X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FDBM%2FDeep%2FEngine.pm;h=0faa0d3d1c162fe4dc80c5067048d6e9456f497d;hb=97d40a0ac43400f8416c3e930c5626c5cb472a55;hp=c180be92c6c01378989dcb0c0e1633c7677918bc;hpb=2603d86e504595571c122c6f7e77b25c1cf4f219;p=dbsrgits%2FDBM-Deep.git diff --git a/lib/DBM/Deep/Engine.pm b/lib/DBM/Deep/Engine.pm index c180be9..0faa0d3 100644 --- a/lib/DBM/Deep/Engine.pm +++ b/lib/DBM/Deep/Engine.pm @@ -1,958 +1,907 @@ package DBM::Deep::Engine; +use 5.006_000; + use strict; +use warnings FATAL => 'all'; + +# Never import symbols into our namespace. We are a class, not a library. +# -RobK, 2008-05-27 +use Scalar::Util (); + +#use Data::Dumper (); -use Fcntl qw( :DEFAULT :flock :seek ); +# File-wide notes: +# * Every method in here assumes that the storage has been appropriately +# safeguarded. This can be anything from flock() to some sort of manual +# mutex. But, it's the caller's responsability to make sure that this has +# been done. -## # Setup file and tag signatures. These should never change. -## sub SIG_FILE () { 'DPDB' } -sub SIG_INTERNAL () { 'i' } +sub SIG_HEADER () { 'h' } sub SIG_HASH () { 'H' } sub SIG_ARRAY () { 'A' } sub SIG_NULL () { 'N' } sub SIG_DATA () { 'D' } sub SIG_INDEX () { 'I' } sub SIG_BLIST () { 'B' } +sub SIG_FREE () { 'F' } sub SIG_SIZE () { 1 } -sub precalc_sizes { - ## - # Precalculate index, bucket and bucket list sizes - ## - my $self = shift; - - $self->{index_size} = (2**8) * $self->{long_size}; - $self->{bucket_size} = $self->{hash_size} + $self->{long_size}; - $self->{bucket_list_size} = $self->{max_buckets} * $self->{bucket_size}; - - return 1; -} - -sub set_pack { - ## - # Set pack/unpack modes (see file header for more) - ## - my $self = shift; - my ($long_s, $long_p, $data_s, $data_p) = @_; - - ## - # Set to 4 and 'N' for 32-bit offset tags (default). Theoretical limit of 4 - # GB per file. - # (Perl must be compiled with largefile support for files > 2 GB) - # - # Set to 8 and 'Q' for 64-bit offsets. Theoretical limit of 16 XB per file. - # (Perl must be compiled with largefile and 64-bit long support) - ## - $self->{long_size} = $long_s ? $long_s : 4; - $self->{long_pack} = $long_p ? $long_p : 'N'; - - ## - # Set to 4 and 'N' for 32-bit data length prefixes. Limit of 4 GB for each - # key/value. Upgrading this is possible (see above) but probably not - # necessary. If you need more than 4 GB for a single key or value, this - # module is really not for you :-) - ## - $self->{data_size} = $data_s ? $data_s : 4; - $self->{data_pack} = $data_p ? $data_p : 'N'; - - return $self->precalc_sizes(); -} - -sub set_digest { - ## - # Set key digest function (default is MD5) - ## - my $self = shift; - my ($digest_func, $hash_size) = @_; - - $self->{digest} = $digest_func ? $digest_func : \&Digest::MD5::md5; - $self->{hash_size} = $hash_size ? $hash_size : 16; - - return $self->precalc_sizes(); -} +our $STALE_SIZE = 2; + +# Please refer to the pack() documentation for further information +my %StP = ( + 1 => 'C', # Unsigned char value (no order needed as it's just one byte) + 2 => 'n', # Unsigned short in "network" (big-endian) order + 4 => 'N', # Unsigned long in "network" (big-endian) order + 8 => 'Q', # Usigned quad (no order specified, presumably machine-dependent) +); +sub StP { $StP{$_[1]} } + +# Import these after the SIG_* definitions because those definitions are used +# in the headers of these classes. -RobK, 2008-06-20 +use DBM::Deep::Engine::Sector::BucketList; +use DBM::Deep::Engine::Sector::FileHeader; +use DBM::Deep::Engine::Sector::Index; +use DBM::Deep::Engine::Sector::Null; +use DBM::Deep::Engine::Sector::Reference; +use DBM::Deep::Engine::Sector::Scalar; +use DBM::Deep::Iterator; + +################################################################################ sub new { my $class = shift; my ($args) = @_; - my $self = bless { - long_size => 4, - long_pack => 'N', - data_size => 4, - data_pack => 'N', - - digest => \&Digest::MD5::md5, - hash_size => 16, - - ## - # Maximum number of buckets per list before another level of indexing is - # done. - # Increase this value for slightly greater speed, but larger database - # files. DO NOT decrease this value below 16, due to risk of recursive - # reindex overrun. - ## - max_buckets => 16, - }, $class; - - $self->precalc_sizes; - - return $self; -} - -sub setup_fh { - my $self = shift; - my ($obj) = @_; - - $self->open( $obj ) if !defined $obj->_fh; + $args->{storage} = DBM::Deep::File->new( $args ) + unless exists $args->{storage}; - my $fh = $obj->_fh; - flock $fh, LOCK_EX; - - unless ( $obj->{base_offset} ) { - seek($fh, 0 + $obj->_root->{file_offset}, SEEK_SET); - my $signature; - my $bytes_read = read( $fh, $signature, length(SIG_FILE)); + my $self = bless { + byte_size => 4, - ## - # File is empty -- write signature and master index - ## - if (!