1 package DBIx::Class::Storage::DBI::Replicated;
5 die('The following modules are required for Replication ' . DBIx::Class::Optional::Dependencies->req_missing_for ('replicated') . "\n" )
6 unless DBIx::Class::Optional::Dependencies->req_ok_for ('replicated');
10 use DBIx::Class::Storage::DBI;
11 use DBIx::Class::Storage::DBI::Replicated::Pool;
12 use DBIx::Class::Storage::DBI::Replicated::Balancer;
13 use DBIx::Class::Storage::DBI::Replicated::Types qw/BalancerClassNamePart DBICSchema DBICStorageDBI/;
14 use MooseX::Types::Moose qw/ClassName HashRef Object/;
15 use Scalar::Util 'reftype';
17 use List::Util qw/min max reduce/;
18 use Context::Preserve 'preserve_context';
21 use namespace::clean -except => 'meta';
27 DBIx::Class::Storage::DBI::Replicated - BETA Replicated database support
31 The Following example shows how to change an existing $schema to a replicated
32 storage type, add some replicated (read-only) databases, and perform reporting
35 You should set the 'storage_type attribute to a replicated type. You should
36 also define your arguments, such as which balancer you want and any arguments
37 that the Pool object should get.
39 my $schema = Schema::Class->clone;
40 $schema->storage_type(['::DBI::Replicated', { balancer_type => '::Random' }]);
41 $schema->connection(...);
43 Next, you need to add in the Replicants. Basically this is an array of
44 arrayrefs, where each arrayref is database connect information. Think of these
45 arguments as what you'd pass to the 'normal' $schema->connect method.
47 $schema->storage->connect_replicants(
48 [$dsn1, $user, $pass, \%opts],
49 [$dsn2, $user, $pass, \%opts],
50 [$dsn3, $user, $pass, \%opts],
53 Now, just use the $schema as you normally would. Automatically all reads will
54 be delegated to the replicants, while writes to the master.
56 $schema->resultset('Source')->search({name=>'etc'});
58 You can force a given query to use a particular storage using the search
59 attribute 'force_pool'. For example:
61 my $rs = $schema->resultset('Source')->search(undef, {force_pool=>'master'});
63 Now $rs will force everything (both reads and writes) to use whatever was setup
64 as the master storage. 'master' is hardcoded to always point to the Master,
65 but you can also use any Replicant name. Please see:
66 L<DBIx::Class::Storage::DBI::Replicated::Pool> and the replicants attribute for more.
68 Also see transactions and L</execute_reliably> for alternative ways to
69 force read traffic to the master. In general, you should wrap your statements
70 in a transaction when you are reading and writing to the same tables at the
71 same time, since your replicants will often lag a bit behind the master.
73 If you have a multi-statement read only transaction you can force it to select
74 a random server in the pool by:
76 my $rs = $schema->resultset('Source')->search( undef,
77 { force_pool => $db->storage->read_handler->next_storage }
82 Warning: This class is marked BETA. This has been running a production
83 website using MySQL native replication as its backend and we have some decent
84 test coverage but the code hasn't yet been stressed by a variety of databases.
85 Individual DBs may have quirks we are not aware of. Please use this in first
86 development and pass along your experiences/bug fixes.
88 This class implements replicated data store for DBI. Currently you can define
89 one master and numerous slave database connections. All write-type queries
90 (INSERT, UPDATE, DELETE and even LAST_INSERT_ID) are routed to master
91 database, all read-type queries (SELECTs) go to the slave database.
93 Basically, any method request that L<DBIx::Class::Storage::DBI> would normally
94 handle gets delegated to one of the two attributes: L</read_handler> or to
95 L</write_handler>. Additionally, some methods need to be distributed
96 to all existing storages. This way our storage class is a drop in replacement
97 for L<DBIx::Class::Storage::DBI>.
99 Read traffic is spread across the replicants (slaves) occurring to a user
100 selected algorithm. The default algorithm is random weighted.
104 The consistency between master and replicants is database specific. The Pool
105 gives you a method to validate its replicants, removing and replacing them
106 when they fail/pass predefined criteria. Please make careful use of the ways
107 to force a query to run against Master when needed.
111 Replicated Storage has additional requirements not currently part of
112 L<DBIx::Class>. See L<DBIx::Class::Optional::Dependencies> for more details.
116 This class defines the following attributes.
120 The underlying L<DBIx::Class::Schema> object this storage is attaching
133 Contains the classname which will instantiate the L</pool> object. Defaults
134 to: L<DBIx::Class::Storage::DBI::Replicated::Pool>.
