The London Perl and Raku Workshop takes place on 26th Oct 2024. If your company depends on Perl, please consider sponsoring and/or attending.


KiokuDB - Object Graph storage engine


If you're new to KiokuDB check out KiokuDB::Tutorial.


    use KiokuDB;

    # use a DSN
    my $d = KiokuDB->connect( $dsn, %args );

    # or manually instantiate a backend
    my $d = KiokuDB->new(
        backend => KiokuDB::Backend::Files->new(
            dir        => "/tmp/foo",
            serializer => "yaml",

    # create a scope object
    my $s = $d->new_scope;

    # takes a snapshot of $some_object
    my $uuid = $d->store($some_object);

    # or with a custom ID:
    $d->store( $id => $some_object ); # $id can be any string

    # retrieve by ID
    my $some_object = $d->lookup($uuid);

    # some backends (like DBI) support simple searches
    $d->search({ name => "foo" });

    # others use GIN queries (DBI supports both)


KiokuDB is a Moose based frontend to various data stores, somewhere in between Tangram and Pixie.

Its purpose is to provide persistence for "regular" objects with as little effort as possible, without sacrificing control over how persistence is actually done, especially for harder to serialize objects.

KiokuDB is also non-invasive: it does not use ties, AUTOLOAD, proxy objects, sv_magic or any other type of trickery.

Many features important for proper Perl space semantics are supported, including shared data, circular structures, weak references, tied structures, etc.

KiokuDB is meant to solve two related persistence problems:

Transparent persistence

Store arbitrary objects without changing their class definitions or worrying about schema details, and without needing to conform to the limitations of a relational model.


Persisting arbitrary objects in a way that is compatible with existing data/code (for example interoprating with another app using CouchDB with JSPON semantics).


In order to use any persistence framework it is important to understand what it does and how it does it.

Systems like Tangram or DBIx::Class generally require explicit meta data and use a schema, which makes them fairly predictable.

When using transparent systems like KiokuDB or Pixie it is more important to understand what's going on behind the scenes in order to avoid surprises and limitations.

An architectural overview is available on the website:

The process is explained here and in the various component documentation in more detail.


When an object is stored using KiokuDB it's collapsed into an KiokuDB::Entry.

An entry is a simplified representation of the object, allowing the data to be saved in formats as simple as JSON.

References to other objects are converted to symbolic references in the entry, so objects can be saved independently of each other.

The entries are given to the KiokuDB::Backend for actual storage.

Collapsing is explained in detail in KiokuDB::Collapser. The way an entry is created varies with the object's class.


When objects are loaded, entries are retrieved from the backend using their UIDs.

When a UID is already loaded (in the live object set of a KiokuDB instance, see KiokuDB::LiveObjects) the live object is used. This way references to shared objects are shared in memory regardless of the order the objects were stored or loaded.

This process is explained in detail in KiokuDB::Linker.


Any object that is passed to store or insert directly is implicitly considered a member of the root set.

This flag implies that the object is an identified resource and should not be garbage collected with any of the proposed garbage collection schemes.

The root flag may be modified explicitly:

    $kiokudb->set_root(@objects); # or unset_root


Lastly, root set membership may also be specified explicitly by the typemap.

A root set member must be explicitly removed using delete or by removing it from the root set. Only non-members of the root set will be purged with any garbage collection scheme.


On supporting backends the txn_do method will execute a block and commit the transaction at its end.

Nesting of txn_do blocks is always supported, though rolling back a nested transaction may produce different results on different backends.

If the backend does not support transactions txn_do simply executes the code block normally.


Most transactional backends are also concurrent.

KiokuDB::Backend::BDB and KiokuDB::Backend::CouchDB default to serializable transaction isolation and do not suffer from deadlocks, but serialization errors may occur, aborting the transaction (in which case the transaction should be tried again).

KiokuDB::Backend::Files provides good concurrency support but will only detect deadlocks on platforms which return EDEADLK from flock. Directory::Transactional may provide alternative mechanisms in the future.

Concurrency support in KiokuDB::Backend::DBI depends on the database. SQLite defaults to serializable transaction isolation out of the box, wheras MySQL and PostgreSQL default to read committed.

Depending on your application read committed isolation may be sufficient, but due to the graph structure nature of the data repeatable reads or serializable level isolation is highly reccomended. Read committed isolation generally works well when each row in the database is more or less independent of others, and various constraints ensure integrity. Unfortunately this is not the case with the graph layout.

To enable stronger isolation guarantees see "Transactions" in KiokuDB::Backend::DBI for per-database pointers.


KiokuDB uses a number of delegates which do the actual work.

Of these only backend is required, the rest have default definitions.

Additional attributes that are not commonly used are listed in "INTERNAL ATTRIBUTES".


This attribute is required.

This must be an object that does KiokuDB::Backend.

The backend handles storage and retrieval of entries.


This is an instance KiokuDB::TypeMap.

The typemap contains entries which control how KiokuDB::Collapser and KiokuDB::Linker handle different types of objects.


An array references of extra classes to allow.

Objects blessed into these classes will be collapsed using KiokuDB::TypeMap::Entry:Naive.


An array references of extra base classes to allow.

Objects derived from these classes will be collapsed using KiokuDB::TypeMap::Entry:Naive.


If true adds KiokuDB::TypeMap::ClassBuilders to the merged typemap.

It's possible to provide a hash reference of options to give to "new" in KiokuDB::TypeMap::ClassBuilders.


Controls whether or not the class versions of objects are checked on load.

Defaults to true.


A table of classes and versions that is passed to the default typemap entry for Moose/Class::MOP objects.