$bytes_read) { - my $loc = $self->_request_space( $obj, length( SIG_FILE ) ); - seek($fh, $loc + $obj->_root->{file_offset}, SEEK_SET); - print( $fh SIG_FILE); + digest => undef, + hash_size => 16, # In bytes + hash_chars => 256, # Number of chars the algorithm uses per byte + max_buckets => 16, + num_txns => 1, # The HEAD + trans_id => 0, # Default to the HEAD - $obj->{base_offset} = $self->_request_space( - $obj, $self->tag_size( $self->{index_size} ), - ); + data_sector_size => 64, # Size in bytes of each data sector - $self->create_tag( - $obj, $obj->_base_offset, $obj->_type, - chr(0) x $self->{index_size}, - ); + entries => {}, # This is the list of entries for transactions + storage => undef, + }, $class; - # Flush the filehandle - my $old_fh = select $fh; - my $old_af = $|; $| = 1; $| = $old_af; - select $old_fh; + # Never allow byte_size to be set directly. + delete $args->{byte_size}; + if ( defined $args->{pack_size} ) { + if ( lc $args->{pack_size} eq 'small' ) { + $args->{byte_size} = 2; + } + elsif ( lc $args->{pack_size} eq 'medium' ) { + $args->{byte_size} = 4; + } + elsif ( lc $args->{pack_size} eq 'large' ) { + $args->{byte_size} = 8; } else { - $obj->{base_offset} = $bytes_read; - - ## - # Check signature was valid - ## - unless ($signature eq SIG_FILE) { - $self->close_fh( $obj ); - $obj->_throw_error("Signature not found -- file is not a Deep DB"); - } + DBM::Deep->_throw_error( "Unknown pack_size value: '$args->{pack_size}'" ); + } + } - ## - # Get our type from master index signature - ## - my $tag = $self->load_tag($obj, $obj->_base_offset) - or $obj->_throw_error("Corrupted file, no master index record"); + # Grab the parameters we want to use + foreach my $param ( keys %$self ) { + next unless exists $args->{$param}; + $self->{$param} = $args->{$param}; + } - unless ($obj->{type} eq $tag->{signature}) { - $obj->_throw_error("File type mismatch"); - } + my %validations = ( + max_buckets => { floor => 16, ceil => 256 }, + num_txns => { floor => 1, ceil => 255 }, + data_sector_size => { floor => 32, ceil => 256 }, + ); + + while ( my ($attr, $c) = each %validations ) { + if ( !defined $self->{$attr} + || !length $self->{$attr} + || $self->{$attr} =~ /\D/ + || $self->{$attr} < $c->{floor} + ) { + $self->{$attr} = '(undef)' if !defined $self->{$attr}; + warn "Floor of $attr is $c->{floor}. Setting it to $c->{floor} from '$self->{$attr}'\n"; + $self->{$attr} = $c->{floor}; + } + elsif ( $self->{$attr} > $c->{ceil} ) { + warn "Ceiling of $attr is $c->{ceil}. Setting it to $c->{ceil} from '$self->{$attr}'\n"; + $self->{$attr} = $c->{ceil}; } } - #XXX We have to make sure we don't mess up when autoflush isn't turned on - unless ( $obj->_root->{inode} ) { - my @stats = stat($obj->_fh); - $obj->_root->{inode} = $stats[1]; - $obj->_root->{end} = $stats[7]; + if ( !$self->{digest} ) { + require Digest::MD5; + $self->{digest} = \&Digest::MD5::md5; } - flock $fh, LOCK_UN; - - return 1; + return $self; } -sub open { - ## - # Open a fh to the database, create if nonexistent. - # Make sure file signature matches DBM::Deep spec. - ## - my $self = shift; - my ($obj) = @_; +################################################################################ - # Theoretically, adding O_BINARY should remove the need for the binmode - # Of course, testing it is going to be ... interesting. - my $flags = O_RDWR | O_CREAT | O_BINARY; - - my $fh; - my $filename = $obj->_root->{file}; - sysopen( $fh, $filename, $flags ) - or $obj->_throw_error("Cannot sysopen file '$filename': $!"); - $obj->_root->{fh} = $fh; +sub read_value { + my $self = shift; + my ($obj, $key) = @_; - #XXX Can we remove this by using the right sysopen() flags? - # Maybe ... q.v. above - binmode $fh; # for win32 + # This will be a Reference sector + my $sector = $self->_load_sector( $obj->_base_offset ) + or return; - if ($obj->_root->{autoflush}) { - my $old = select $fh; - $|=1; - select $old; + if ( $sector->staleness != $obj->_staleness ) { + return; } - return 1; -} + my $key_md5 = $self->_apply_digest( $key ); -sub close_fh { - my $self = shift; - my ($obj) = @_; + my $value_sector = $sector->get_data_for({ + key_md5 => $key_md5, + allow_head => 1, + }); - if ( my $fh = $obj->_root->{fh} ) { - close $fh; - } - $obj->_root->{fh} = undef; + unless ( $value_sector ) { + $value_sector = DBM::Deep::Engine::Sector::Null->new({ + engine => $self, + data => undef, + }); - return 1; -} + $sector->write_data({ + key_md5 => $key_md5, + key => $key, + value => $value_sector, + }); + } -sub tag_size { - my $self = shift; - my ($size) = @_; - return SIG_SIZE + $self->{data_size} + $size; + return $value_sector->data; } -sub create_tag { - ## - # Given offset, signature and content, create tag and write to disk - ## +sub get_classname { my $self = shift; - my ($obj, $offset, $sig, $content) = @_; - my $size = length($content); + my ($obj) = @_; - my $fh = $obj->_fh; + # This will be a Reference sector + my $sector = $self->_load_sector( $obj->_base_offset ) + or DBM::Deep->_throw_error( "How did get_classname fail (no sector for '$obj')?!" ); - seek($fh, $offset + $obj->_root->{file_offset}, SEEK_SET); - print( $fh $sig . pack($self->{data_pack}, $size) . $content ); + if ( $sector->staleness != $obj->_staleness ) { + return; + } - return { - signature => $sig, - size => $size, - offset => $offset + SIG_SIZE + $self->{data_size}, - content => $content - }; + return $sector->get_classname; } -sub load_tag { - ## - # Given offset, load single tag and return signature, size and data - ## +sub make_reference { my $self = shift; - my ($obj, $offset) = @_; + my ($obj, $old_key, $new_key) = @_; -# print join(':',map{$_||''}caller(1)), $/; + # This will be a Reference sector + my $sector = $self->_load_sector( $obj->_base_offset ) + or DBM::Deep->_throw_error( "How did make_reference fail (no sector