Cayenne Guide

A Cayenne project is an XML representation of a model connecting database schema with Java classes. A project is normally created and manipulated via CayenneModeler GUI and then used to initialize Cayenne runtime. A project is made of one or more files. There’s always a root project descriptor file in any valid project. It is normally called cayenne-xyz.xml, where "xyz" is the name of the project.

Project descriptor can reference DataMap files, one per DataMap. DataMap files are normally called xyz.map.xml, where "xyz" is the name of the DataMap. For legacy reasons this naming convention is different from the convention for the root project descriptor above, and we may align it in the future versions. Here is how a typical project might look on the file system:

DataMap are referenced by name in the root descriptor:

Map files are resolved by Cayenne by appending ".map.xml" extension to the map name, and resolving the resulting string relative to the root descriptor URI. The following sections discuss various ORM model objects, without regards to their XML representation. XML format details are really unimportant to the Cayenne users.

DataMap is a container of persistent entities and other object-relational metadata. DataMap provides developers with a scope to organize their entities, but it does not provide a namespace for entities. In fact all DataMaps present in runtime are combined in a single namespace. Each DataMap must be associated with a DataNode. This is how Cayenne knows which database to use when running a query.

DataNode is a model of a database. It is actually pretty simple. It has a name and a JDBC DataSource. Unlike the rest of the objects discussed here, DataNodes are not a part of the XML mapping project. They are defined in the code, when Cayenne runtime is created (see Starting Cayenne). Most applications only have one DataNode, though there may be any number of nodes if needed.

DbEntity is a model of a single DB table or view. DbEntity is made of DbAttributes that correspond to columns, and DbRelationships that map PK/FK pairs. DbRelationships are not strictly tied to FK constraints in DB, and should be mapped for all logical "relationships" between the tables.

ObjEntity is a model of a single persistent Java class. ObjEntity is made of ObjAttributes and ObjRelationships. Both correspond to entity class properties. However ObjAttributes represent "simple" properties (normally things like String, numbers, dates, etc.), while ObjRelationships correspond to properties that have a type of another entity.

ObjEntity maps to one or more DbEntities. There’s always one "root" DbEntity for each ObjEntity. ObjAttribiute maps to a DbAttribute or an Embeddable. Most often mapped DbAttribute is from the root DbEntity. Sometimes mapping is done to a DbAttribute from another DbEntity somehow related to the root DbEntity. Such ObjAttribute is called "flattened". Similarly ObjRelationship maps either to a single DbRelationship, or to a chain of DbRelationships ("flattened" ObjRelationship).

ObjEntities may also contain mapping of their lifecycle callback methods.

Embeddable is a model of a Java class that acts as a single attribute of an ObjEntity, but maps to multiple columns in the database.

A model of a stored procedure in the database.

A model of a query. Cayenne allows queries to be mapped in Cayenne project, or created in the code. Depending on the circumstances the users may take one or the other approach.

CayenneModeler GUI tool is intended to work with object relational mapping projects. While you can edit your XML by hand, it is rarely needed, as the Modeler is a pretty advanced tool included in Cayenne distribution. To obtain CayenneModeler, download Cayenne distribution archive from https://cayenne.apache.org/download.html matching the OS you are using. Of course Java needs to be installed on the machine where you are going to run the Modeler.

A path to an existing Cayenne project file can be passed as the only argument to open the project on startup:

On macOS the same applies to the .app bundle:

Alternative approach to using build tools is doing reverse engineering from CayenneModeler.

You can find the reverse engineering tool on the DB Import tab of a DataMap.

Before using Cayenne in you code you need to generate java source code for persistent objects. This can be done with Modeler GUI or via cgen maven/ant plugin.

To generate classes in the Modeler select a DataMap, go to the Class Generation tab and click Generate.

The default settings are suitable for almost all cases. Change them only if you know what exactly you need to customize. The "Advanced options" and "Templates options" on the same tab let you control almost all aspects of code generation, including custom templates for Java code. See default Cayenne templates on GitHub as an example.

Normally each ObjEntity is mapped to a specific Java class (such as Artist or Painting) that explicitly declare all entity properties as pairs of getters and setters. However, Cayenne allows to map a completely generic class to any number of entities. The only expectation is that a generic class implements org.apache.cayenne.Persistent. So an ideal candidate for a generic class is GenericPersistentObject, or some custom subclass of GenericPersistentObject.

