1 package DBM::Deep::Engine;
8 our $VERSION = q(0.99_01);
10 use Fcntl qw( :DEFAULT :flock );
14 # * To add to bucket_size, make sure you modify the following:
17 # - add_bucket() - where the buckets are printed
20 # Setup file and tag signatures. These should never change.
22 sub SIG_FILE () { 'DPDB' }
23 sub SIG_HEADER () { 'h' }
24 sub SIG_INTERNAL () { 'i' }
25 sub SIG_HASH () { 'H' }
26 sub SIG_ARRAY () { 'A' }
27 sub SIG_NULL () { 'N' }
28 sub SIG_DATA () { 'D' }
29 sub SIG_INDEX () { 'I' }
30 sub SIG_BLIST () { 'B' }
31 sub SIG_FREE () { 'F' }
32 sub SIG_KEYS () { 'K' }
45 digest => \&Digest::MD5::md5,
49 # Maximum number of buckets per blist before another level of indexing is
50 # done. Increase this value for slightly greater speed, but larger database
51 # files. DO NOT decrease this value below 16, due to risk of recursive
60 if ( defined $args->{pack_size} ) {
61 if ( lc $args->{pack_size} eq 'small' ) {
62 $args->{long_size} = 2;
63 $args->{long_pack} = 'n';
65 elsif ( lc $args->{pack_size} eq 'medium' ) {
66 $args->{long_size} = 4;
67 $args->{long_pack} = 'N';
69 elsif ( lc $args->{pack_size} eq 'large' ) {
70 $args->{long_size} = 8;
71 $args->{long_pack} = 'Q';
74 die "Unknown pack_size value: '$args->{pack_size}'\n";
78 # Grab the parameters we want to use
79 foreach my $param ( keys %$self ) {
80 next unless exists $args->{$param};
81 $self->{$param} = $args->{$param};
83 Scalar::Util::weaken( $self->{obj} ) if $self->{obj};
85 if ( $self->{max_buckets} < 16 ) {
86 warn "Floor of max_buckets is 16. Setting it to 16 from '$self->{max_buckets}'\n";
87 $self->{max_buckets} = 16;
93 sub _fileobj { return $_[0]{fileobj} }
98 # The 2**8 here indicates the number of different characters in the
99 # current hashing algorithm
100 #XXX Does this need to be updated with different hashing algorithms?
101 $self->{index_size} = (2**8) * $self->{long_size};
102 $self->{bucket_size} = $self->{hash_size} + $self->{long_size} * 3;
103 $self->{bucket_list_size} = $self->{max_buckets} * $self->{bucket_size};
108 sub write_file_header {
111 my $loc = $self->_fileobj->request_space( length( SIG_FILE ) + 21 );
113 $self->_fileobj->print_at( $loc,
116 pack('N', 1), # header version
117 pack('N', 12), # header size
118 pack('N', 0), # currently running transaction IDs
119 pack('n', $self->{long_size}),
120 pack('A', $self->{long_pack}),
121 pack('n', $self->{data_size}),
122 pack('A', $self->{data_pack}),
123 pack('n', $self->{max_buckets}),
126 $self->_fileobj->set_transaction_offset( 13 );
131 sub read_file_header {
134 my $buffer = $self->_fileobj->read_at( 0, length(SIG_FILE) + 9 );
135 return unless length($buffer);
137 my ($file_signature, $sig_header, $header_version, $size) = unpack(
141 unless ( $file_signature eq SIG_FILE ) {
142 $self->_fileobj->close;
143 $self->_throw_error( "Signature not found -- file is not a Deep DB" );
146 unless ( $sig_header eq SIG_HEADER ) {
147 $self->_fileobj->close;
148 $self->_throw_error( "Old file version found." );
151 my $buffer2 = $self->_fileobj->read_at( undef, $size );
152 my ($running_transactions, @values) = unpack( 'N n A n A n', $buffer2 );
154 $self->_fileobj->set_transaction_offset( 13 );
156 if ( @values < 5 || grep { !defined } @values ) {
157 $self->_fileobj->close;
158 $self->_throw_error("Corrupted file - bad header");
161 #XXX Add warnings if values weren't set right
162 @{$self}{qw(long_size long_pack data_size data_pack max_buckets)} = @values;
164 return length($buffer) + length($buffer2);
171 # Need to remove use of $fh here
172 my $fh = $self->_fileobj->{fh};
