NAME DBM::Deep - A pure perl multi-level hash/array DBM SYNOPSIS use DBM::Deep; my $db = new DBM::Deep "foo.db"; $db->{key} = 'value'; # tie() style print $db->{key}; $db->put('key', 'value'); # OO style print $db->get('key'); # true multi-level support $db->{my_complex} = [ 'hello', { perl => 'rules' }, 42, 99 ]; DESCRIPTION A unique flat-file database module, written in pure perl. True multi-level hash/array support (unlike MLDBM, which is faked), hybrid OO / tie() interface, cross-platform FTPable files, and quite fast. Can handle millions of keys and unlimited hash levels without significant slow-down. Written from the ground-up in pure perl -- this is NOT a wrapper around a C-based DBM. Out-of-the-box compatibility with Unix, Mac OS X and Windows. INSTALLATION Hopefully you are using CPAN's excellent Perl module, which will download and install the module for you. If not, get the tarball, and run these commands: tar zxf DBM-Deep-* cd DBM-Deep-* perl Makefile.PL make make test make install SETUP Construction can be done OO-style (which is the recommended way), or using Perl's tie() function. Both are examined here. OO CONSTRUCTION The recommended way to construct a DBM::Deep object is to use the new() method, which gets you a blessed, tied hash or array reference. my $db = new DBM::Deep "foo.db"; This opens a new database handle, mapped to the file "foo.db". If this file does not exist, it will automatically be created. DB files are opened in "r+" (read/write) mode, and the type of object returned is a hash, unless otherwise specified (see OPTIONS below). You can pass a number of options to the constructor to specify things like locking, autoflush, etc. This is done by passing an inline hash: my $db = new DBM::Deep( file => "foo.db", locking => 1, autoflush => 1 ); Notice that the filename is now specified *inside* the hash with the "file" parameter, as opposed to being the sole argument to the constructor. This is required if any options are specified. See OPTIONS below for the complete list. You can also start with an array instead of a hash. For this, you must specify the "type" parameter: my $db = new DBM::Deep( file => "foo.db", type => DBM::Deep::TYPE_ARRAY ); Note: Specifing the "type" parameter only takes effect when beginning a new DB file. If you create a DBM::Deep object with an existing file, the "type" will be loaded from the file header, and ignored if it is passed to the constructor. TIE CONSTRUCTION Alternatively, you can create a DBM::Deep handle by using Perl's built-in tie() function. This is not ideal, because you get only a basic, tied hash (or array) which is not blessed, so you can't call any functions on it. my %hash; tie %hash, "DBM::Deep", "foo.db"; my @array; tie @array, "DBM::Deep", "bar.db"; As with the OO constructor, you can replace the DB filename parameter with a hash containing one or more options (see OPTIONS just below for the complete list). tie %hash, "DBM::Deep", { file => "foo.db", locking => 1, autoflush => 1 }; OPTIONS There are a number of options that can be passed in when constructing your DBM::Deep objects. These apply to both the OO- and tie- based approaches. * file Filename of the DB file to link the handle to. You can pass a full absolute filesystem path, partial path, or a plain filename if the file is in the current working directory. This is a required parameter. * mode File open mode (read-only, read-write, etc.) string passed to Perl's FileHandle module. This is an optional parameter, and defaults to "r+" (read/write). Note: If the default (r+) mode is selected, the file will also be auto- created if it doesn't exist. * type This parameter specifies what type of object to create, a hash or array. Use one of these two constants: "DBM::Deep::TYPE_HASH" or "DBM::Deep::TYPE_ARRAY". This only takes effect when beginning a new file. This is an optional parameter, and defaults to hash. * locking Specifies whether locking is to be enabled. DBM::Deep uses Perl's Fnctl flock() function to lock the database in exclusive mode for writes, and shared mode for reads. Pass any true value to enable. This affects the base DB handle *and any child hashes or arrays* that use the same DB file. This is an optional parameter, and defaults to 0 (disabled). See LOCKING below for more. * autoflush Specifies whether autoflush is to be enabled on the underlying FileHandle. This obviously slows down write operations, but is required if you may have multiple processes accessing the same DB file (also consider enable *locking* or at least *volatile*). Pass any true value to enable. This is an optional parameter, and defaults to 0 (disabled). * volatile If *volatile* mode is enabled, DBM::Deep will stat() the DB file before each STORE() operation. This is required if an outside force may change the size of the file between