141 default=>'DBIx::Class::Storage::DBI::Replicated::Pool',
143 'create_pool' => 'new',
149 Contains a hashref of initialized information to pass to the Balancer object.
150 See L<DBIx::Class::Storage::DBI::Replicated::Pool> for available arguments.
164 The replication pool requires a balance class to provider the methods for
165 choose how to spread the query load across each replicant in the pool.
169 has 'balancer_type' => (
171 isa=>BalancerClassNamePart,
174 default=> 'DBIx::Class::Storage::DBI::Replicated::Balancer::First',
176 'create_balancer' => 'new',
182 Contains a hashref of initialized information to pass to the Balancer object.
183 See L<DBIx::Class::Storage::DBI::Replicated::Balancer> for available arguments.
187 has 'balancer_args' => (
197 Is a L<DBIx::Class::Storage::DBI::Replicated::Pool> or derived class. This is a
198 container class for one or more replicated databases.
204 isa=>'DBIx::Class::Storage::DBI::Replicated::Pool',
215 Is a L<DBIx::Class::Storage::DBI::Replicated::Balancer> or derived class. This
216 is a class that takes a pool (L<DBIx::Class::Storage::DBI::Replicated::Pool>)
222 isa=>'DBIx::Class::Storage::DBI::Replicated::Balancer',
224 handles=>[qw/auto_validate_every/],
229 The master defines the canonical state for a pool of connected databases. All
230 the replicants are expected to match this databases state. Thus, in a classic
231 Master / Slaves distributed system, all the slaves are expected to replicate
232 the Master's state as quick as possible. This is the only database in the
233 pool of databases that is allowed to handle write traffic.
243 =head1 ATTRIBUTES IMPLEMENTING THE DBIx::Storage::DBI INTERFACE
245 The following methods are delegated all the methods required for the
246 L<DBIx::Class::Storage::DBI> interface.
250 my $method_dispatch = {
262 deployment_statements
265 build_datetime_parser
280 with_deferred_fk_checks
288 relname_to_table_alias
290 _default_dbi_connect_attributes
292 _dbic_connect_attributes
298 bind_attribute_by_data_type
302 _dbh_execute_for_fetch
304 _dbh_execute_inserts_with_no_binds
305 _select_args_to_query
308 _normalize_connect_info
322 /, Class::MOP::Class->initialize('DBIx::Class::Storage::DBIHacks')->get_method_list ],
327 _dbh_columns_info_for
330 unimplemented => [qw/
331 _arm_global_destructor
334 get_use_dbms_capability
335 set_use_dbms_capability
341 _determine_connector_driver
342 _extract_driver_from_connect_info
344 _warn_undetermined_driver
351 _perform_autoinc_retrieval
352 _autoinc_supplied_for_op
365 # the capability framework
366 # not sure if CMOP->initialize does evil things to DBIC::S::DBI, fix if a problem
368 { $_ =~ /^ _ (?: use | supports | determine_supports ) _ /x }
369 ( Class::MOP::Class->initialize('DBIx::Class::Storage::DBI')->get_all_method_names )
373 if (DBIx::Class::_ENV_::DBICTEST) {
376 for my $type (keys %$method_dispatch) {
377 for (@{$method_dispatch->{$type}}) {
378 push @{$seen->{$_}}, $type;
382 if (my @dupes = grep { @{$seen->{$_}} > 1 } keys %$seen) {
384 'The following methods show up multiple times in ::Storage::DBI::Replicated handlers:',
385 (map { "$_: " . (join ', ', @{$seen->{$_}}) } sort @dupes),
390 if (my @cant = grep { ! DBIx::Class::Storage::DBI->can($_) } keys %$seen) {
392 '::Storage::DBI::Replicated specifies handling of the following *NON EXISTING* ::Storage::DBI methods:',
399 for my $method (@{$method_dispatch->{unimplemented}}) {
400 __PACKAGE__->meta->add_method($method, sub {
402 $self->throw_exception("$method() must not be called on ".(blessed $self).' objects');
