package Data::ObjectDriver::Driver::SimplePartition; use strict; use warnings; use base qw( Data::ObjectDriver::Driver::Partition ); use Carp qw( croak ); use Data::Dumper; use Data::ObjectDriver::Driver::MultiPartition; sub init { my $driver = shift; my %param = @_; my $class = delete $param{using} or croak "using is required"; my @extra = %param; $param{get_driver} = _make_get_driver($class, \@extra); $driver->SUPER::init(%param); return $driver; } sub _make_get_driver { my($class, $extra) = @_; $extra ||= []; ## Make sure we've loaded the parent class that contains information ## about our partitioning scheme. croak "Bogus classname." unless $class =~ /^[\w:]+$/; "eval $class; 1;" or die "Failed to load parent class: $@\n"; my $col = $class->primary_key_tuple->[0]; my $get_driver = $class->properties->{partition_get_driver} or croak "Partitioning driver not defined for $class"; my $number = $class->properties->{number_of_partitions}; my $mp_driver = Data::ObjectDriver::Driver::MultiPartition->new( partitions => [ map { $get_driver->($_, @$extra) } (1 .. $number) ] ); return sub { my($terms, $args) = @_; my $parent_id; if (ref($terms) eq 'HASH') { $parent_id = $terms->{ $col }; } elsif (ref($terms) eq 'ARRAY') { ## An array ref could either be a multiple-column primary key OR ## a list of primary keys. With a multiple-column primary key, the ## $id is an array ref, where the first column is always the ## parent_id. $parent_id = ref($terms->[0]) eq 'ARRAY' ? $terms->[0][0] : $terms->[0]; } if ($parent_id) { my $parent = $class->driver->lookup($class, $parent_id) or croak "Member of $class with ID $parent_id not found"; return $get_driver->( $parent->partition_id, @$extra ); } else { unless($args->{multi_partition}) { croak "Cannot extract $col from terms ", Dumper($terms); } return $mp_driver; } }; } 1; __END__ =head1 NAME Data::ObjectDriver::Driver::SimplePartition - basic partitioned object driver =head1 SYNOPSIS package ParentObject; use base qw( Data::ObjectDriver::BaseObject ); __PACKAGE__->install_properties({ columns => [ 'parent_id', 'partition_id', ... ], ... driver => Data::ObjectDriver::Driver::DBI->new( @$GLOBAL_DB_INFO ), primary_key => 'parent_id', }); __PACKAGE__->has_partitions( number => scalar @PARTITIONS, get_driver => \&get_driver_by_partition, ); package SomeObject; use base qw( Data::ObjectDriver::BaseObject ); __PACKAGE__->install_properties({ ... driver => Data::ObjectDriver::Driver::SimplePartition->new( using => 'ParentObject' ), primary_key => ['parent_id', 'object_id'], }); =head1 DESCRIPTION I is a basic driver for objects partitioned into separate databases. See L for more about partitioning databases. I helps you partition objects into databases based on their association with one record of a I class. If your classes don't meet the requirements imposed by I, you can still write your own partitioning driver. See L. =head1 SUGGESTED PRACTICES Often this is used for user partitioning, where the parent class is your user account class; all records of other classes that are "owned" by that user are partitioned into the same database. This allows you to scale horizontally with the number of users, at the cost of complicating querying multiple users' data together. I will load the related instance of the parent class every time it needs to find the partition for a related object. Consider using a minimal mapping class for the parent, keeping as much data as possible in other related classes. For example, if C were your parent class, you might keep I the user ID and other data used to find users (such as login name and email address) in C, keeping further profile data in another C class. As all the partitioned classes related to a given parent class will share the same C logic to turn a partition ID into a driver, you might put the C function in the parent class, or use a custom subclass of I that contains and automatically specifies the C function. =head1 USAGE =head2 Data::ObjectDriver::Driver::SimplePartition->new(%params) Creates a new basic partitioning driver for a particular class. The required members of C<%params> are: =over 4 =item * C The name of the parent class on which the driven class is partitioned. Using a class as a parent partitioned class requires these properties to be defined: =over 4 =item * C The parent class must have a C column containing a partition identifier. This identifier is passed to the C function to identify a driver to return. =item * C The parent class's primary key must be a simple single-column key, and that column must be the same as the referencing column in the partitioned classes. =item * C The C property must be a function that returns an object driver, given a partition ID and any extra parameters given to the C constructor. This property can also be defined as C in a call to Chas_partitions()>. See L. =back =back You can also include any further optional parameters you like. They will be passed to the partitioned class's C function as given. A I driver will require these properties to be defined for partitioned classes: =over 4 =item * C Your primary key should be a complex primary key (arrayref) with the simple key of the parent object for the first field. =back =head1 DIAGNOSTICS =over 4 =item * C The C parameter to the I constructor is required to create the partitioned class's C function. Perhaps you omitted it, or your subclass of I did not properly specify it to its parent's constructor. =item * C The parent class name you specified in your C parameter does not appear to be a valid class name. If you are automatically generating parent class names, check that your method of converting strings to class names is correct. =item * C> The parent class you specified in your C parameter could not be loaded, for the given reason. Perhaps you didn't include its location in your library path. =item * C> The partitioned class named in the error is configured to use the I driver but does not have a C set. Check that you intended to use I with that class or, if you're automatically specifying the C function, that your technique is working correctly. =item * C from terms I> The I driver could not determine from the given search terms or object key what the ID of the related parent record was. Check that your columns in the partitioned and parent classes share the same name, and that your application includes the parent ID in all C calls for the partitioned class and instances of partitioned objects before attempting to save them. Optionaly you can enable a basic support of search accross multiple partition by passing the 'multi_partition' arg (true value) to the search query. =item * C with ID I not found> The parent record associated with the partitioned object could not be loaded. Perhaps your application deleted the parent record without removing its associated partitioned objects first. =back =head1 BUGS AND LIMITATIONS There are no known bugs in this module. =head1 SEE ALSO L =head1 LICENSE I is free software; you may redistribute it and/or modify it under the same terms as Perl itself. =head1 AUTHOR & COPYRIGHT Except where otherwise noted, I is Copyright 2005-2006 Six Apart, cpan@sixapart.com. All rights reserved. =cut