transactions. Locking also implicitly enables volatile. This is useful if you want to use a different locking system or write your own. Pass any true value to enable. This is an optional parameter, and defaults to 0 (disabled). * filter_* See FILTERS below. * debug Currently, *debug* mode does nothing more than print all errors to STDERR. However, it may be expanded in the future to log more debugging information. Pass any true value to enable. This is an optional paramter, and defaults to 0 (disabled). TIE INTERFACE With DBM::Deep you can access your databases using Perl's standard hash/array syntax. Because all Deep objects are *tied* to hashes or arrays, you can treat them as such. Deep will intercept all reads/writes and direct them to the right place -- the DB file. This has nothing to do with the "TIE CONSTRUCTION" section above. This simply tells you how to use DBM::Deep using regular hashes and arrays, rather than calling functions like "get()" and "put()" (although those work too). It is entirely up to you how to want to access your databases. HASHES You can treat any DBM::Deep object like a normal Perl hash reference. Add keys, or even nested hashes (or arrays) using standard Perl syntax: my $db = new DBM::Deep "foo.db"; $db->{mykey} = "myvalue"; $db->{myhash} = {}; $db->{myhash}->{subkey} = "subvalue"; print $db->{myhash}->{subkey} . "\n"; You can even step through hash keys using the normal Perl "keys()" function: foreach my $key (keys %$db) { print "$key: " . $db->{$key} . "\n"; } Remember that Perl's "keys()" function extracts *every* key from the hash and pushes them onto an array, all before the loop even begins. If you have an extra large hash, this may exhaust Perl's memory. Instead, consider using Perl's "each()" function, which pulls keys/values one at a time, using very little memory: while (my ($key, $value) = each %$db) { print "$key: $value\n"; } ARRAYS As with hashes, you can treat any DBM::Deep object like a normal Perl array reference. This includes inserting, removing and manipulating elements, and the "push()", "pop()", "shift()", "unshift()" and "splice()" functions. The object must have first been created using type "DBM::Deep::TYPE_ARRAY", or simply be a nested array reference inside a hash. Example: my $db = new DBM::Deep( file => "foo-array.db", type => DBM::Deep::TYPE_ARRAY ); $db->[0] = "foo"; push @$db, "bar", "baz"; unshift @$db, "bah"; my $last_elem = pop @$db; # baz my $first_elem = shift @$db; # bah my $second_elem = $db->[1]; # bar my $num_elements = scalar @$db; OO INTERFACE In addition to the *tie()* interface, you can also use a standard OO interface to manipulate all aspects of DBM::Deep databases. Each type of object (hash or array) has its own methods, but both types share the following common methods: "put()", "get()", "exists()", "delete()" and "clear()". * put() Stores a new hash key/value pair, or sets an array element value. Takes two arguments, the hash key or array index, and the new value. The value can be a scalar, hash ref or array ref. Returns true on success, false on failure. $db->put("foo", "bar"); # for hashes $db->put(1, "bar"); # for arrays * get() Fetches the value of a hash key or array element. Takes one argument: the hash key or array index. Returns a scalar, hash ref or array ref, depending on the data type stored. my $value = $db->get("foo"); # for hashes my $value = $db->get(1); # for arrays * exists() Checks if a hash key or array index exists. Takes one argument: the hash key or array index. Returns true if it exists, false if not. if ($db->exists("foo")) { print "yay!\n"; } # for hashes if ($db->exists(1)) { print "yay!\n"; } # for arrays * delete() Deletes one hash key/value pair or array element. Takes one argument: the hash key or array index. Returns true on success, false if not found. For arrays, the remaining elements located after the deleted element are NOT moved over. The deleted element is essentially just undefined, which is exactly how Perl's internal arrays work. Please note that the space occupied by the deleted key/value or element is not reused again -- see "UNUSED SPACE RECOVERY" below for details and workarounds. $db->delete("foo"); # for hashes $db->delete(1); # for arrays * clear() Deletes all hash keys or array elements. Takes no arguments. No return value. Please note that the space occupied by the deleted keys/values or elements is not reused again -- see "UNUSED SPACE RECOVERY" below for details and workarounds. $db->clear(); # hashes or arrays HASHES For hashes, DBM::Deep supports all the common methods described above, and the following additional methods: "first_key()" and "next_key()". * first_key() Returns the "first" key in the hash. As with built-in Perl hashes, keys are fetched in an undefined order (which appears random). Takes no