175 #XXX The duplication of calculate_sizes needs to go away
176 unless ( $obj->{base_offset} ) {
177 my $bytes_read = $self->read_file_header;
179 $self->calculate_sizes;
182 # File is empty -- write header and master index
185 $self->_fileobj->audit( "# Database created on" );
187 $self->write_file_header;
189 $obj->{base_offset} = $self->_fileobj->request_space(
190 $self->tag_size( $self->{index_size} ),
194 $obj->_base_offset, $obj->_type,
195 chr(0)x$self->{index_size},
198 # Flush the filehandle
199 my $old_fh = select $fh;
200 my $old_af = $|; $| = 1; $| = $old_af;
204 $obj->{base_offset} = $bytes_read;
207 # Get our type from master index header
209 my $tag = $self->load_tag($obj->_base_offset);
212 $self->_throw_error("Corrupted file, no master index record");
215 unless ($obj->_type eq $tag->{signature}) {
217 $self->_throw_error("File type mismatch");
222 $self->calculate_sizes;
225 #XXX We have to make sure we don't mess up when autoflush isn't turned on
226 $self->_fileobj->set_inode;
236 return SIG_SIZE + $self->{data_size} + $size;
241 # Given offset, signature and content, create tag and write to disk
244 my ($offset, $sig, $content) = @_;
245 my $size = length( $content );
247 $self->_fileobj->print_at(
249 $sig, pack($self->{data_pack}, $size), $content,
252 return unless defined $offset;
256 #XXX Is this even used?
258 offset => $offset + SIG_SIZE + $self->{data_size},
265 # Given offset, load single tag and return signature, size and data
270 my $fileobj = $self->_fileobj;
272 my ($sig, $size) = unpack(
273 "A $self->{data_pack}",
274 $fileobj->read_at( $offset, SIG_SIZE + $self->{data_size} ),
279 #XXX Is this even used?
281 offset => $offset + SIG_SIZE + $self->{data_size},
282 content => $fileobj->read_at( undef, $size ),
288 # Adds one key/value pair to bucket list, given offset, MD5 digest of key,
289 # plain (undigested) key and value.
292 my ($tag, $md5, $plain_key, $value, $deleted, $orig_key) = @_;
295 # This verifies that only supported values will be stored.
297 my $r = Scalar::Util::reftype( $value );
300 last if $r eq 'HASH';
301 last if $r eq 'ARRAY';
304 "Storage of references of type '$r' is not supported."
308 my $fileobj = $self->_fileobj;
310 my $actual_length = $self->_length_needed( $value, $plain_key );
312 #ACID - This is a mutation. Must only find the exact transaction
313 my ($subloc, $offset, $size,$is_deleted) = $self->_find_in_buckets( $tag, $md5, 1 );
316 if ( $fileobj->transaction_id == 0 ) {
317 @transactions = $fileobj->current_transactions;
320 # $self->_release_space( $size, $subloc );
321 # Updating a known md5
322 #XXX This needs updating to use _release_space
325 if ($actual_length <= $size) {
329 $location = $fileobj->request_space( $actual_length );
331 $fileobj->print_at( $tag->{offset} + $offset + $self->{hash_size},
332 pack($self->{long_pack}, $location ),
333 pack($self->{long_pack}, $actual_length ),
334 pack('n n', $fileobj->transaction_id, $deleted ),
338 my $old_value = $self->read_from_loc( $subloc, $orig_key );
339 for ( @transactions ) {
340 my $tag2 = $self->load_tag( $tag->{offset} - SIG_SIZE - $self->{data_size} );
341 $fileobj->{transaction_id} = $_;
342 $self->add_bucket( $tag2, $md5, $orig_key, $old_value, undef, $orig_key );
343 $fileobj->{transaction_id} = 0;
345 $tag = $self->load_tag( $tag->{offset} - SIG_SIZE - $self->{data_size} );
348 elsif ( defined $offset ) {
349 $location = $fileobj->request_space( $actual_length );
351 $fileobj->print_at( $tag->{offset} + $offset,
353 pack($self->{long_pack}, $location ),
354 pack($self->{long_pack}, $actual_length ),