for '$obj')?!" ); - my $fh = $obj->_fh; + if ( $sector->staleness != $obj->_staleness ) { + return; + } - seek($fh, $offset + $obj->_root->{file_offset}, SEEK_SET); + my $old_md5 = $self->_apply_digest( $old_key ); - #XXX I'm not sure this check will work if autoflush isn't enabled ... - return if eof $fh; + my $value_sector = $sector->get_data_for({ + key_md5 => $old_md5, + allow_head => 1, + }); - my $b; - read( $fh, $b, SIG_SIZE + $self->{data_size} ); - my ($sig, $size) = unpack( "A $self->{data_pack}", $b ); + unless ( $value_sector ) { + $value_sector = DBM::Deep::Engine::Sector::Null->new({ + engine => $self, + data => undef, + }); - my $buffer; - read( $fh, $buffer, $size); + $sector->write_data({ + key_md5 => $old_md5, + key => $old_key, + value => $value_sector, + }); + } - return { - signature => $sig, - size => $size, - offset => $offset + SIG_SIZE + $self->{data_size}, - content => $buffer - }; + if ( $value_sector->isa( 'DBM::Deep::Engine::Sector::Reference' ) ) { + $sector->write_data({ + key => $new_key, + key_md5 => $self->_apply_digest( $new_key ), + value => $value_sector, + }); + $value_sector->increment_refcount; + } + else { + $sector->write_data({ + key => $new_key, + key_md5 => $self->_apply_digest( $new_key ), + value => $value_sector->clone, + }); + } } -sub _length_needed { +sub key_exists { my $self = shift; - my ($obj, $value) = @_; - - my $is_dbm_deep = eval { - local $SIG{'__DIE__'}; - $value->isa( 'DBM::Deep' ); - }; + my ($obj, $key) = @_; - my $internal_ref = $is_dbm_deep && ($value->_root eq $obj->_root); + # This will be a Reference sector + my $sector = $self->_load_sector( $obj->_base_offset ) + or return ''; - if ( $internal_ref ) { - return $self->{long_size}; + if ( $sector->staleness != $obj->_staleness ) { + return ''; } - my $r = Scalar::Util::reftype( $value ) || ''; - unless ( $r eq 'HASH' || $r eq 'ARRAY' ) { - return length( $value ); - } + my $data = $sector->get_data_for({ + key_md5 => $self->_apply_digest( $key ), + allow_head => 1, + }); - my $actual_length = $self->{index_size}; + # exists() returns 1 or '' for true/false. + return $data ? 1 : ''; +} - # if autobless is enabled, must also take into consideration - # the class name, as it is stored along with key/value. - if ( $obj->_root->{autobless} ) { - my $value_class = Scalar::Util::blessed($value); - if ( defined $value_class && !$value->isa('DBM::Deep') ) { - $actual_length += length($value_class); - } +sub delete_key { + my $self = shift; + my ($obj, $key) = @_; + + my $sector = $self->_load_sector( $obj->_base_offset ) + or return; + + if ( $sector->staleness != $obj->_staleness ) { + return; } - return $actual_length; + return $sector->delete_key({ + key_md5 => $self->_apply_digest( $key ), + allow_head => 0, + }); } -sub add_bucket { - ## - # Adds one key/value pair to bucket list, given offset, MD5 digest of key, - # plain (undigested) key and value. - ## +sub write_value { my $self = shift; - my ($obj, $tag, $md5, $plain_key, $value) = @_; + my ($obj, $key, $value) = @_; - # This verifies that only supported values will be stored. + my $r = Scalar::Util::reftype( $value ) || ''; { - my $r = Scalar::Util::reftype( $value ); - last if !defined $r; - + last if $r eq ''; last if $r eq 'HASH'; last if $r eq 'ARRAY'; - $obj->_throw_error( - "Storage of variables of type '$r' is not supported." + DBM::Deep->_throw_error( + "Storage of references of type '$r' is not supported." ); } - my $location = 0; - my $result = 2; + # This will be a Reference sector + my $sector = $self->_load_sector( $obj->_base_offset ) + or DBM::Deep->_throw_error( "1: Cannot write to a deleted spot in DBM::Deep." ); - my $root = $obj->_root; + if ( $sector->staleness != $obj->_staleness ) { + DBM::Deep->_throw_error( "2: Cannot write to a deleted spot in DBM::Deep." ); + } - my $fh = $obj->_fh; + my ($class, $type); + if ( !defined $value ) { + $class = 'DBM::Deep::Engine::Sector::Null'; + } + elsif ( $r eq 'ARRAY' || $r eq 'HASH' ) { + my $tmpvar; + if ( $r eq 'ARRAY' ) { + $tmpvar = tied @$value; + } elsif ( $r eq 'HASH' ) { + $tmpvar = tied %$value; + } - my ($subloc, $offset) = $self->_find_in_buckets( $tag, $md5 ); + if ( $tmpvar ) { + my $is_dbm_deep = eval { local $SIG{'__DIE__'}; $tmpvar->isa( 'DBM::Deep' ); }; - # Updating a known md5 - if ( $subloc ) { - $result = 1; + unless ( $is_dbm_deep ) { + DBM::Deep->_throw_error( "Cannot store something that is tied." ); + } - seek($fh, $subloc + SIG_SIZE + $root->{file_offset}, SEEK_SET); - my $size; - read( $fh, $size, $self->{data_size}); - $size = unpack($self->{data_pack}, $size); + unless ( $tmpvar->_engine->storage == $self->storage ) { + DBM::Deep->_throw_error( "Cannot store values across DBM::Deep files. Please use export() instead." ); + } - my $actual_length = $self->_length_needed( $obj, $value ); + # First, verify if we're storing the same thing to this spot. If we are, then + # this should be a no-op. -EJS, 2008-05-19 + my $loc = $sector->get_data_location_for({ + key_md5 => $self->_apply_digest( $key ), + allow_head => 1, + }); - if ($actual_length <= $size) { - $location = $subloc; - } - else { - $location = $root->{end}; - seek( - $fh, - $tag->{offset} + $offset + $self->{hash_size} + $root->{file_offset}, - SEEK_SET, - ); - print( $fh pack($self->{long_pack}, $location) ); + if ( defined($loc) && $loc == $tmpvar->_base_offset ) { + return 1; + } + + #XXX Can this use $loc? + my $value_sector = $self->_load_sector( $tmpvar->_base_offset ); + $sector->write_data({ + key => $key, + key_md5 => $self->_apply_digest( $key ), + value => $value_sector, + }); + $value_sector->increment_refcount; + + return 1; } - } - # Adding a new md5 - elsif ( defined $offset ) { - $location = $root->{end}; - seek( $fh, $tag->{offset} + $offset + $root->{file_offset}, SEEK_SET ); - print( $fh $md5 . pack($self->{long_pack}, $location) ); + $class = 'DBM::Deep::Engine::Sector::Reference'; + $type = substr( $r, 0, 1 ); } - # If bucket didn't fit into list, split into a new index level else { - $self->split_index( $obj, $md5, $tag ); - - $location = $root->{end}; + if ( tied($value) ) { + DBM::Deep->_throw_error( "Cannot store something that is tied." ); + } + $class = 'DBM::Deep::Engine::Sector::Scalar'; } - $self->write_value( $obj, $location, $plain_key, $value ); + # Create this after loading the reference sector in case something bad happens. + # This way, we won't allocate value sector(s) needlessly. + my $value_sector = $class->new({ + engine => $self, + data => $value, + type => $type, + }); + + $sector->write_data({ + key => $key, + key_md5 => $self->_apply_digest( $key ), + value => $value_sector, + }); + + # This code is to make sure we write all the values in the $value to the disk + # and to make sure all changes to $value after the assignment are reflected + # on disk. This may be counter-intuitive at first, but it is correct dwimmery. + # NOTE - simply tying $value won't perform a STORE on each value. Hence, the + # copy to a temp value. + if ( $r eq 'ARRAY' ) { + my @temp = @$value; + tie @$value, 'DBM::Deep', { + base_offset => $value_sector->offset, + staleness => $value_sector->staleness, + storage => $self->storage, + engine => $self, + }; + @$value = @temp; + bless $value, 'DBM::Deep::Array' unless Scalar::Util::blessed( $value ); + } + elsif ( $r eq 'HASH' ) { + my %temp = %$value; + tie %$value, 'DBM::Deep', { + base_offset => $value_sector->offset, + staleness => $value_sector->staleness, + storage => $self->storage, + engine => $self, + }; + + %$value = %temp; + bless $value, 'DBM::Deep::Hash' unless Scalar::Util::blessed( $value ); + } - return $result; + return 1; } -sub write_value { +# XXX Add staleness here +sub get_next_key { my $self = shift; - my ($obj, $location, $key, $value) = @_; + my ($obj, $prev_key) = @_; + + # XXX Need to add logic about resetting the iterator if any key in the reference has changed + unless ( $prev_key ) { + $obj->{iterator} = DBM::Deep::Iterator->new({ + base_offset => $obj->_base_offset, + engine => $self, + }); + } - my $fh = $obj->_fh; - my $root = $obj->_root; + return $obj->{iterator}->get_next_key( $obj ); +} - my $is_dbm_deep = eval { - local $SIG{'__DIE__'}; - $value->isa( 'DBM::Deep' ); - }; +################################################################################ - my $internal_ref = $is_dbm_deep && ($value->_root eq $root); +sub setup_fh { + my $self = shift; + my ($obj) = @_; + + return 1 if $obj->_base_offset; - seek($fh, $location + $root->{file_offset}, SEEK_SET); + my $header = $self->_load_header; - ## - # Write signature based on content type, set content length and write - # actual value. - ## - my $r = Scalar::Util::reftype($value) || ''; - my $content_length; - if ( $internal_ref ) { - print( $fh SIG_INTERNAL ); - print( $fh pack($self->{data_pack}, $self->{long_size}) ); - print( $fh pack($self->{long_pack}, $value->_base_offset) ); - $content_length = $self->{long_size}; + # Creating a new file + if ( $header->is_new ) { + # 1) Create Array/Hash entry + my $sector = DBM::Deep::Engine::Sector::Reference->new({ + engine => $self, + type => $obj->_type, + }); + $obj->{base_offset} = $sector->offset; + $obj->{staleness} = $sector->staleness; + + $self->flush; } + # Reading from an existing file else { - if ($r eq 'HASH') { - print( $fh SIG_HASH ); - print( $fh pack($self->{data_pack}, $self->{index_size}) . chr(0) x $self->{index_size} ); - $content_length = $self->{index_size}; - } - elsif ($r eq 'ARRAY') { - print( $fh SIG_ARRAY ); - print( $fh pack($self->{data_pack}, $self->{index_size}) . chr(0) x $self->{index_size} ); - $content_length = $self->{index_size}; + $obj->{base_offset} = $header->size; + my $sector = DBM::Deep::Engine::Sector::Reference->new({ + engine => $self, + offset => $obj->_base_offset, + }); + unless ( $sector ) { + DBM::Deep->_throw_error("Corrupted file, no master index record"); } - elsif (!defined($value)) { - print( $fh SIG_NULL ); - print( $fh pack($self->{data_pack}, 0) ); - $content_length = 0; - } - else { - print( $fh SIG_DATA ); - print( $fh pack($self->{data_pack}, length($value)) . $value ); - $content_length = length($value); - } - } - ## - # Plain key is stored AFTER value, as keys are typically fetched less often. - ## - print( $fh pack($self->{data_pack}, length($key)) . $key ); - - ## - # If value is blessed, preserve class name - ## - if ( $root->{autobless} ) { - my $value_class = Scalar::Util::blessed($value); - if ( defined $value_class && !