408 Defines an object that implements the read side of L<BIx::Class::Storage::DBI>.
412 has 'read_handler' => (
416 handles=>$method_dispatch->{reader},
421 Defines an object that implements the write side of L<BIx::Class::Storage::DBI>,
422 as well as methods that don't write or read that can be called on only one
423 storage, methods that return a C<$dbh>, and any methods that don't make sense to
428 has 'write_handler' => (
432 handles=>$method_dispatch->{writer},
437 has _master_connect_info_opts =>
438 (is => 'rw', isa => HashRef, default => sub { {} });
440 =head2 around: connect_info
442 Preserves master's C<connect_info> options (for merging with replicants.)
443 Also sets any Replicated-related options from connect_info, such as
444 C<pool_type>, C<pool_args>, C<balancer_type> and C<balancer_args>.
448 around connect_info => sub {
449 my ($next, $self, $info, @extra) = @_;
451 $self->throw_exception(
452 'connect_info can not be retrieved from a replicated storage - '
453 . 'accessor must be called on a specific pool instance'
454 ) unless defined $info;
456 my $merge = Hash::Merge->new('LEFT_PRECEDENT');
459 for my $arg (@$info) {
460 next unless (reftype($arg)||'') eq 'HASH';
461 %opts = %{ $merge->merge($arg, \%opts) };
465 if (@opts{qw/pool_type pool_args/}) {
466 $self->pool_type(delete $opts{pool_type})
470 $merge->merge((delete $opts{pool_args} || {}), $self->pool_args)
473 ## Since we possibly changed the pool_args, we need to clear the current
474 ## pool object so that next time it is used it will be rebuilt.
478 if (@opts{qw/balancer_type balancer_args/}) {
479 $self->balancer_type(delete $opts{balancer_type})
480 if $opts{balancer_type};
482 $self->balancer_args(
483 $merge->merge((delete $opts{balancer_args} || {}), $self->balancer_args)
486 $self->balancer($self->_build_balancer)
490 $self->_master_connect_info_opts(\%opts);
492 return preserve_context {
493 $self->$next($info, @extra);
495 # Make sure master is blessed into the correct class and apply role to it.
496 my $master = $self->master;
497 $master->_determine_driver;
498 Moose::Meta::Class->initialize(ref $master);
500 DBIx::Class::Storage::DBI::Replicated::WithDSN->meta->apply($master);
502 # link pool back to master
503 $self->pool->master($master);
509 This class defines the following methods.
513 L<DBIx::Class::Schema> when instantiating its storage passed itself as the
514 first argument. So we need to massage the arguments a bit so that all the
515 bits get put into the correct places.
520 my ($class, $schema, $storage_type_args, @args) = @_;
531 Lazy builder for the L</master> attribute.
537 my $master = DBIx::Class::Storage::DBI->new($self->schema);
543 Lazy builder for the L</pool> attribute.
549 $self->create_pool(%{$self->pool_args});
552 =head2 _build_balancer
554 Lazy builder for the L</balancer> attribute. This takes a Pool object so that
555 the balancer knows which pool it's balancing.
559 sub _build_balancer {
561 $self->create_balancer(
563 master=>$self->master,
564 %{$self->balancer_args},
568 =head2 _build_write_handler
570 Lazy builder for the L</write_handler> attribute. The default is to set this to
575 sub _build_write_handler {
576 return shift->master;
579 =head2 _build_read_handler
581 Lazy builder for the L</read_handler> attribute. The default is to set this to
586 sub _build_read_handler {
587 return shift->balancer;
590 =head2 around: connect_replicants
592 All calls to connect_replicants needs to have an existing $schema tacked onto
593 top of the args, since L<DBIx::Storage::DBI> needs it, and any C<connect_info>
594 options merged with the master, with replicant opts having higher priority.
598 around connect_replicants => sub {
599 my ($next, $self, @args) = @_;
602 $r = [ $r ] unless reftype $r eq 'ARRAY';
604 $self->throw_exception('coderef replicant connect_info not supported')
605 if ref $r->[0] && reftype $r->[0] eq 'CODE';
607 # any connect_info options?