355 pack('n n', $fileobj->transaction_id, $deleted ),
358 for ( @transactions ) {
359 my $tag2 = $self->load_tag( $tag->{offset} - SIG_SIZE - $self->{data_size} );
360 $fileobj->{transaction_id} = $_;
361 $self->add_bucket( $tag2, $md5, '', '', 1, $orig_key );
362 $fileobj->{transaction_id} = 0;
364 $tag = $self->load_tag( $tag->{offset} - SIG_SIZE - $self->{data_size} );
366 # If bucket didn't fit into list, split into a new index level
367 # split_index() will do the $self->_fileobj->request_space() call
368 #XXX It also needs to be transactionally aware
370 $location = $self->split_index( $md5, $tag );
373 $self->write_value( $location, $plain_key, $value, $orig_key );
380 my ($location, $key, $value, $orig_key) = @_;
382 my $fileobj = $self->_fileobj;
384 my $dbm_deep_obj = _get_dbm_object( $value );
385 if ( $dbm_deep_obj && $dbm_deep_obj->_fileobj ne $fileobj ) {
386 $self->_throw_error( "Cannot cross-reference. Use export() instead" );
390 # Write signature based on content type, set content length and write
393 my $r = Scalar::Util::reftype( $value ) || '';
394 if ( $dbm_deep_obj ) {
395 $self->write_tag( $location, SIG_INTERNAL,pack($self->{long_pack}, $dbm_deep_obj->_base_offset) );
397 elsif ($r eq 'HASH') {
398 if ( !$dbm_deep_obj && tied %{$value} ) {
399 $self->_throw_error( "Cannot store something that is tied" );
401 $self->write_tag( $location, SIG_HASH, chr(0)x$self->{index_size} );
403 elsif ($r eq 'ARRAY') {
404 if ( !$dbm_deep_obj && tied @{$value} ) {
405 $self->_throw_error( "Cannot store something that is tied" );
407 $self->write_tag( $location, SIG_ARRAY, chr(0)x$self->{index_size} );
409 elsif (!defined($value)) {
410 $self->write_tag( $location, SIG_NULL, '' );
413 $self->write_tag( $location, SIG_DATA, $value );
417 # Plain key is stored AFTER value, as keys are typically fetched less often.
419 $fileobj->print_at( undef, pack($self->{data_pack}, length($key)) . $key );
421 # Internal references don't care about autobless
422 return 1 if $dbm_deep_obj;
425 # If value is blessed, preserve class name
427 if ( $fileobj->{autobless} ) {
428 if ( defined( my $c = Scalar::Util::blessed($value) ) ) {
429 $fileobj->print_at( undef, chr(1), pack($self->{data_pack}, length($c)) . $c );
432 $fileobj->print_at( undef, chr(0) );
437 # Tie the passed in reference so that changes to it are reflected in the
438 # datafile. The use of $location as the base_offset will act as the
439 # the linkage between parent and child.
441 # The overall assignment is a hack around the fact that just tying doesn't
442 # store the values. This may not be the wrong thing to do.
446 tie %$value, 'DBM::Deep', {
447 base_offset => $location,
449 parent => $self->{obj},
450 parent_key => $orig_key,
454 elsif ($r eq 'ARRAY') {
456 tie @$value, 'DBM::Deep', {
457 base_offset => $location,
459 parent => $self->{obj},
460 parent_key => $orig_key,
470 my ($md5, $tag) = @_;
472 my $fileobj = $self->_fileobj;
474 my $loc = $fileobj->request_space(
475 $self->tag_size( $self->{index_size} ),
478 $fileobj->print_at( $tag->{ref_loc}, pack($self->{long_pack}, $loc) );
480 my $index_tag = $self->write_tag(
482 chr(0)x$self->{index_size},
485 my $newtag_loc = $fileobj->request_space(
486 $self->tag_size( $self->{bucket_list_size} ),
489 my $keys = $tag->{content}
490 . $md5 . pack($self->{long_pack}, $newtag_loc)
491 . pack($self->{long_pack}, 0) # size
492 . pack($self->{long_pack}, 0); # transaction ID
496 # The <= here is deliberate - we have max_buckets+1 keys to iterate
497 # through, unlike every other loop that uses max_buckets as a stop.
