;
while($node) {
if($node->[$DIRTY]) {
$self->print("flush dirty write() [$node->[$KEY], $node->[$VALUE]]")
if ($self->{dbg} > 1);
$self->write($node->[$KEY], $node->[$VALUE]);
$node->[$DIRTY] = 0;
$flush_count++;
}
$node = $node->[$AFTER];
}
$flush_count;
}
sub print {
my($self, $msg) = @_;
print "$self: $msg\n";
}
sub pretty_self {
my($self) = @_;
my(@prints);
for(sort keys %{$self}) {
next unless defined $self->{$_};
push(@prints, "$_=>$self->{$_}");
}
"{ " . join(", ", @prints) . " }";
}
sub pretty_chains {
my($self) = @_;
my($str);
my $k = $self->FIRSTKEY();
$str .= "[head]->";
my($curr_node) = $self->{head};
while($curr_node) {
$str .= "[$curr_node->[$KEY],$curr_node->[$VALUE]]->";
$curr_node = $curr_node->[$AFTER];
}
$str .= "[tail]->";
$curr_node = $self->{tail};
while($curr_node) {
$str .= "[$curr_node->[$KEY],$curr_node->[$VALUE]]->";
$curr_node = $curr_node->[$BEFORE];
}
$str .= "[head]";
$str;
}
1;
__END__
=head1 INSTALLATION
Tie::Cache installs easily using the make or nmake commands as
shown below. Otherwise, just copy Cache.pm to $PERLLIB/site/Tie
> perl Makefile.PL
> make
> make test
> make install
* use nmake for win32
** you can also just copy Cache.pm to $perllib/Tie
=head1 BENCMARKS
There is another simpler LRU cache implementation in CPAN,
Tie::Cache::LRU, which has the same basic size limiting
functionality, and for this functionality, the exact same
interface.
Through healthy competition, Michael G Schwern got
Tie::Cache::LRU mostly faster than Tie::Cache on reads & writes:
Cache Size 5000 Tie::Cache 0.17 Tie::Cache::LRU 0.21
10000 Writes 1.55 CPU sec 1.10 CPU sec
40000 Reads 1.82 CPU sec 1.58 CPU sec
10000 Deletes 0.55 CPU sec 0.59 CPU sec
Unless you are using TRUE CACHE or MaxBytes functionality,
using Tie::Cache::LRU should be an easy replacement for Tie::Cache.
=head1 TRUE CACHE
To use class as a true cache, which acts as the sole interface
for some data set, subclass the real cache off Tie::Cache,
with @ISA = qw( 'Tie::Cache' ) notation. Then override
the read() method for behavior when there is a cache miss,
and the write() method for behavior when the cache's data
changes.
When WriteSync is 1 or TRUE (DEFAULT), write() is called immediately
when data in the cache is modified. If set to 0, data that has
been modified in the cache gets written out when the entries are deleted or
during the DESTROY phase of the cache object, usually at the end of
a script.
To have the dirty data write() periodically while WriteSync is set to 0,
there is a flush() cache API call that will flush the dirty writes
in this way. Just call the flush() API like:
my $write_flush_count = tied(%cache)->flush();
The flush() API was added in the .17 release thanks to Rob Bloodgood.
=head1 TRUE CACHE EXAMPLE
use Tie::Cache;
# personalize the Tie::Cache object, by inheriting from it
package My::Cache;
@ISA = qw(Tie::Cache);
# override the read() and write() member functions
# these tell the cache what to do with a cache miss or flush
sub read {
my($self, $key) = @_;
print "cache miss for $key, read() data\n";
rand() * $key;
}
sub write {
my($self, $key, $value) = @_;
print "flushing [$key, $value] from cache, write() data\n";
}
my $cache_size = $ARGV[0] || 2;
my $num_to_cache = $ARGV[1] || 4;
my $Debug = $ARGV[2] || 1;
tie %cache, 'My::Cache', $cache_size, {Debug => $Debug};
# load the cache with new data, each through its contents,
# and then reload in reverse order.
for(1..$num_to_cache) { print "read data $_: $cache{$_}\n" }
while(my($k, $v) = each %cache) { print "each data $k: $v\n"; }
for(my $i=$num_to_cache; $i>0; $i--) { print "read data $i: $cache{$i}\n"; }
# flush writes now, trivial use since will happen in DESTROY() anyway
tied(%cache)->flush();
# clear cache in 2 ways, write will flush out to disk
%cache = ();
undef %cache;
=head1 NOTES
Many thanks to all those who helped me make this module a reality,
including:
:) Tom Hukins who provided me insight and motivation for
finishing this module.
:) Jamie McCarthy, for trying to make Tie::Cache be all
that it can be.
:) Rob Fugina who knows how to "TRULY CACHE".
:) Rob Bloodgood, for the TRUE CACHE flush() API
=head1 AUTHOR
Please send any questions or comments to Joshua Chamas
at chamas@alumni.stanford.org
=head1 COPYRIGHT
Copyright (c) 1999-2002 Joshua Chamas, Chamas Enterprises Inc.
Sponsored by development on NodeWorks http://www.nodeworks.com
All rights reserved. This program is free software;
you can redistribute it and/or modify it under the same
terms as Perl itself.
=cut