$value->isa( 'DBM::Deep' ) ) { - ## - # Blessed ref -- will restore later - ## - print( $fh chr(1) ); - print( $fh pack($self->{data_pack}, length($value_class)) . $value_class ); - $content_length += 1; - $content_length += $self->{data_size} + length($value_class); + unless ($obj->_type eq $sector->type) { + DBM::Deep->_throw_error("File type mismatch"); } - else { - print( $fh chr(0) ); - $content_length += 1; - } - } - ## - # If this is a new content area, advance EOF counter - ## - if ($location == $root->{end}) { - $root->{end} += SIG_SIZE; - $root->{end} += $self->{data_size} + $content_length; - $root->{end} += $self->{data_size} + length($key); - } - - ## - # If content is a hash or array, create new child DBM::Deep object and - # pass each key or element to it. - ## - if ( ! $internal_ref ) { - if ($r eq 'HASH') { - my $branch = DBM::Deep->new( - type => DBM::Deep->TYPE_HASH, - base_offset => $location, - root => $root, - ); - foreach my $key (keys %{$value}) { - $branch->STORE( $key, $value->{$key} ); - } - } - elsif ($r eq 'ARRAY') { - my $branch = DBM::Deep->new( - type => DBM::Deep->TYPE_ARRAY, - base_offset => $location, - root => $root, - ); - my $index = 0; - foreach my $element (@{$value}) { - $branch->STORE( $index, $element ); - $index++; - } - } + $obj->{staleness} = $sector->staleness; } + $self->storage->set_inode; + return 1; } -sub split_index { +sub begin_work { my $self = shift; - my ($obj, $md5, $tag) = @_; - - my $fh = $obj->_fh; - my $root = $obj->_root; - my $keys = $tag->{content}; - - seek($fh, $tag->{ref_loc} + $root->{file_offset}, SEEK_SET); - - my $loc = $self->_request_space( - $obj, $self->tag_size( $self->{index_size} ), - ); - - print( $fh pack($self->{long_pack}, $loc) ); - - my $index_tag = $self->create_tag( - $obj, $loc, SIG_INDEX, - chr(0) x $self->{index_size}, - ); - - my @offsets = (); - - $keys .= $md5 . pack($self->{long_pack}, 0); + my ($obj) = @_; - BUCKET: - for (my $i = 0; $i <= $self->{max_buckets}; $i++) { - my ($key, $old_subloc) = $self->_get_key_subloc( $keys, $i ); + if ( $self->trans_id ) { + DBM::Deep->_throw_error( "Cannot begin_work within an active transaction" ); + } - next BUCKET unless $key; + my @slots = $self->read_txn_slots; + my $found; + for my $i ( 0 .. $#slots ) { + next if $slots[$i]; - my $num = ord(substr($key, $tag->{ch} + 1, 1)); + $slots[$i] = 1; + $self->set_trans_id( $i + 1 ); + $found = 1; + last; + } + unless ( $found ) { + DBM::Deep->_throw_error( "Cannot allocate transaction ID" ); + } + $self->write_txn_slots( @slots ); - if ($offsets[$num]) { - my $offset = $offsets[$num] + SIG_SIZE + $self->{data_size}; - seek($fh, $offset + $root->{file_offset}, SEEK_SET); - my $subkeys; - read( $fh, $subkeys, $self->{bucket_list_size}); + if ( !$self->trans_id ) { + DBM::Deep->_throw_error( "Cannot begin_work - no available transactions" ); + } - for (my $k=0; $k<$self->{max_buckets}; $k++) { - my ($temp, $subloc) = $self->_get_key_subloc( $subkeys, $k ); + return; +} - if (!$subloc) { - seek($fh, $offset + ($k * $self->{bucket_size}) + $root->{file_offset}, SEEK_SET); - print( $fh $key . pack($self->{long_pack}, $old_subloc || $root->{end}) ); - last; - } - } # k loop - } - else { - $offsets[$num] = $root->{end}; - seek($fh, $index_tag->{offset} + ($num * $self->{long_size}) + $root->{file_offset}, SEEK_SET); +sub rollback { + my $self = shift; + my ($obj) = @_; - my $loc = $self->_request_space( - $obj, $self->tag_size( $self->{bucket_list_size} ), - ); + if ( !$self->trans_id ) { + DBM::Deep->_throw_error( "Cannot rollback without an active transaction" ); + } - print( $fh pack($self->{long_pack}, $loc) ); + foreach my $entry ( @{ $self->get_entries } ) { + my ($sector, $idx) = split ':', $entry; + $self->_load_sector( $sector )->rollback( $idx ); + } - my $blist_tag = $self->create_tag( - $obj, $loc, SIG_BLIST, - chr(0) x $self->{bucket_list_size}, - ); + $self->clear_entries; - seek($fh, $blist_tag->{offset} + $root->{file_offset}, SEEK_SET); - print( $fh $key . pack($self->{long_pack}, $old_subloc || $root->{end}) ); - } - } # i loop + my @slots = $self->read_txn_slots; + $slots[$self->trans_id-1] = 0; + $self->write_txn_slots( @slots ); + $self->inc_txn_staleness_counter( $self->trans_id ); + $self->set_trans_id( 0 ); - return; + return 1; } -sub read_from_loc { +sub commit { my $self = shift; - my ($obj, $subloc) = @_; + my ($obj) = @_; - my $fh = $obj->_fh; + if ( !$self->trans_id ) { + DBM::Deep->_throw_error( "Cannot commit without an active transaction" ); + } - ## - # Found match -- seek to offset and read signature - ## - my $signature; - seek($fh, $subloc + $obj->_root->{file_offset}, SEEK_SET); - read( $fh, $signature, SIG_SIZE); + foreach my $entry ( @{ $self->get_entries } ) { + my ($sector, $idx) = split ':', $entry; + $self->_load_sector( $sector )->commit( $idx ); + } - ## - # If value is a hash or array, return new DBM::Deep object with correct offset - ## - if (($signature eq SIG_HASH) || ($signature eq SIG_ARRAY)) { - my $obj = DBM::Deep->new( - type => $signature, - base_offset => $subloc, - root => $obj->_root, - ); + $self->clear_entries; - if ($obj->_root->{autobless}) { - ## - # Skip over value and plain key to see if object needs - # to be re-blessed - ## - seek($fh, $self->{data_size} + $self->{index_size}, SEEK_CUR); - - my $size; - read( $fh, $size, $self->{data_size}); $size = unpack($self->{data_pack}, $size); - if ($size) { seek($fh, $size, SEEK_CUR); } - - my $bless_bit; - read( $fh, $bless_bit, 1); - if (ord($bless_bit)) { - ## - # Yes, object needs to be