498 for (my $i = 0; $i <= $self->{max_buckets}; $i++) {
499 my ($key, $old_subloc, $size) = $self->_get_key_subloc( $keys, $i );
501 die "[INTERNAL ERROR]: No key in split_index()\n" unless $key;
502 die "[INTERNAL ERROR]: No subloc in split_index()\n" unless $old_subloc;
504 my $num = ord(substr($key, $tag->{ch} + 1, 1));
507 my $subkeys = $fileobj->read_at( $newloc[$num], $self->{bucket_list_size} );
509 # This is looking for the first empty spot
510 my ($subloc, $offset, $size) = $self->_find_in_buckets(
511 { content => $subkeys }, '',
515 $newloc[$num] + $offset,
516 $key, pack($self->{long_pack}, $old_subloc),
522 my $loc = $fileobj->request_space(
523 $self->tag_size( $self->{bucket_list_size} ),
527 $index_tag->{offset} + ($num * $self->{long_size}),
528 pack($self->{long_pack}, $loc),
531 my $blist_tag = $self->write_tag(
533 chr(0)x$self->{bucket_list_size},
536 $fileobj->print_at( $blist_tag->{offset}, $key . pack($self->{long_pack}, $old_subloc) );
538 $newloc[$num] = $blist_tag->{offset};
541 $self->_release_space(
542 $self->tag_size( $self->{bucket_list_size} ),
543 $tag->{offset} - SIG_SIZE - $self->{data_size},
551 my ($subloc, $orig_key) = @_;
553 my $fileobj = $self->_fileobj;
555 my $signature = $fileobj->read_at( $subloc, SIG_SIZE );
558 # If value is a hash or array, return new DBM::Deep object with correct offset
560 if (($signature eq SIG_HASH) || ($signature eq SIG_ARRAY)) {
561 my $new_obj = DBM::Deep->new({
563 base_offset => $subloc,
564 fileobj => $self->_fileobj,
565 parent => $self->{obj},
566 parent_key => $orig_key,
569 if ($new_obj->_fileobj->{autobless}) {
571 # Skip over value and plain key to see if object needs
574 $fileobj->increment_pointer( $self->{data_size} + $self->{index_size} );
576 my $size = $fileobj->read_at( undef, $self->{data_size} );
577 $size = unpack($self->{data_pack}, $size);
578 if ($size) { $fileobj->increment_pointer( $size ); }
580 my $bless_bit = $fileobj->read_at( undef, 1 );
581 if ( ord($bless_bit) ) {
584 $fileobj->read_at( undef, $self->{data_size} ),
588 $new_obj = bless $new_obj, $fileobj->read_at( undef, $size );
595 elsif ( $signature eq SIG_INTERNAL ) {
596 my $size = $fileobj->read_at( undef, $self->{data_size} );
597 $size = unpack($self->{data_pack}, $size);
600 my $new_loc = $fileobj->read_at( undef, $size );
601 $new_loc = unpack( $self->{long_pack}, $new_loc );
602 return $self->read_from_loc( $new_loc, $orig_key );
609 # Otherwise return actual value
611 elsif ( $signature eq SIG_DATA ) {
612 my $size = $fileobj->read_at( undef, $self->{data_size} );
613 $size = unpack($self->{data_pack}, $size);
615 my $value = $size ? $fileobj->read_at( undef, $size ) : '';
620 # Key exists, but content is null
625 sub get_bucket_value {
627 # Fetch single value given tag and MD5 digested key.
630 my ($tag, $md5, $orig_key) = @_;
632 #ACID - This is a read. Can find exact or HEAD
633 my ($subloc, $offset, $size, $is_deleted) = $self->_find_in_buckets( $tag, $md5 );
636 #XXX Need to use real key
637 # $self->add_bucket( $tag, $md5, $orig_key, undef, undef, $orig_key );
640 elsif ( !$is_deleted ) {
641 return $self->read_from_loc( $subloc, $orig_key );
649 # Delete single key/value pair given tag and MD5 digested key.
652 my ($tag, $md5, $orig_key) = @_;
654 #ACID - Although this is a mutation, we must find any transaction.
655 # This is because we need to mark something as deleted that is in the HEAD.