arguments, returns the key as a scalar value. my $key = $db->first_key(); * next_key() Returns the "next" key in the hash, given the previous one as the sole argument. Returns undef if there are no more keys to be fetched. $key = $db->next_key($key); Here are some examples of using hashes: my $db = new DBM::Deep "foo.db"; $db->put("foo", "bar"); print "foo: " . $db->get("foo") . "\n"; $db->put("baz", {}); # new child hash ref $db->get("baz")->put("buz", "biz"); print "buz: " . $db->get("baz")->get("buz") . "\n"; my $key = $db->first_key(); while ($key) { print "$key: " . $db->get($key) . "\n"; $key = $db->next_key($key); } if ($db->exists("foo")) { $db->delete("foo"); } ARRAYS For arrays, DBM::Deep supports all the common methods described above, and the following additional methods: "length()", "push()", "pop()", "shift()", "unshift()" and "splice()". * length() Returns the number of elements in the array. Takes no arguments. my $len = $db->length(); * push() Adds one or more elements onto the end of the array. Accepts scalars, hash refs or array refs. No return value. $db->push("foo", "bar", {}); * pop() Fetches the last element in the array, and deletes it. Takes no arguments. Returns undef if array is empty. Returns the element value. my $elem = $db->pop(); * shift() Fetches the first element in the array, deletes it, then shifts all the remaining elements over to take up the space. Returns the element value. This method is not recommended with large arrays -- see "LARGE ARRAYS" below for details. my $elem = $db->shift(); * unshift() Inserts one or more elements onto the beginning of the array, shifting all existing elements over to make room. Accepts scalars, hash refs or array refs. No return value. This method is not recommended with large arrays -- see below for details. $db->unshift("foo", "bar", {}); * splice() Performs exactly like Perl's built-in function of the same name. See "perldoc -f splice" for usage -- it is too complicated to document here. This method is not recommended with large arrays -- see "LARGE ARRAYS" below for details. Here are some examples of using arrays: my $db = new DBM::Deep( file => "foo.db", type => DBM::Deep::TYPE_ARRAY ); $db->push("bar", "baz"); $db->unshift("foo"); $db->put(3, "buz"); my $len = $db->length(); print "length: $len\n"; # 4 for (my $k=0; $k<$len; $k++) { print "$k: " . $db->get($k) . "\n"; } $db->splice(1, 2, "biz", "baf"); while (my $elem = shift @$db) { print "shifted: $elem\n"; } LOCKING Enable automatic file locking by passing a true value to the "locking" parameter when constructing your DBM::Deep object (see SETUP above). my $db = new DBM::Deep( file => "foo.db", locking => 1 ); This causes Deep to "flock()" the underlying FileHandle object with exclusive mode for writes, and shared mode for reads. This is required if you have multiple processes accessing the same database file, to avoid file corruption. Please note that "flock()" does NOT work for files over NFS. See "DB OVER NFS" below for more. EXPLICIT LOCKING You can explicitly lock a database, so it remains locked for multiple transactions. This is done by calling the "lock()" method, and passing an optional lock mode argument (defaults to exclusive mode). This is particularly useful for things like counters, where the current value needs to be fetched, then incremented, then stored again. $db->lock(); my $counter = $db->get("counter"); $counter++; $db->put("counter", $counter); $db->unlock(); # or... $db->lock(); $db->{counter}++; $db->unlock(); You can pass "lock()" an optional argument, which specifies which mode to use (exclusive or shared). Use one of these two constants: "DBM::Deep::LOCK_EX" or "DBM::Deep::LOCK_SH". These are passed directly to "flock()", and are the same as the constants defined in Perl's "Fcntl" module. $db->lock( DBM::Deep::LOCK_SH ); # something here $db->unlock(); If you want to implement your own file locking scheme, be sure to create your DBM::Deep objects setting the "volatile" option to true. This hints to Deep that the DB file may change between transactions. See "LOW-LEVEL ACCESS" below for more. IMPORTING/EXPORTING You can import existing complex structures by calling the "import()" method, and export an entire database into an in-memory structure using the "export()" method. Both are examined here. IMPORTING Say you have an existing hash with nested hashes/arrays inside it. Instead of walking the structure and adding keys/elements to the database as you go, simply pass a reference to the "import()" method. This recursively adds everything to an existing DBM::Deep object for you. Here is an example: my $struct = { key1 => "value1", key2 => "value2", array1 => [ "elem0", "elem1", "elem2" ], hash1 => { subkey1 => "subvalue1", subkey2 => "subvalue2" } }; my $db = new DBM::Deep "foo.db"; $db->import( $struct ); print $db->{key1} . "\n"; # prints "value1" This recursively imports the entire $struct object into $db, including all nested hashes and arrays. If the DBM::Deep object contains exsiting data, keys are merged with the existing ones, replacing if they already exist. The "import()" method can be called on any database level (not just the base level), and works with both hash and array DB types. Note: Make sure your existing structure has no circular references in it. These will cause an infinite loop when importing. EXPORTING Calling the "export()" method on an existing DBM::Deep object will return a reference to a new in-memory copy of the datbase. The export is done recursively, so all nested hashes/arrays are all exported to standard Perl objects. Here is an example: my $db = new DBM::Deep "foo.db"; $db->{key1} = "value1"; $db->{key2} = "value2"; $db->{hash1} = {}; $db->{hash1}->{subkey1} = "subvalue1"; $db->{hash1}->{subkey2} = "subvalue2"; my $struct = $db->export(); print $struct->{key1} . "\n"; # prints "value1" This makes a complete copy of the database in memory, and returns a reference to it. The "export()" method can be called on any database level (not just the base level), and works with both hash and array DB types. Be careful of large databases -- you can store a lot more data in a DBM::Deep object than an in-memory Perl structure. Note: Make sure your database has no circular references in it. These will cause an infinite loop when exporting. FILTERS DBM::Deep has a number of hooks where you can specify your own Perl function to perform filtering on incoming or outgoing data. This is a perfect way to extend the engine, and implement things like real-time compression or encryption. Filtering applies to the base DB level, and all child hashes / arrays. Filter hooks can be specified when your DBM::Deep object is first constructed, or by calling the "set_filter()" method at any time. There are four available filter hooks, described below: * filter_store_key This filter is called whenever a hash key is stored. It is passed the incoming key, and expected to return a transformed key. * filter_store_value This filter is called whenever a hash key or array element is stored. It is passed the incoming value, and expected to return a transformed value. * filter_fetch_key This filter is called whenever a hash key is fetched (i.e. via "first_key()" or "next_key()"). It is passed the transformed key, and expected to return the plain key. * filter_fetch_value This filter is called whenever a hash key or array element is fetched. It is passed the transformed value, and expected to return the plain value. Here are the two ways to setup a filter hook: my $db = new DBM::Deep( file => "foo.db", filter_store_value => \&my_filter_store, filter_fetch_value => \&my_filter_fetch ); # or... $db->set_filter( "filter_store_value", \&my_filter_store ); $db->set_filter( "filter_fetch_value", \&my_filter_fetch ); Your filter function will be called only when dealing with SCALAR keys or values. When nested hashes and arrays are being stored/fetched, filtering is bypassed. Filters are called as static functions, passed a single SCALAR argument, and expected to return a single SCALAR value. If you want to remove a filter, set the function reference to "undef": $db->set_filter( "filter_store_value", undef ); REAL-TIME ENCRYPTION EXAMPLE Here is a working example that uses the *Crypt::Blowfish* module to do real-time encryption / decryption of keys & values with DBM::Deep Filters. Please visit for more on *Crypt::Blowfish*. You'll also need the *Crypt::CBC* module. use DBM::Deep; use Crypt::Blowfish; use Crypt::CBC; my $cipher = new Crypt::CBC({ 'key' => 'my secret key', 'cipher' => 'Blowfish', 'iv' => '$KJh#(}q', 'regenerate_key' => 0, 'padding' => 'space', 'prepend_iv' => 0 }); my $db = new DBM::Deep( file => "foo-encrypt.db", filter_store_key => \&my_encrypt, filter_store_value => \&my_encrypt, filter_fetch_key => \&my_decrypt, filter_fetch_value => \&my_decrypt, ); $db->{key1} = "value1"; $db->{key2} = "value2"; print "key1: " . $db->{key1} . "\n"; print "key2: " . $db->{key2} . "\n"; undef $db; exit; sub my_encrypt { return $cipher->encrypt( $_[0] ); } sub my_decrypt { return $cipher->decrypt( $_[0] ); } REAL-TIME COMPRESSION EXAMPLE Here is a working example that uses the *Compress::Zlib* module to do real-time compression / decompression of keys & values with DBM::Deep Filters. Please visit for more on *Compress::Zlib*. use DBM::Deep; use Compress::Zlib; my $db = new DBM::Deep( file => "foo-compress.db", filter_store_key => \&my_compress, filter_store_value => \&my_compress, filter_fetch_key => \&my_decompress, filter_fetch_value => \&my_decompress, ); $db->{key1} = "value1"; $db->{key2} = "value2"; print "key1: " . $db->{key1} . "\n"; print "key2: " . $db->{key2} . "\n"; undef $db; exit; sub my_compress { return Compress::Zlib::memGzip( $_[0] ) ; } sub my_decompress { return Compress::Zlib::memGunzip( $_[0] ) ; } Note: Filtering of keys only applies to hashes. Array "keys" are actually numerical index numbers, and are not filtered. ERROR HANDLING Most DBM::Deep methods return a true value for success, and a false value for failure. Upon failure, the actual error message is stored in an internal scalar, which can be fetched by calling the "error()" method. my $db = new DBM::Deep "foo.db"; # hash $db->push("foo"); # ILLEGAL -- array only func print $db->error(); # prints error message You can then call "clear_error()" to clear the current error state. $db->clear_error(); It is always a good idea to check the error state upon object creation. Deep immediately tries to "open()" the FileHandle, so if you don't have sufficient permissions or some other filesystem error occurs, you should act accordingly before trying to access the database. my $db = new DBM::Deep("foo.db"); if ($db->error()) { die "ERROR: " . $db->error(); } If you set the "debug" option to true when creating your DBM::Deep object, all errors are printed to STDERR. LARGEFILE SUPPORT If you have a 64-bit system, and your Perl is compiled with both LARGEFILE and 64-bit support, you *may* be able to create databases larger than 2 GB. DBM::Deep by default uses 32-bit file offset tags, but these can be changed by calling the static "set_pack()" method before you do anything else. DBM::Deep::set_pack(8, 'Q'); This tells DBM::Deep to pack all file offsets with 8-byte (64-bit) quad words instead of 32-bit longs. After setting these values your DB files have a theoretical maximum size of 16 XB (exabytes). Note: Changing these values will NOT work for existing database files. Only change this for new files, and make sure it stays set consistently throughout the file's life. If you do set these values, you can no longer access 32-bit DB files. You can, however, call "set_pack(4, 'N')" to change back to 32-bit mode. Note: I have not personally tested files > 2 GB -- all my systems have only a 32-bit Perl. If anyone tries this, please tell me what happens! LOW-LEVEL ACCESS If you require low-level access to the underlying FileHandle that Deep uses, you can call the "fh()" method, which returns the handle: my $fh = $db->fh(); This method can be called on the root level of the datbase, or any child hashes or arrays. All levels share a *root* structure, which contains things like the FileHandle, a reference counter, and all your options you specified when you created the object. You can get access to this root structure by calling the "root()" method. my $root = $db->root(); This is useful for changing options after the object has already been created, such as enabling/disabling locking, volatile or debug modes. You can also store your own temporary user data in this structure (be wary of name collision), which is then accessible from any child hash or array. CUSTOM DIGEST ALGORITHM DBM::Deep by default uses the *Message Digest 5* (MD5) algorithm for hashing keys. However you can override this, and use another algorithm (such as SHA-256) or even write your own. But please note that Deep currently expects zero collisions, so your algorithm has to be *perfect*, so to speak. Collision detection may be introduced in a later version. You can specify a custom digest algorithm by calling the static "set_digest()" function, passing a reference to a subroutine, and the length of the algorithm's hashes (in bytes). This is a global static function, which affects ALL Deep objects. Here is a working example that uses a 256-bit hash from the *Digest::SHA256* module. Please see for more. use DBM::Deep; use Digest::SHA256; my $context = Digest::SHA256::new(256); DBM::Deep::set_digest( \&my_digest, 32 ); my $db = new DBM::Deep "foo-sha.db"; $db->{key1} = "value1"; $db->{key2} = "value2"; print "key1: " . $db->{key1} . "\n"; print "key2: " . $db->{key2} . "\n"; undef $db; exit; sub my_digest { return substr( $context->hash($_[0]), 0, 32 ); } Note: Your returned digest strings must be EXACTLY the number of bytes you specify in the "set_digest()" function (in this case 32). CIRCULAR REFERENCES DBM::Deep has experimental support for circular references. Meaning you can have a nested hash key or array element that points to a parent object. This relationship is stored in the DB file, and is preserved between sessions. Here is an example: my $db = new DBM::Deep "foo.db"; $db->{foo} = "bar"; $db->{circle} = $db; # ref to self print $db->{foo} . "\n"; # prints "foo" print $db->{circle}->{foo} . "\n"; # prints "foo" again One catch is, passing the object to a function that recursively walks the object tree (such as *Data::Dumper* or even the built-in "optimize()" or "export()" methods) will result in an infinite loop. The other catch is, if you fetch the *key* of a circular reference (i.e. using the "first_key()" or "next_key()" methods), you will get the *target object's key*, not the ref's key. This gets even more interesting with the above example, where the *circle* key points to the base DB object, which technically doesn't have a key. So I made DBM::Deep return "[base]" as the key name in that special case. CAVEATS / ISSUES / BUGS This section describes all the known issues with DBM::Deep. It you have found something that is not listed here, please send e-mail to jhuckaby@cpan.org. UNUSED SPACE RECOVERY One major caveat with Deep is that space occupied by existing keys and values is not recovered when they are deleted. Meaning if you keep deleting and adding new keys, your file will continuously grow. I am working on this, but in the meantime you can call the built-in "optimize()" method from time to time (perhaps in a crontab or something) to recover all your unused space. $db->optimize(); # returns true on success This rebuilds the ENTIRE database into a new file, then moves it on top of the original. The new file will have no unused space, thus it will take up as little disk space as possible. Please note that this operation can take a long time for large files, and you need enough disk space to temporarily hold 2 copies of your DB file. The temporary file is created in the same directory as the original, named with a ".tmp" extension, and is deleted when the operation completes. Oh, and if locking is enabled, the DB is automatically locked for the entire duration of the copy. WARNING: Only call optimize() on the top-level node of the database, and make sure there are no child references lying around. Deep keeps a reference counter, and if it is greater than 1, optimize() will abort and return undef. AUTOVIVIFICATION Unfortunately, autovivification doesn't always work. This appears to be a bug in Perl's tie() system, as *Jakob Schmidt* encountered the very same issue with his *DWH_FIle* module (see ), and it is also mentioned in the BUGS section for the *MLDBM* module ). Basically, your milage may vary when issuing statements like this: $db->{a} = { b => [ 1, 2, { c => [ 'd', { e => 'f' } ] } ] }; This causes 3 hashes and 2 arrays to be created in the database all in one fell swoop, and all nested within each other. Perl *may* choke on this, and fail to create one or more of the nested structures. This doesn't appear to be a bug in DBM::Deep, but I am still investigating it. The problem is intermittent. For safety, I recommend creating nested structures using a series of commands instead of just one, which will always work: $db->{a} = {}; $db->{a}->{b} = []; my $b = $db->{a}->{b}; $b->[0] = 1; $b->[1] = 2; $b->[2] = {}; $b->[2]->{c} = []; my $c = $b->[2]->{c}; $c->[0] = 'd'; $c->[1] = {}; $c->[1]->{e} = 'f'; undef $c; undef $b; Also, you can just create the whole structure in memory using a temporary variable, then use DBM::Deep's "import()" method to import the entire thing into the database: my $temp = { a => { b => [ 1, 2, { c => [ 'd', { e => 'f' } ] } ] } }; $db->import( $temp ); Note: I have yet to recreate this bug with Perl 5.8.1. Perhaps the issue has been resolved? Will update as events warrant. FILE CORRUPTION The current level of error handling in Deep is minimal. Files *are* checked for a 32-bit signature on open(), but other corruption in files can cause segmentation faults. Deep may try to seek() past the end of a file, or get stuck in an infinite loop depending on the level of corruption. File write operations are not checked for failure (for speed), so if you happen to run out of disk space, Deep will probably fail in a bad way. These things will be addressed in a later version of DBM::Deep. DB OVER NFS Beware of using DB files over NFS. Deep uses flock(), which works well on local filesystems, but will NOT protect you from file corruption over NFS. I've heard about setting up your NFS server with a locking daemon, then using lockf() to lock your files, but your milage may vary there as well. From what I understand, there is no real way to do it. However, if you need access to the underlying FileHandle in Deep for using some other kind of locking scheme like lockf(), see the "LOW-LEVEL ACCESS" section above. COPYING OBJECTS Beware of copying tied objects in Perl. Very strange things can happen. Instead, use Deep's "clone()" method which safely copies the object and returns a new, blessed, tied hash or array to the same level in the DB. my $copy = $db->clone(); LARGE ARRAYS Beware of using "shift()", "unshift()" or "splice()" with large arrays. These functions cause every element in the array to move, which can be murder on DBM::Deep, as every element has to be fetched from disk, then stored again