re-blessed - ## - my $class_name; - read( $fh, $size, $self->{data_size}); $size = unpack($self->{data_pack}, $size); - if ($size) { read( $fh, $class_name, $size); } - if ($class_name) { $obj = bless( $obj, $class_name ); } - } - } + my @slots = $self->read_txn_slots; + $slots[$self->trans_id-1] = 0; + $self->write_txn_slots( @slots ); + $self->inc_txn_staleness_counter( $self->trans_id ); + $self->set_trans_id( 0 ); - return $obj; - } - elsif ( $signature eq SIG_INTERNAL ) { - my $size; - read( $fh, $size, $self->{data_size}); - $size = unpack($self->{data_pack}, $size); + return 1; +} - if ( $size ) { - my $new_loc; - read( $fh, $new_loc, $size ); - $new_loc = unpack( $self->{long_pack}, $new_loc ); +sub read_txn_slots { + my $self = shift; + return $self->_load_header->read_txn_slots(@_); +} - return $self->read_from_loc( $obj, $new_loc ); - } - else { - return; - } - } - ## - # Otherwise return actual value - ## - elsif ($signature eq SIG_DATA) { - my $size; - read( $fh, $size, $self->{data_size}); - $size = unpack($self->{data_pack}, $size); +sub write_txn_slots { + my $self = shift; + return $self->_load_header->write_txn_slots(@_); +} - my $value = ''; - if ($size) { read( $fh, $value, $size); } - return $value; - } +sub get_running_txn_ids { + my $self = shift; + my @transactions = $self->read_txn_slots; + my @trans_ids = map { $_+1} grep { $transactions[$_] } 0 .. $#transactions; +} - ## - # Key exists, but content is null - ## - return; +sub get_txn_staleness_counter { + my $self = shift; + return $self->_load_header->get_txn_staleness_counter(@_); } -sub get_bucket_value { - ## - # Fetch single value given tag and MD5 digested key. - ## +sub inc_txn_staleness_counter { my $self = shift; - my ($obj, $tag, $md5) = @_; + return $self->_load_header->inc_txn_staleness_counter(@_); +} - my ($subloc, $offset) = $self->_find_in_buckets( $tag, $md5 ); - if ( $subloc ) { - return $self->read_from_loc( $obj, $subloc ); - } - return; +sub get_entries { + my $self = shift; + return [ keys %{ $self->{entries}{$self->trans_id} ||= {} } ]; } -sub delete_bucket { - ## - # Delete single key/value pair given tag and MD5 digested key. - ## +sub add_entry { my $self = shift; - my ($obj, $tag, $md5) = @_; + my ($trans_id, $loc, $idx) = @_; - my ($subloc, $offset) = $self->_find_in_buckets( $tag, $md5 ); - if ( $subloc ) { - my $fh = $obj->_fh; - seek($fh, $tag->{offset} + $offset + $obj->_root->{file_offset}, SEEK_SET); - print( $fh substr($tag->{content}, $offset + $self->{bucket_size} ) ); - print( $fh chr(0) x $self->{bucket_size} ); + return unless $trans_id; - return 1; - } - return; + $self->{entries}{$trans_id} ||= {}; + $self->{entries}{$trans_id}{"$loc:$idx"} = undef; } -sub bucket_exists { - ## - # Check existence of single key given tag and MD5 digested key. - ## +# If the buckets are being relocated because of a reindexing, the entries +# mechanism needs to be made aware of it. +sub reindex_entry { my $self = shift; - my ($obj, $tag, $md5) = @_; + my ($old_loc, $old_idx, $new_loc, $new_idx) = @_; + + TRANS: + while ( my ($trans_id, $locs) = each %{ $self->{entries} } ) { + if ( exists $locs->{"$old_loc:$old_idx"} ) { + delete $locs->{"$old_loc:$old_idx"}; + $locs->{"$new_loc:$new_idx"} = undef; + next TRANS; + } + } +} - my ($subloc, $offset) = $self->_find_in_buckets( $tag, $md5 ); - return $subloc && 1; +sub clear_entries { + my $self = shift; + delete $self->{entries}{$self->trans_id}; } -sub find_bucket_list { - ## - # Locate offset for bucket list, given digested key - ## +################################################################################ + +sub _apply_digest { my $self = shift; - my ($obj, $md5, $args) = @_; - $args = {} unless $args; + return $self->{digest}->(@_); +} - ## - # Locate offset for bucket list using digest index system - ## - my $tag = $self->load_tag($obj, $obj->_base_offset) - or $obj->_throw_error( "INTERNAL ERROR - Cannot find tag" ); +sub _add_free_blist_sector { shift->_add_free_sector( 0, @_ ) } +sub _add_free_data_sector { shift->_add_free_sector( 1, @_ ) } +sub _add_free_index_sector { shift->_add_free_sector( 2, @_ ) } +sub _add_free_sector { shift->_load_header->add_free_sector( @_ ) } + +sub _request_blist_sector { shift->_request_sector( 0, @_ ) } +sub _request_data_sector { shift->_request_sector( 1, @_ ) } +sub _request_index_sector { shift->_request_sector( 2, @_ ) } +sub _request_sector { shift->_load_header->request_sector( @_ ) } + +################################################################################ + +{ + my %t = ( + SIG_ARRAY => 'Reference', + SIG_HASH => 'Reference', + SIG_BLIST => 'BucketList', + SIG_INDEX => 'Index', + SIG_NULL => 'Null', + SIG_DATA => 'Scalar', + ); - my $ch = 0; - while ($tag->{signature} ne SIG_BLIST) { - my $num = ord substr($md5, $ch, 1); + my %class_for; + while ( my ($k,$v) = each %t ) { + $class_for{ DBM::Deep::Engine->$k } = "DBM::Deep::Engine::Sector::$v"; + } - my $ref_loc = $tag->{offset} + ($num * $self->{long_size}); - $tag = $self->index_lookup( $obj, $tag, $num ); + sub load_sector { + my $self = shift; + my ($offset) = @_; + + my $data = $self->get_data( $offset ) + or return;#die "Cannot read from '$offset'\n"; + my $type = substr( $$data, 0, 1 ); + my $class = $class_for{ $type }; + return $class->new({ + engine => $self, + type => $type, + offset => $offset, + }); + } + *_load_sector = \&load_sector; - if (!