609 $i++ while $i < @$r && (reftype($r->[$i])||'') ne 'HASH';
612 $r->[$i] = {} unless $r->[$i];
614 # merge if two hashes
615 my @hashes = @$r[$i .. $#{$r}];
617 $self->throw_exception('invalid connect_info options')
618 if (grep { reftype($_) eq 'HASH' } @hashes) != @hashes;
620 $self->throw_exception('too many hashrefs in connect_info')
623 my $merge = Hash::Merge->new('LEFT_PRECEDENT');
624 my %opts = %{ $merge->merge(reverse @hashes) };
627 splice @$r, $i+1, ($#{$r} - $i), ();
629 # make sure master/replicants opts don't clash
630 my %master_opts = %{ $self->_master_connect_info_opts };
631 if (exists $opts{dbh_maker}) {
632 delete @master_opts{qw/dsn user password/};
634 delete $master_opts{dbh_maker};
637 %opts = %{ $merge->merge(\%opts, \%master_opts) };
643 $self->$next($self->schema, @args);
648 Returns an array of all the connected storage backends. The first element
649 in the returned array is the master, and the rest are each of the
656 return grep {defined $_ && blessed $_} (
658 values %{ $self->replicants },
662 =head2 execute_reliably ($coderef, ?@args)
664 Given a coderef, saves the current state of the L</read_handler>, forces it to
665 use reliable storage (e.g. sets it to the master), executes a coderef and then
666 restores the original state.
672 $schema->resultset('User')->create({name=>$name});
673 my $user_rs = $schema->resultset('User')->find({name=>$name});
677 my $user_rs = $schema->storage->execute_reliably($reliably, 'John');
679 Use this when you must be certain of your database state, such as when you just
680 inserted something and need to get a resultset including it, etc.
684 sub execute_reliably {
688 unless( ref $coderef eq 'CODE') {
689 $self->throw_exception('Second argument must be a coderef');
692 ## replace the current read handler for the remainder of the scope
693 local $self->{read_handler} = $self->master;
699 $self->throw_exception("coderef returned an error: $_");
703 =head2 set_reliable_storage
705 Sets the current $schema to be 'reliable', that is all queries, both read and
706 write are sent to the master
710 sub set_reliable_storage {
712 my $schema = $self->schema;
713 my $write_handler = $self->schema->storage->write_handler;
715 $schema->storage->read_handler($write_handler);
718 =head2 set_balanced_storage
720 Sets the current $schema to be use the </balancer> for all reads, while all
721 writes are sent to the master only
725 sub set_balanced_storage {
727 my $schema = $self->schema;
728 my $balanced_handler = $self->schema->storage->balancer;
730 $schema->storage->read_handler($balanced_handler);
735 Check that the master and at least one of the replicants is connected.
742 $self->master->connected &&
743 $self->pool->connected_replicants;
746 =head2 ensure_connected
748 Make sure all the storages are connected.
752 sub ensure_connected {
754 foreach my $source ($self->all_storages) {
755 $source->ensure_connected(@_);
761 Set the limit_dialect for all existing storages
767 foreach my $source ($self->all_storages) {
768 $source->limit_dialect(@_);
770 return $self->master->limit_dialect;
775 Set the quote_char for all existing storages
781 foreach my $source ($self->all_storages) {
782 $source->quote_char(@_);
784 return $self->master->quote_char;
789 Set the name_sep for all existing storages
795 foreach my $source ($self->all_storages) {
796 $source->name_sep(@_);
798 return $self->master->name_sep;
803 Set the schema object for all existing storages
809 foreach my $source ($self->all_storages) {
810 $source->set_schema(@_);
816 set a debug flag across all storages
823 foreach my $source ($self->all_storages) {
827 return $self->master->debug;
838 return $self->master->debugobj(@_);
849 return $self->master->debugfh(@_);
860 return $self->master->debugcb(@_);
865 disconnect everything
871 foreach my $source ($self->all_storages) {
872 $source->disconnect(@_);
878 set cursor class on all storages, or return master's
883 my ($self, $cursor_class) = @_;
886 $_->cursor_class($cursor_class) for $self->all_storages;
888 $self->master->cursor_class;
893 set cursor class on all storages, or return master's, alias for L</cursor_class>
899 my ($self, $cursor_class) = @_;
902 $_->cursor($cursor_class) for $self->all_storages;
904 $self->master->cursor;
909 sets the L<DBIx::Class::Storage::DBI/unsafe> option on all storages or returns
910 master's current setting
918 $_->unsafe(@_) for $self->all_storages;
921 return $self->master->unsafe;
924 =head2 disable_sth_caching