in a different location. This may be addressed in a later version. PERFORMANCE This section discusses DBM::Deep's speed and memory usage. SPEED Obviously, DBM::Deep isn't going to be as fast as some C-based DBMs, such as the almighty *BerkeleyDB*. But it makes up for it in features like true multi-level hash/array support, and cross-platform FTPable files. Even so, DBM::Deep is still pretty fast, and the speed stays fairly consistent, even with huge databases. Here is some test data: Adding 1,000,000 keys to new DB file... At 100 keys, avg. speed is 2,703 keys/sec At 200 keys, avg. speed is 2,642 keys/sec At 300 keys, avg. speed is 2,598 keys/sec At 400 keys, avg. speed is 2,578 keys/sec At 500 keys, avg. speed is 2,722 keys/sec At 600 keys, avg. speed is 2,628 keys/sec At 700 keys, avg. speed is 2,700 keys/sec At 800 keys, avg. speed is 2,607 keys/sec At 900 keys, avg. speed is 2,190 keys/sec At 1,000 keys, avg. speed is 2,570 keys/sec At 2,000 keys, avg. speed is 2,417 keys/sec At 3,000 keys, avg. speed is 1,982 keys/sec At 4,000 keys, avg. speed is 1,568 keys/sec At 5,000 keys, avg. speed is 1,533 keys/sec At 6,000 keys, avg. speed is 1,787 keys/sec At 7,000 keys, avg. speed is 1,977 keys/sec At 8,000 keys, avg. speed is 2,028 keys/sec At 9,000 keys, avg. speed is 2,077 keys/sec At 10,000 keys, avg. speed is 2,031 keys/sec At 20,000 keys, avg. speed is 1,970 keys/sec At 30,000 keys, avg. speed is 2,050 keys/sec At 40,000 keys, avg. speed is 2,073 keys/sec At 50,000 keys, avg. speed is 1,973 keys/sec At 60,000 keys, avg. speed is 1,914 keys/sec At 70,000 keys, avg. speed is 2,091 keys/sec At 80,000 keys, avg. speed is 2,103 keys/sec At 90,000 keys, avg. speed is 1,886 keys/sec At 100,000 keys, avg. speed is 1,970 keys/sec At 200,000 keys, avg. speed is 2,053 keys/sec At 300,000 keys, avg. speed is 1,697 keys/sec At 400,000 keys, avg. speed is 1,838 keys/sec At 500,000 keys, avg. speed is 1,941 keys/sec At 600,000 keys, avg. speed is 1,930 keys/sec At 700,000 keys, avg. speed is 1,735 keys/sec At 800,000 keys, avg. speed is 1,795 keys/sec At 900,000 keys, avg. speed is 1,221 keys/sec At 1,000,000 keys, avg. speed is 1,077 keys/sec This test was performed on a PowerMac G4 1gHz running Mac OS X 10.3.2 & Perl 5.8.1, with an 80GB Ultra ATA/100 HD spinning at 7200RPM. The hash keys and values were between 6 - 12 chars in length. The DB file ended up at 210MB. Run time was 12 min 3 sec. MEMORY USAGE One of the great things about DBM::Deep is that it uses very little memory. Even with huge databases (1,000,000+ keys) you will not see much increased memory on your process. Deep relies solely on the filesystem for storing and fetching data. Here is output from */usr/bin/top* before even opening a database handle: PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME COMMAND 22831 root 11 0 2716 2716 1296 R 0.0 0.2 0:07 perl Basically the process is taking 2,716K of memory. And here is the same process after storing and fetching 1,000,000 keys: PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME COMMAND 22831 root 14 0 2772 2772 1328 R 0.0 0.2 13:32 perl Notice the memory usage increased by only 56K. Test was performed on a 700mHz x86 box running Linux RedHat 7.2 & Perl 5.6.1. DB FILE FORMAT In case you were interested in the underlying DB file format, it is documented here in this section. You don't need to know this to use the module, it's just included for reference. SIGNATURE DBM::Deep files always start with a 32-bit signature to identify the file type. This is at offset 0. The signature is "DPDB" in network byte order. This is checked upon each file open(). TAG The DBM::Deep file is in a *tagged format*, meaning each section of the file has a standard header containing the type of data, the length of data, and then the data itself. The type is a single character (1 byte), the length is a 32-bit unsigned long in network byte order, and the data is, well, the data. Here is how it unfolds: MASTER INDEX Immediately after the 32-bit file signature is the *Master Index* record. This is a standard tag header followed by 1024 bytes (in 32-bit mode) or 2048 bytes (in 64-bit mode) of data. The type is *H* for hash or *A* for array, depending on how the DBM::Deep object was constructed. The index works by looking at a *MD5 Hash* of the hash key (or array index number). The first 8-bit char of the MD5 signature is the offset into the index, multipled by 4 in 32-bit mode, or 8 in 64-bit mode. The value of the index element is a file offset of the next tag for the key/element in question, which is usually a *Bucket List* tag (see below). The next tag *could* be another index, depending on how many keys/elements exist. See RE-INDEXING below for details. BUCKET LIST A *Bucket List* is a collection of 16 MD5 hashes for keys/elements, plus file offsets to