$tag) { - return if !$args->{create}; + sub load_header { + my $self = shift; - my $fh = $obj->_fh; - seek($fh, $ref_loc + $obj->_root->{file_offset}, SEEK_SET); + #XXX Does this mean we make too many objects? -RobK, 2008-06-23 + return DBM::Deep::Engine::Sector::FileHeader->new({ + engine => $self, + offset => 0, + }); + } + *_load_header = \&load_header; - my $loc = $self->_request_space( - $obj, $self->tag_size( $self->{bucket_list_size} ), - ); + sub get_data { + my $self = shift; + my ($offset, $size) = @_; + return unless defined $offset; - print( $fh pack($self->{long_pack}, $loc) ); + unless ( exists $self->sector_cache->{$offset} ) { + # Don't worry about the header sector. It will manage itself. + return unless $offset; - $tag = $self->create_tag( - $obj, $loc, SIG_BLIST, - chr(0) x $self->{bucket_list_size}, - ); + if ( !defined $size ) { + my $type = $self->storage->read_at( $offset, 1 ) + or die "($offset): Cannot read from '$offset' to find the type\n"; - $tag->{ref_loc} = $ref_loc; - $tag->{ch} = $ch; + if ( $type eq $self->SIG_FREE ) { + return; + } - last; + my $class = $class_for{$type} + or die "($offset): Cannot find class for '$type'\n"; + $size = $class->size( $self ) + or die "($offset): '$class' doesn't return a size\n"; + $self->sector_cache->{$offset} = $type . $self->storage->read_at( undef, $size - 1 ); + } + else { + $self->sector_cache->{$offset} = $self->storage->read_at( $offset, $size ) + or return; + } } - $tag->{ch} = $ch++; - $tag->{ref_loc} = $ref_loc; + return \$self->sector_cache->{$offset}; } - - return $tag; } -sub index_lookup { - ## - # Given index tag, lookup single entry in index and return . - ## +sub sector_cache { my $self = shift; - my ($obj, $tag, $index) = @_; - - my $location = unpack( - $self->{long_pack}, - substr( - $tag->{content}, - $index * $self->{long_size}, - $self->{long_size}, - ), - ); + return $self->{sector_cache} ||= {}; +} - if (!$location) { return; } +sub clear_sector_cache { + my $self = shift; + $self->{sector_cache} = {}; +} - return $self->load_tag( $obj, $location ); +sub dirty_sectors { + my $self = shift; + return $self->{dirty_sectors} ||= {}; } -sub traverse_index { - ## - # Scan index and recursively step into deeper levels, looking for next key. - ## +sub clear_dirty_sectors { my $self = shift; - my ($obj, $offset, $ch, $force_return_next) = @_; + $self->{dirty_sectors} = {}; +} - my $tag = $self->load_tag($obj, $offset ); +sub add_dirty_sector { + my $self = shift; + my ($offset) = @_; - my $fh = $obj->_fh; + $self->dirty_sectors->{ $offset } = undef; +} - if ($tag->{signature} ne SIG_BLIST) { - my $content = $tag->{content}; - my $start = $obj->{return_next} ? 0 : ord(substr($obj->{prev_md5}, $ch, 1)); +sub flush { + my $self = shift; - for (my $idx = $start; $idx < (2**8); $idx++) { - my $subloc = unpack( - $self->{long_pack}, - substr( - $content, - $idx * $self->{long_size}, - $self->{long_size}, - ), - ); + my $sectors = $self->dirty_sectors; + for my $offset (sort { $a <=> $b } keys %{ $sectors }) { + $self->storage->print_at( $offset, $self->sector_cache->{$offset} ); + } - if ($subloc) { - my $result = $self->traverse_index( - $obj, $subloc, $ch + 1, $force_return_next, - ); + # Why do we need to have the storage flush? Shouldn't autoflush take care of things? + # -RobK, 2008-06-26 + $self->storage->flush; - if (defined($result)) { return $result; } - } - } # index loop + $self->clear_dirty_sectors; - $obj->{return_next} = 1; - } # tag is an index + $self->clear_sector_cache; +} - else { - my $keys = $tag->{content}; - if ($force_return_next) { $obj->{return_next} = 1; } - - ## - # Iterate through buckets, looking for a key match - ## - for (my $i = 0; $i < $self->{max_buckets}; $i++) { - my ($key, $subloc) = $self->_get_key_subloc( $keys, $i ); - - # End of bucket list -- return to outer loop - if (!$subloc) { - $obj->{return_next} = 1; - last; - } - # Located previous key -- return next one found - elsif ($key eq $obj->{prev_md5}) { - $obj->{return_next} = 1; - next; - } - # Seek to bucket location and skip over signature - elsif ($obj->{return_next}) { - seek($fh, $subloc + $obj->_root->{file_offset}, SEEK_SET); - - # Skip over value to get to plain key - my $sig; - read( $fh, $sig, SIG_SIZE ); - - my $size; - read( $fh, $size, $self->{data_size}); - $size = unpack($self->{data_pack}, $size); - if ($size) { seek($fh, $size, SEEK_CUR); } - - # Read in plain key and return as scalar - my $plain_key; - read( $fh, $size, $self->{data_size}); - $size = unpack($self->{data_pack}, $size); - if ($size) { read( $fh, $plain_key, $size); } - - return $plain_key; - } - } +################################################################################ - $obj->{return_next} = 1; - } # tag is a bucket list +sub lock_exclusive { + my $self = shift; + my ($obj) = @_; + return $self->storage->lock_exclusive( $obj ); +} - return; +sub lock_shared { + my $self = shift; + my ($obj) = @_; + return $self->storage->lock_shared( $obj ); } -sub get_next_key { - ## - # Locate next key, given digested previous one - ## +sub unlock { my $self = shift; my ($obj) = @_; - $obj->{prev_md5} = $_[1] ? $_[1] : undef; - $obj->{return_next} = 0; + my $rv = $self->storage->unlock( $obj ); - ## - # If the previous key was not specifed, start at the top and - # return the first one found. - ## - if (!