fixed regression in the random balancer that I created when I removed the shuffle...

[dbsrgits/DBIx-Class.git] / lib / DBIx / Class / Storage / DBI / Replicated / Balancer.pm

package DBIx::Class::Storage::DBI::Replicated::Balancer;

use Moose::Role;
requires 'next_storage';

=head1 NAME

DBIx::Class::Storage::DBI::Replicated::Balancer; A Software Load Balancer 

=head1 SYNOPSIS

This role is used internally by L<DBIx::Class::Storage::DBI::Replicated>.
    
=head1 DESCRIPTION

Given a pool (L<DBIx::Class::Storage::DBI::Replicated::Pool>) of replicated
database's (L<DBIx::Class::Storage::DBI::Replicated::Replicant>), defines a
method by which query load can be spread out across each replicant in the pool.

=head1 ATTRIBUTES

This class defines the following attributes.

=head2 auto_validate_every ($seconds)

If auto_validate has some sort of value, run the L<validate_replicants> every
$seconds.  Be careful with this, because if you set it to 0 you will end up
validating every query.

=cut

has 'auto_validate_every' => (
  is=>'rw',
  isa=>'Int',
  predicate=>'has_auto_validate_every',
);

=head2 master

The L<DBIx::Class::Storage::DBI> object that is the master database all the
replicants are trying to follow.  The balancer needs to know it since it's the
ultimate fallback.

=cut

has 'master' => (
  is=>'ro',
  isa=>'DBIx::Class::Storage::DBI',
  required=>1,
);

=head2 pool

The L<DBIx::Class::Storage::DBI::Replicated::Pool> object that we are trying to
balance.

=cut

has 'pool' => (
  is=>'ro',
  isa=>'DBIx::Class::Storage::DBI::Replicated::Pool',
  required=>1,
);

=head2 current_replicant

Replicant storages (slaves) handle all read only traffic.  The assumption is
that your database will become readbound well before it becomes write bound
and that being able to spread your read only traffic around to multiple 
databases is going to help you to scale traffic.

This attribute returns the next slave to handle a read request.  Your L</pool>
attribute has methods to help you shuffle through all the available replicants
via it's balancer object.

=cut

has 'current_replicant' => (
  is=> 'rw',
  isa=>'DBIx::Class::Storage::DBI',
  lazy_build=>1,
  handles=>[qw/
    select
    select_single
    columns_info_for
  /],
);

=head1 METHODS

This class defines the following methods.

=head2 _build_current_replicant

Lazy builder for the L</current_replicant_storage> attribute.

=cut

sub _build_current_replicant {
  my $self = shift @_;
  $self->next_storage;
}

=head2 next_storage

This method should be defined in the class which consumes this role.

Given a pool object, return the next replicant that will serve queries.  The
default behavior is to grap the first replicant it finds but you can write 
your own subclasses of L<DBIx::Class::Storage::DBI::Replicated::Balancer> to 
support other balance systems.

This returns from the pool of active replicants.  If there are no active
replicants, then you should have it return the master as an ultimate fallback.

=head2 around: next_storage

Advice on next storage to add the autovalidation.  We have this broken out so
that it's easier to break out the auto validation into a role.

This also returns the master in the case that none of the replicants are active
or just just forgot to create them :)

=cut

around 'next_storage' => sub {
  my ($next_storage, $self, @args) = @_;
  my $now = time;
    
  ## Do we need to validate the replicants?
  if(
     $self->has_auto_validate_every && 
     ($self->auto_validate_every + $self->pool->last_validated) <= $now
  ) {
      $self->pool->validate_replicants;
  }
    
  ## Get a replicant, or the master if none
  if(my $next = $self->$next_storage(@args)) {
    return $next;
  } else {
    return $self->master;
  }
};

=head2 increment_storage

Rolls the Storage to whatever is next in the queue, as defined by the Balancer.

=cut

sub increment_storage {
  my $self = shift @_;
  my $next_replicant = $self->next_storage;
  $self->current_replicant($next_replicant);
}

=head2 around: select

Advice on the select attribute.  Each time we use a replicant
we need to change it via the storage pool algorithm.  That way we are spreading
the load evenly (hopefully) across existing capacity.

=cut

around 'select' => sub {
  my ($select, $self, @args) = @_;
  
  if (my $forced_pool = $args[-1]->{force_pool}) {
    delete $args[-1]->{force_pool};
    return $self->_get_forced_pool($forced_pool)->select(@args); 
  } else {
    $self->increment_storage;
    return $self->$select(@args);
  }
};

=head2 around: select_single

Advice on the select_single attribute.  Each time we use a replicant
we need to change it via the storage pool algorithm.  That way we are spreading
the load evenly (hopefully) across existing capacity.

=cut

around 'select_single' => sub {
  my ($select_single, $self, @args) = @_;
  
  if (my $forced_pool = $args[-1]->{force_pool}) {
    delete $args[-1]->{force_pool};
    return $self->_get_forced_pool($forced_pool)->select_single(@args); 
  } else {
    $self->increment_storage;
    return $self->$select_single(@args);
  }
};

=head2 before: columns_info_for

Advice on the current_replicant_storage attribute.  Each time we use a replicant
we need to change it via the storage pool algorithm.  That way we are spreading
the load evenly (hopefully) across existing capacity.

=cut

before 'columns_info_for' => sub {
  my $self = shift @_;
  $self->increment_storage;
};

=head2 _get_forced_pool ($name)

Given an identifier, find the most correct storage object to handle the query.

=cut

sub _get_forced_pool {
  my ($self, $forced_pool) = @_;
  if(blessed $forced_pool) {
    return $forced_pool;
  } elsif($forced_pool eq 'master') {
    return $self->master;
  } elsif(my $replicant = $self->pool->replicants($forced_pool)) {
    return $replicant;
  } else {
    $self->master->throw_exception("$forced_pool is not a named replicant.");
  }   
}

=head1 AUTHOR

John Napiorkowski <john.napiorkowski@takkle.com>

=head1 LICENSE

You may distribute this code under the same terms as Perl itself.

=cut

1;