where the actual data is stored. It starts with a standard tag header, with type *B*, and a data size of 320 bytes in 32-bit mode, or 384 bytes in 64-bit mode. Each MD5 hash is stored in full (16 bytes), plus the 32-bit or 64-bit file offset for the *Bucket* containing the actual data. When the list fills up, a *Re-Index* operation is performed (See RE-INDEXING below). BUCKET A *Bucket* is a tag containing a key/value pair (in hash mode), or a index/value pair (in array mode). It starts with a standard tag header with type *D* for scalar data (string, binary, etc.), or it could be a nested hash (type *H*) or array (type *A*). The value comes just after the tag header. The size reported in the tag header is only for the value, but then, just after the value is another size (32-bit unsigned long) and then the plain key itself. Since the value is likely to be fetched more often than the plain key, I figured it would be *slightly* faster to store the value first. If the type is *H* (hash) or *A* (array), the value is another *Master Index* record for the nested structure, where the process begins all over again. RE-INDEXING After a *Bucket List* grows to 16 records, its allocated space in the file is exhausted. Then, when another key/element comes in, the list is converted to a new index record. However, this index will look at the next char in the MD5 hash, and arrange new Bucket List pointers accordingly. This process is called *Re-Indexing*. Basically, a new index tag is created at the file EOF, and all 17 (16 + new one) keys/elements are removed from the old Bucket List and inserted into the new index. Several new Bucket Lists are created in the process, as a new MD5 char from the key is being examined (it is unlikely that the keys will all share the same next char of their MD5s). Because of the way the *MD5* algorithm works, it is impossible to tell exactly when the Bucket Lists will turn into indexes, but the first round tends to happen right around 4,000 keys. You will see a *slight* decrease in performance here, but it picks back up pretty quick (see SPEED above). Then it takes a lot more keys to exhaust the next level of Bucket Lists. It's right around 900,000 keys. This process can continue nearly indefinitely -- right up until the point the *MD5* signatures start colliding with each other, and this is EXTREMELY rare -- like winning the lottery 5 times in a row AND getting struck by lightning while you are walking to cash in your tickets. Theoretically, since *MD5* hashes are 128-bit values, you *could* have up to 340,282,366,921,000,000,000,000,000,000,000,000,000 keys/elements (I believe this is 340 unodecillion, but don't quote me). STORING When a new key/element is stored, the key (or index number) is first ran through *Digest::MD5* to get a 128-bit signature (example, in hex: b05783b0773d894396d475ced9d2f4f6). Then, the *Master Index* record is checked for the first char of the signature (in this case *b*). If it does not exist, a new *Bucket List* is created for our key (and the next 15 future keys that happen to also have *b* as their first MD5 char). The entire MD5 is written to the *Bucket List* along with the offset of the new *Bucket* record (EOF at this point, unless we are replacing an existing *Bucket*), where the actual data will be stored. FETCHING Fetching an existing key/element involves getting a *Digest::MD5* of the key (or index number), then walking along the indexes. If there are enough keys/elements in this DB level, there might be nested indexes, each linked to a particular char of the MD5. Finally, a *Bucket List* is pointed to, which contains up to 16 full MD5 hashes. Each is checked for equality to the key in question. If we found a match, the *Bucket* tag is loaded, where the value and plain key are stored. Fetching the plain key occurs when calling the *first_key()* and *next_key()* methods. In this process the indexes are walked systematically, and each key fetched in increasing MD5 order (which is why it appears random). Once the *Bucket* is found, the value is skipped the plain key returned instead. Note: Do not count on keys being fetched as if the MD5 hashes were alphabetically sorted. This only happens on an index-level -- as soon as the *Bucket Lists* are hit, the keys will come out in the order they went in -- so it's pretty much undefined how the keys will come out -- just like Perl's built-in hashes. AUTHOR Joseph Huckaby, jhuckaby@cpan.org Special thanks to Adam Sah and Rich Gaushell! You know why :-) SEE ALSO perltie(1), Tie::Hash(3), Digest::MD5(3), Fcntl(3), flock(2), lockf(3), nfs(5), Digest::SHA256(3), Crypt::Blowfish(3), Compress::Zlib(3) LICENSE Copyright (c) 2002-2004 Joseph Huckaby. All Rights Reserved. This is free software, you may use it and distribute it under the same terms as Perl itself.