When a class version has changed between the time that an object was stored and the time it's being retrieved, the data must be converted.

See KiokuDB::TypeMap::Entry::MOP for more details.


connect $dsn, %args

DWIM wrapper for new.

$dsn represents some sort of backend (much like DBI dsns map to DBDs).

An example DSN is:

    my $dir = KiokuDB->connect("bdb:dir=path/to/data/");

The backend moniker name is extracted by splitting on the colon. The rest of the string is passed to new_from_dsn, which is documented in more detail in KiokuDB::Backend.

Typically DSN arguments are separated by ;, with = separating keys and values. Arguments with no value are assumed to denote boolean truth (e.g. jspon:dir=foo;pretty means dir => "foo", pretty => 1). However, a backend may override the default parsing, so this is not guaranteed.

Extra arguments are passed both to the backend constructor, and the KiokuDB constructor.

Note that if you need a typemap you still need to pass it in:

    KiokuDB->connect( $dsn, typemap => $typemap );

The DSN can also be a valid JSON string taking one of the following forms:

    dsn => '["dbi:SQLite:foo",{"schema":"MyApp::DB"}]'

    dsn => '{"dsn":"dbi:SQLite:foo","schema":"MyApp::DB"}'

This allows more complicated arguments to be specified accurately, or arbitrary options to be specified when the backend has nonstandard DSN parsing (for instance KiokuDB::Backend::DBI simply passes the string to DBI, so this is necessary in order to specify options on the command line).

configure $config_file, %args


new %args

Creates a new directory object.



Creates a new object scope. Handled by live_objects.

The object scope artificially bumps up the reference count of objects to ensure that they live at least as long as the scope does.

This ensures that weak references aren't deleted prematurely, and the object graph doesn't get corrupted without needing to create circular structures and cleaning up leaks manually.

lookup @ids

Fetches the objects for the specified IDs from the live object set or from storage.

store @objects
store %objects
store_nonroot @objects
store_nonroot %objects

Recursively collapses @objects and inserts or updates the entries.

This performs a full update of every reachable object from @objects, snapshotting everything.

Strings found in the object list are assumed to be IDs for the following objects.

The nonroot variant will not mark the objects as members of the root set (therefore they will be subject to garbage collection).

update @objects

Performs a shallow update of @objects (referants are not updated).

It is an error to update an object not in the database.

deep_update @objects

Update @objects and all of the objects they reference. All references objects must already be in the database.

insert @objects
insert %objects
insert_nonroot @objects
insert_nonroot %objects

Inserts objects to the database.

It is an error to insert objects that are already in the database, all elements of @objects must be new, but their referants don't have to be.

@objects will be collapsed recursively, but the collapsing stops at known objects, which will not be updated.

The nonroot variant will not mark the objects as members of the root set (therefore they will be subject to garbage collection).

delete @objects_or_ids

Deletes the specified objects from the store.

Note that this can cause lookup errors if the object you are deleting is referred to by another object, because that link will be broken.

set_root @objects
unset_root @objects

Modify the root flag on the associated entries.

update must be called for the change to take effect.

txn_do $code, %args
txn_do %args
scoped_txn $code

Executes $code within the scope of a transaction.

This requires that the backend supports transactions (KiokuDB::Backend::Role::TXN).

If the backend does not support transactions, the code block will simply be invoked.

Transactions may be nested.

If the scope argument is true an implicit call to new_scope will be made, keeping the scope for the duration of the transaction.

The return value is propagated from the code block, with handling of list/scalar/void context.

scoped_txn is like txn_do but sets scope to true.


These methods simply call the corresponding methods on the backend.

Like txn_do these methods are no-ops if the backend does not support transactions.

search \%proto
search @args

Searching requires a backend that supports querying.

The \%proto form is currently unspecified but in the future should provide a simple but consistent way of looking up objects by attributes.

The second form is backend specific querying, for instance Search::GIN::Query objects passed to KiokuDB::Backend::BDB::GIN or the generic GIN backend wrapper KiokuDB::GIN.

Returns a Data::Stream::Bulk of the results.


Returns a Data::Stream::Bulk of all the root objects in the database.


Returns a Data::Stream::Bulk of all the objects in the database.

grep $filter

Returns a Data::Stream::Bulk of the objects in root_set filtered by $filter.

scan $callback

Iterates the root set calling $callback for each object.


Delegates to KiokuDB::LiveObjects


Returns $self.

This is used when setting up KiokuDB::Role::API delegation chains. Calling directory on any level of delegator will always return the real KiokuDB instance no matter how deep.



If set at compile time, the default UUID generation role will use serial IDs, instead of UUIDs.

This is useful for testing, since the same IDs will be issued each run, but is utterly broken in the face of concurrency.


These attributes are documented for completeness and should typically not be needed.



The collapser prepares objects for storage, by creating KiokuDB::Entry objects to pass to the backend.



The linker links entries into functioning instances, loading necessary dependencies from the backend.



The live object set keeps track of objects and entries for the linker and the resolver.

It also creates scope objects that help ensure objects don't garbage collect too early ("new_scope" in KiokuDB::LiveObjects, KiokuDB::LiveObjects::Scope), and transaction scope objects used by txn_do (KiokuDB::LiveObjects::TXNScope).


An instance of KiokuDB::TypeMap::Resolver. Handles actual lookup and compilation of typemap entries, using the user typemap.


Prior Art on the CPAN


Polymorphic retrieval is possible with DBIx::Class::DynamicSubclass



KiokuDB is maintained using Git. Information about the repository is available on


Yuval Kogman <>


    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.