KiokuDB::Backend::DBI - DBI backend for KiokuDB

    my $dir = KiokuDB->connect(
        "dbi:mysql:foo",
        user     => "blah",
        password => "moo',
        columns  => [
            # specify extra columns for the 'entries' table
            # in the same format you pass to DBIC's add_columns

            name => {
                data_type => "varchar",
                is_nullable => 1, # probably important
            },
        ],
    );

    $dir->search({ name => "foo" }); # SQL::Abstract

This backend for KiokuDB leverages existing DBI accessible databases.

The schema is based on two tables, entries and gin_index (the latter is only used if a Search::GIN extractor is specified).

The entries table has two main columns, id and data (currently in JSPON format, in the future the format will be pluggable), and additional user specified columns.

The user specified columns are extracted from inserted objects using a callback (or just copied for simple scalars), allowing SQL where clauses to be used for searching.

The columns are specified using a DBIx::Class::ResultSource instance.

One additional column info parameter is used, extract, which is called as a method on the inserted object with the column name as the only argument. The return value from this callback will be used to populate the column.

If the column extractor is omitted then the column will contain a copy of the entry data key by the same name, if it is a plain scalar. Otherwise the column will be NULL.

These columns are only used for lookup purposes, only data is consulted when loading entries.

This backend is layered on top of DBIx::Class::Storage::DBI and reused DBIx::Class::Schema for DDL.

Because of this objects from a DBIx::Class::Schema can refer to objects in the KiokuDB entries table, and vice versa.

For more details see DBIx::Class::Schema::KiokuDB.

This driver has been tested with MySQL 5 (4.1 should be the minimal supported version), SQLite 3, and PostgresSQL 8.3.

The SQL code is reasonably portable and should work with most databases. Binary column support is required when using the Storable serializer.

For reasons of performance and ease of use database vendors ship with read committed transaction isolation by default.

This means that read locks are not acquired when data is fetched from the database, allowing it to be updated by another writer. If the current transaction then updates the value it will be silently overwritten.

IMHO this is a much bigger problem when the data is unstructured. This is because data is loaded and fetched in potentially smaller chunks, increasing the risk of phantom reads.

Unfortunately enabling truly isolated transaction semantics means that txn_commit may fail due to a lock contention, forcing you to repeat your transaction. Arguably this is more correct "read comitted", which can lead to race conditions.

Enabling repeatable read or serializable transaction isolation prevents transactions from interfering with eachother, by ensuring all data reads are performed with a shared lock.

For more information on isolation see http://en.wikipedia.org/wiki/Isolation_(computer_science)

SQLite provides serializable isolation by default.

http://www.sqlite.org/pragma.html#pragma_read_uncommitted

MySQL provides read committed isolation by default.

Serializable level isolation can be enabled by by default by changing the transaction-isolation global variable,

http://dev.mysql.com/doc/refman/5.1/en/set-transaction.html#isolevel_serializable

PostgreSQL provides read committed isolation by default.

Repeatable read or serializable isolation can be enabled by setting the default transaction isolation level, or using the SET TRANSACTION SQL statement.

http://www.postgresql.org/docs/8.3/interactive/transaction-iso.html, http://www.postgresql.org/docs/8.3/interactive/runtime-config-client.html#GUC-DEFAULT-TRANSACTION-ISOLATION

schema

Created automatically.

This is DBIx::Class::Schema object that is used for schema deployment, connectivity and transaction handling.

connect_info

An array reference whose contents are passed to connect in DBIx::Class::Schema.

If omitted will be created from the attrs dsn, user, password and dbi_attrs.

dsn

user

password

dbi_attrs

Convenience attrs for connecting using connect in KiokuDB.

User in connect_info's builder.

columns

Additional columns, see "COLUMN EXTRACTIONS".

serializer

KiokuDB::Serializer. Coerces from a string, too:

    KiokuDB->connect("dbi:...", serializer => "storable");

Defaults to KiokuDB::Serializer::JSON.

create

If true the existence of the tables will be checked for and the DB will be deployed if not.

Defaults to false.

extract

An optional Search::GIN::Extract used to create the gin_index entries.

Usually Search::GIN::Extract::Callback.

schema_hook

A hook that is called on the backend object as a method with the schema as the argument just before connecting.

If you need to modify the schema in some way (adding indexes or constraints) this is where it should be done.

for_update

If true (the defaults), will cause all select statement to be issued with a FOR UPDATE modifier on MySQL, Postgres and Oracle.

This is highly reccomended because these database provide low isolation guarantees as configured out the box, and highly interlinked graph databases are much more susceptible to corruption because of lack of transcational isolation than normalized relational databases.

sqlite_sync_mode

If this attribute is set and the underlying database is SQLite, then PRAGMA syncrhonous=... will be issued with this value.

Can be OFF, NORMAL or FULL (SQLite's default), or 0, 1, or 2.

See http://www.sqlite.org/pragma.html#pragma_synchronous.

mysql_strict

If true (the default), sets MySQL's strict mode.

This is HIGHLY reccomended, or you may enjoy some of MySQL's more interesting features, like automatic data loss when the columns are too narrow.

See http://dev.mysql.com/doc/refman/5.0/en/server-sql-mode.html and DBIx::Class::Storage::DBI::mysql for more details.

on_connect_call

See DBIx::Class::Storage::DBI.

This attribute is constructed based on the values of mysql_version and sqlite_sync_mode, but may be overridden if you need more control.

dbic_attrs

See DBIx::Class::Storage::DBI.

Defaults to

    { on_connect_call => $self->on_connect_call }

batch_size

SQL that deals with entries run in batches of the amount provided in batch_size. If it is not provided, the statements will run in a single batch.

This solves the issue with SQLite where lists can only handle 999 elements at a time. batch_size will be set to 999 by default if the driver in use is SQLite.

See KiokuDB::Backend and the various roles for more info.

deploy

Calls deploy in DBIx::Class::Schema.

Deployment to MySQL requires that you specify something like:

    $dir->backend->deploy({ producer_args => { mysql_version => 4 } });

because MySQL versions before 4 did not have support for boolean types, and the schema emitted by SQL::Translator will not work with the queries used.

drop_tables

Drops the entries and gin_index tables.

You are problably using MySQL, which comes with a helpful data compression feature: when your serialized objects are larger than the maximum size of a BLOB column MySQL will simply shorten it for you.

Why BLOB defaults to 64k, and how on earth someone would consider silent data truncation a sane default I could never fathom, but nevertheless MySQL does allow you to disable this by setting the "strict" SQL mode in the configuration.

To resolve the actual problem (though this obviously won't repair your lost data), alter the entries table so that the data column uses the nonstandard LONGBLOB datatype.

http://github.com/nothingmuch/kiokudb-backend-dbi

Yuval Kogman <nothingmuch@woobling.org>

    Copyright (c) 2008, 2009 Yuval Kogman, Infinity Interactive. All
    rights reserved This program is free software; you can redistribute
    it and/or modify it under the same terms as Perl itself.