$obj->{prev_md5}) { - $obj->{prev_md5} = chr(0) x $self->{hash_size}; - $obj->{return_next} = 1; - } + $self->flush if $rv; - return $self->traverse_index( $obj, $obj->_base_offset, 0 ); + return $rv; } -# Utilities - -sub _get_key_subloc { - my $self = shift; - my ($keys, $idx) = @_; - - my ($key, $subloc) = unpack( - "a$self->{hash_size} $self->{long_pack}", - substr( - $keys, - ($idx * $self->{bucket_size}), - $self->{bucket_size}, - ), - ); +################################################################################ - return ($key, $subloc); -} +sub storage { $_[0]{storage} } +sub byte_size { $_[0]{byte_size} } +sub hash_size { $_[0]{hash_size} } +sub hash_chars { $_[0]{hash_chars} } +sub num_txns { $_[0]{num_txns} } +sub max_buckets { $_[0]{max_buckets} } +sub blank_md5 { chr(0) x $_[0]->hash_size } +sub data_sector_size { $_[0]{data_sector_size} } -sub _find_in_buckets { +# This is a calculated value +sub txn_bitfield_len { my $self = shift; - my ($tag, $md5) = @_; - - BUCKET: - for ( my $i = 0; $i < $self->{max_buckets}; $i++ ) { - my ($key, $subloc) = $self->_get_key_subloc( $tag->{content}, $i ); + unless ( exists $self->{txn_bitfield_len} ) { + my $temp = ($self->num_txns) / 8; + if ( $temp > int( $temp ) ) { + $temp = int( $temp ) + 1; + } + $self->{txn_bitfield_len} = $temp; + } + return $self->{txn_bitfield_len}; +} - return ($subloc, $i * $self->{bucket_size}) unless $subloc; +sub trans_id { $_[0]{trans_id} } +sub set_trans_id { $_[0]{trans_id} = $_[1] } - next BUCKET if $key ne $md5; +sub trans_loc { $_[0]{trans_loc} } +sub set_trans_loc { $_[0]{trans_loc} = $_[1] } - return ($subloc, $i * $self->{bucket_size}); - } +sub chains_loc { $_[0]{chains_loc} } +sub set_chains_loc { $_[0]{chains_loc} = $_[1] } - return; -} +sub cache { $_[0]{cache} ||= {} } +sub clear_cache { %{$_[0]->cache} = () } -sub _request_space { +sub _dump_file { my $self = shift; - my ($obj, $size) = @_; + $self->flush; - my $loc = $obj->_root->{end}; - $obj->_root->{end} += $size; + # Read the header + my $header_sector = DBM::Deep::Engine::Sector::FileHeader->new({ + engine => $self, + }); - return $loc; -} + my %types = ( + 0 => 'B', + 1 => 'D', + 2 => 'I', + ); -sub _release_space { - my $self = shift; - my ($obj, $size, $loc) = @_; + my %sizes = ( + 'D' => $self->data_sector_size, + 'B' => DBM::Deep::Engine::Sector::BucketList->new({engine=>$self,offset=>1})->size, + 'I' => DBM::Deep::Engine::Sector::Index->new({engine=>$self,offset=>1})->size, + ); - return; -} + my $return = ""; -1; -__END__ + # Filesize + $return .= "Size: " . (-s $self->storage->{fh}) . $/; -# This will be added in later, after more refactoring is done. This is an early -# attempt at refactoring on the physical level instead of the virtual level. -sub _read_at { - my $self = shift; - my ($obj, $spot, $amount, $unpack) = @_; + # Header values + $return .= "NumTxns: " . $self->num_txns . $/; - my $fh = $obj->_fh; - seek( $fh, $spot + $obj->_root->{file_offset}, SEEK_SET ); + # Read the free sector chains + my %sectors; + foreach my $multiple ( 0 .. 2 ) { + $return .= "Chains($types{$multiple}):"; + my $old_loc = $self->chains_loc + $multiple * $self->byte_size; + while ( 1 ) { + my $loc = unpack( + $StP{$self->byte_size}, + $self->storage->read_at( $old_loc, $self->byte_size ), + ); - my $buffer; - my $bytes_read = read( $fh, $buffer, $amount ); + # We're now out of free sectors of this kind. + unless ( $loc ) { + last; + } - if ( $unpack ) { - $buffer = unpack( $unpack, $buffer ); + $sectors{ $types{$multiple} }{ $loc } = undef; + $old_loc = $loc + SIG_SIZE + $STALE_SIZE; + $return .= " $loc"; + } + $return .= $/; } - if ( wantarray ) { - return ($buffer, $bytes_read); - } - else { - return $buffer; + my $spot = $header_sector->size; + SECTOR: + while ( $spot < $self->storage->{end} ) { + # Read each sector in order. + my $sector = $self->_load_sector( $spot ); + if ( !$sector ) { + # Find it in the free-sectors that were found already + foreach my $type ( keys %sectors ) { + if ( exists $sectors{$type}{$spot} ) { + my $size = $sizes{$type}; + $return .= sprintf "%08d: %s %04d\n", $spot, 'F' . $type, $size; + $spot += $size; + next SECTOR; + } + } + + die "********\n$return\nDidn't find free sector for $spot in chains\n********\n"; + } + else { + $return .= sprintf "%08d: %s %04d", $spot, $sector->type, $sector->size; + if ( $sector->type eq 'D' ) { + $return .= ' ' . $sector->data; + } + elsif ( $sector->type eq 'A' || $sector->type eq 'H' ) { + $return .= ' REF: ' . $sector->get_refcount; + } + elsif ( $sector->type eq 'B' ) { + foreach my $bucket ( $sector->chopped_up ) { + $return .= "\n "; + $return .= sprintf "%08d", unpack($StP{$self->byte_size}, + substr( $bucket->[-1], $self->hash_size, $self->byte_size), + ); + my $l = unpack( $StP{$self->byte_size}, + substr( $bucket->[-1], + $self->hash_size + $self->byte_size, + $self->byte_size, + ), + ); + $return .= sprintf " %08d", $l; + foreach my $txn ( 0 .. $self->num_txns - 2 ) { + my $l = unpack( $StP{$self->byte_size}, + substr( $bucket->[-1], + $self->hash_size + 2 * $self->byte_size + $txn * ($self->byte_size + $STALE_SIZE), + $self->byte_size, + ), + ); + $return .= sprintf " %08d", $l; + } + } + } + $return .= $/; + + $spot += $sector->size; + } } + + return $return; } + +1; +__END__