926 sets the L<DBIx::Class::Storage::DBI/disable_sth_caching> option on all storages
927 or returns master's current setting
931 sub disable_sth_caching {
935 $_->disable_sth_caching(@_) for $self->all_storages;
938 return $self->master->disable_sth_caching;
941 =head2 lag_behind_master
943 returns the highest Replicant L<DBIx::Class::Storage::DBI/lag_behind_master>
948 sub lag_behind_master {
951 return max map $_->lag_behind_master, $self->replicants;
954 =head2 is_replicating
956 returns true if all replicants return true for
957 L<DBIx::Class::Storage::DBI/is_replicating>
964 return (grep $_->is_replicating, $self->replicants) == ($self->replicants);
967 =head2 connect_call_datetime_setup
969 calls L<DBIx::Class::Storage::DBI/connect_call_datetime_setup> for all storages
973 sub connect_call_datetime_setup {
975 $_->connect_call_datetime_setup for $self->all_storages;
980 $_->_populate_dbh for $self->all_storages;
985 $_->_connect for $self->all_storages;
990 $_->_rebless for $self->all_storages;
993 sub _determine_driver {
995 $_->_determine_driver for $self->all_storages;
998 sub _driver_determined {
1002 $_->_driver_determined(@_) for $self->all_storages;
1005 return $self->master->_driver_determined;
1011 $_->_init for $self->all_storages;
1014 sub _run_connection_actions {
1017 $_->_run_connection_actions for $self->all_storages;
1020 sub _do_connection_actions {
1024 $_->_do_connection_actions(@_) for $self->all_storages;
1028 sub connect_call_do_sql {
1030 $_->connect_call_do_sql(@_) for $self->all_storages;
1033 sub disconnect_call_do_sql {
1035 $_->disconnect_call_do_sql(@_) for $self->all_storages;
1038 sub _seems_connected {
1041 return min map $_->_seems_connected, $self->all_storages;
1047 return min map $_->_ping, $self->all_storages;
1050 # not using the normalized_version, because we want to preserve
1051 # version numbers much longer than the conventional xxx.yyyzzz
1052 my $numify_ver = sub {
1054 my @numparts = split /\D+/, $ver;
1055 my $format = '%d.' . (join '', ('%06d') x (@numparts - 1));
1057 return sprintf $format, @numparts;
1062 if (not $self->_dbh_details->{info}) {
1063 $self->_dbh_details->{info} = (
1064 reduce { $a->[0] < $b->[0] ? $a : $b }
1065 map [ $numify_ver->($_->{dbms_version}), $_ ],
1066 map $_->_server_info, $self->all_storages
1070 return $self->next::method;
1073 sub _get_server_version {
1076 return $self->_server_info->{dbms_version};
1081 Due to the fact that replicants can lag behind a master, you must take care to
1082 make sure you use one of the methods to force read queries to a master should
1083 you need realtime data integrity. For example, if you insert a row, and then
1084 immediately re-read it from the database (say, by doing
1085 L<< $result->discard_changes|DBIx::Class::Row/discard_changes >>)
1086 or you insert a row and then immediately build a query that expects that row
1087 to be an item, you should force the master to handle reads. Otherwise, due to
1088 the lag, there is no certainty your data will be in the expected state.
1090 For data integrity, all transactions automatically use the master storage for
1091 all read and write queries. Using a transaction is the preferred and recommended
1092 method to force the master to handle all read queries.
1094 Otherwise, you can force a single query to use the master with the 'force_pool'
1097 my $result = $resultset->search(undef, {force_pool=>'master'})->find($pk);
1099 This attribute will safely be ignored by non replicated storages, so you can use
1100 the same code for both types of systems.
1102 Lastly, you can use the L</execute_reliably> method, which works very much like
1105 For debugging, you can turn replication on/off with the methods L</set_reliable_storage>
1106 and L</set_balanced_storage>, however this operates at a global level and is not
1107 suitable if you have a shared Schema object being used by multiple processes,
1108 such as on a web application server. You can get around this limitation by
1109 using the Schema clone method.
1111 my $new_schema = $schema->clone;
1112 $new_schema->set_reliable_storage;
1114 ## $new_schema will use only the Master storage for all reads/writes while
1115 ## the $schema object will use replicated storage.
1119 John Napiorkowski <john.napiorkowski@takkle.com>
1121 Based on code originated by:
1123 Norbert Csongrádi <bert@cpan.org>
1124 Peter Siklósi <einon@einon.hu>
1128 You may distribute this code under the same terms as Perl itself.
1132 __PACKAGE__->meta->make_immutable;