656 my ($subloc, $offset, $size,$is_deleted) = $self->_find_in_buckets( $tag, $md5 );
660 my $fileobj = $self->_fileobj;
663 if ( $fileobj->transaction_id == 0 ) {
664 @transactions = $fileobj->current_transactions;
667 if ( $fileobj->transaction_id == 0 ) {
668 my $value = $self->read_from_loc( $subloc, $orig_key );
670 for (@transactions) {
671 $fileobj->{transaction_id} = $_;
672 #XXX Need to use real key
673 $self->add_bucket( $tag, $md5, $orig_key, $value, undef, $orig_key );
674 $fileobj->{transaction_id} = 0;
676 $tag = $self->load_tag( $tag->{offset} - SIG_SIZE - $self->{data_size} );
678 #XXX This needs _release_space() for the value and anything below
680 $tag->{offset} + $offset,
681 substr( $tag->{content}, $offset + $self->{bucket_size} ),
682 chr(0) x $self->{bucket_size},
686 $self->add_bucket( $tag, $md5, '', '', 1, $orig_key );
694 # Check existence of single key given tag and MD5 digested key.
697 my ($tag, $md5) = @_;
699 #ACID - This is a read. Can find exact or HEAD
700 my ($subloc, $offset, $size, $is_deleted) = $self->_find_in_buckets( $tag, $md5 );
701 return ($subloc && !$is_deleted) && 1;
706 # Locate offset for bucket list, given digested key
709 my ($offset, $md5, $args) = @_;
710 $args = {} unless $args;
713 # Locate offset for bucket list using digest index system
715 my $tag = $self->load_tag( $offset )
716 or $self->_throw_error( "INTERNAL ERROR - Cannot find tag" );
719 while ($tag->{signature} ne SIG_BLIST) {
720 my $num = ord substr($md5, $ch, 1);
722 my $ref_loc = $tag->{offset} + ($num * $self->{long_size});
723 $tag = $self->index_lookup( $tag, $num );
726 return if !$args->{create};
728 my $loc = $self->_fileobj->request_space(
729 $self->tag_size( $self->{bucket_list_size} ),
732 $self->_fileobj->print_at( $ref_loc, pack($self->{long_pack}, $loc) );
734 $tag = $self->write_tag(
736 chr(0)x$self->{bucket_list_size},
739 $tag->{ref_loc} = $ref_loc;
746 $tag->{ref_loc} = $ref_loc;
754 # Given index tag, lookup single entry in index and return .
757 my ($tag, $index) = @_;
759 my $location = unpack(
763 $index * $self->{long_size},
768 if (!$location) { return; }
770 return $self->load_tag( $location );
775 # Scan index and recursively step into deeper levels, looking for next key.
778 my ($obj, $offset, $ch, $force_return_next) = @_;
780 my $tag = $self->load_tag( $offset );
782 if ($tag->{signature} ne SIG_BLIST) {
783 my $content = $tag->{content};
784 my $start = $obj->{return_next} ? 0 : ord(substr($obj->{prev_md5}, $ch, 1));
786 for (my $idx = $start; $idx < (2**8); $idx++) {
791 $idx * $self->{long_size},
797 my $result = $self->traverse_index(
798 $obj, $subloc, $ch + 1, $force_return_next,
801 if (defined($result)) { return $result; }
805 $obj->{return_next} = 1;
807 # This is the bucket list
809 my $keys = $tag->{content};
810 if ($force_return_next) { $obj->{return_next} = 1; }
813 # Iterate through buckets, looking for a key match
815 my $transaction_id = $self->_fileobj->transaction_id;
816 for (my $i = 0; $i < $self->{max_buckets}; $i++) {
817 my ($key, $subloc, $size, $trans_id, $is_deleted) = $self->_get_key_subloc( $keys, $i );
820 #XXX Need to find all the copies of this key to find out if $transaction_id has it
821 #XXX marked as deleted, in use, or what.
822 next if $trans_id && $trans_id != $transaction_id;
824 # End of bucket list -- return to outer loop
826 $obj->{return_next} = 1;
829 # Located previous key -- return next one found
830 elsif ($key eq $obj->{prev_md5}) {
831 $obj->{return_next} = 1;
834 # Seek to bucket location and skip over signature
835 elsif ($obj->{return_next}) {
836 my $fileobj = $self->_fileobj;
838 # Skip over value to get to plain key
839 my $sig = $fileobj->read_at( $subloc, SIG_SIZE );
841 my $size = $fileobj->read_at( undef, $self->{data_size} );
842 $size = unpack($self->{data_pack}, $size);
843 if ($size) { $fileobj->increment_pointer( $size ); }
845 # Read in plain key and return as scalar
846 $size = $fileobj->read_at( undef, $self->{data_size} );
847 $size = unpack($self->{data_pack}, $size);
850 if ($size) { $plain_key = $fileobj->read_at( undef, $size); }
855 $obj->{return_next} = 1;
863 # Locate next key, given digested previous one
868 $obj->{prev_md5} = $_[1] ? $_[1] : undef;
869 $obj->{return_next} = 0;
872 # If the previous key was not specifed, start at the top and
873 # return the first one found.
875 if (!$obj->{prev_md5}) {
876 $obj->{prev_md5} = chr(0) x $self->{hash_size};
877 $obj->{return_next} = 1;
880 return $self->traverse_index( $obj, $obj->_base_offset, 0 );
885 sub _get_key_subloc {
887 my ($keys, $idx) = @_;
889 my ($key, $subloc, $size, $transaction_id, $is_deleted) = unpack(
890 # This is 'a', not 'A'. Please read the pack() documentation for the
891 # difference between the two and why it's important.
892 "a$self->{hash_size} $self->{long_pack}2 n2",
895 ($idx * $self->{bucket_size}),
896 $self->{bucket_size},
900 return ($key, $subloc, $size, $transaction_id, $is_deleted);
903 sub _find_in_buckets {
905 my ($tag, $md5, $exact) = @_;
908 my $trans_id = $self->_fileobj->transaction_id;
913 for ( my $i = 0; $i < $self->{max_buckets}; $i++ ) {
914 my ($key, $subloc, $size, $transaction_id, $is_deleted) = $self->_get_key_subloc(
918 my @rv = ($subloc, $i * $self->{bucket_size}, $size, $is_deleted);
921 if ( !$exact && @zero && $trans_id ) {
922 @rv = ($zero[2], $zero[0] * $self->{bucket_size},$zero[3],$is_deleted);
927 next BUCKET if $key ne $md5;
929 # Save off the HEAD in case we need it.
930 @zero = ($i,$key,$subloc,$size,$transaction_id,$is_deleted) if $transaction_id == 0;
932 next BUCKET if $transaction_id != $trans_id;
942 my ($size, $loc) = @_;
946 $self->_fileobj->print_at( $loc,
948 pack($self->{long_pack}, $size ),
949 pack($self->{long_pack}, $next_loc ),
956 die "DBM::Deep: $_[1]\n";
959 sub _get_dbm_object {
964 if ($item->isa( 'DBM::Deep' )) {
971 my $r = Scalar::Util::reftype( $item ) || '';
972 if ( $r eq 'HASH' ) {
975 my $obj = tied(%$item);
976 if ($obj->isa( 'DBM::Deep' )) {
983 elsif ( $r eq 'ARRAY' ) {
986 my $obj = tied(@$item);
987 if ($obj->isa( 'DBM::Deep' )) {
1000 my ($value, $key) = @_;
1002 my $is_dbm_deep = eval {
1003 local $SIG{'__DIE__'};
1004 $value->isa( 'DBM::Deep' );
1008 + $self->{data_size} # size for value
1009 + $self->{data_size} # size for key
1010 + length( $key ); # length of key
1012 if ( $is_dbm_deep && $value->_fileobj eq $self->_fileobj ) {
1013 # long_size is for the internal reference
1014 return $len + $self->{long_size};
1017 if ( $self->_fileobj->{autobless} ) {
1018 # This is for the bit saying whether or not this thing is blessed.
1022 my $r = Scalar::Util::reftype( $value ) || '';
1023 unless ( $r eq 'HASH' || $r eq 'ARRAY' ) {
1024 if ( defined $value ) {
1025 $len += length( $value );
1030 $len += $self->{index_size};
1032 # if autobless is enabled, must also take into consideration
1033 # the class name as it is stored after the key.
1034 if ( $self->_fileobj->{autobless} ) {
1035 my $c = Scalar::Util::blessed($value);
1036 if ( defined $c && !$is_dbm_deep ) {
1037 $len += $self->{data_size} + length($c);