If you don’t enter anything for Java Class of an ObjEntity, Cayenne assumes generic mapping and uses the following implicit rules to determine a class of a generic object. If DataMap "Custom Superclass" is set, runtime uses this class to instantiate new objects. If not, org.apache.cayenne.GenericPersistentObject is used.

Class generation procedures (either done in the Modeler or with Ant or Maven) would skip entities that are mapped to GenericPersistentObject explicitly or have no class mapping.

Cayenne supports three PK generation strategies:

Strategy should be set per each DbEntity independently.

To add Cayenne to your Maven project, include cayenne in your POM:

To add Cayenne to your Gradle project, include cayenne module:

If your environment requires manual dependency management (like Ant), check lib and lib/third-party folders of Cayenne distribution. It contains all Cayenne jars as well as the minimal set of third-party libraries to get you started.

In runtime Cayenne is accessed via org.apache.cayenne.CayenneRuntime. CayenneRuntime is created by calling a convenient builder:

The parameter you pass to the builder is a location of the main project file. Location is a '/'-separated path (same path separator is used on UNIX and Windows) that is resolved relative to the application classpath. The project file can be placed in the root package or in a subpackage (e.g. in the code above it is in "com/example" subpackage).

CayenneRuntime encapsulates a single Cayenne stack. Most applications will just have one CayenneRuntime using it to create as many ObjectContexts as needed, access the Dependency Injection (DI) container and work with other Cayenne features. Internally CayenneRuntime is just a thin wrapper around the DI container. Detailed features of the container are discussed in Customizing Cayenne Runtime chapter. Here we’ll just show an example of how an application might turn on external transactions:

It is a good idea to shut down the runtime when it is no longer needed, usually before the application itself is shutdown:

When a runtime object has the same scope as the application, this may not be always necessary, however in some cases it is essential, and is generally considered a good practice. E.g. in a web container hot redeploy of a webapp will cause resource leaks and eventual OutOfMemoryError if the application fails to shutdown CayenneRuntime.

The mapping project does not contain any database connection information. The database is defined in the code, when the runtime is created, as a DataNode. In the simplest case a DataNode is created from a DataSource. It can be any DataSource, e.g., the one provided by your application container, or the one built with CayenneDataSource:

When a DataNode needs to be customized, e.g., to let Cayenne create the DB schema on the first connection, to set a specific DbAdapter, or to give the node a name, use a DataNodeDescriptor:

There can be more than one DataNode. Each one is linked to its DataMaps by name. The "default" DataNode is optional, and is used for all the DataMaps that are not linked to a node explicitly:

CayenneRuntime requires at least one mapping project to run. But it can also take multiple projects and merge them together in a single configuration. This way different parts of a database can be mapped independently from each other (even by different software providers), and combined in runtime when assembling an application. Doing it is as easy as passing multiple project locations to CayenneRuntime builder:

When the projects are merged, the following rules are applied:

ObjectContext is an interface that users normally work with to access the database. It provides the API to execute database operations and to manage persistent objects. A context is obtained from the CayenneRuntime:

The call above creates a new instance of ObjectContext that can access the database via this runtime. ObjectContext is a single "work area" in Cayenne, storing persistent objects. ObjectContext guarantees that, for each database row with a unique ID, it will contain at most one instance of an object, thus ensuring object graph consistency between multiple selects (a feature called "uniquing"). At the same time, different ObjectContexts will have independent copies of objects for each unique database row. This allows users to isolate object changes from one another by using separate ObjectContexts.

These properties directly affect the strategies for scoping and sharing (or not sharing) ObjectContexts. Contexts that are only used to fetch objects from the database and whose objects are never modified by the application can be shared between multiple users (and multiple threads). Contexts that store modified objects should be accessed only by a single user (e.g. a web application user might reuse a context instance between multiple web requests in the same HttpSession, thus carrying uncommitted changes to objects from request to request, until they decide to commit them or roll them back). Even for a single user it might make sense to use multiple ObjectContexts (e.g. request-scoped contexts to allow concurrent requests from the browser that change and commit objects independently).

ObjectContext does not permanently hold any of the application resources. So it does not have to be closed. If the context is not used anymore, it should simply be allowed to go out of scope and get garbage collected, just like any other Java object.

Cayenne can persist Java objects that implement the org.apache.cayenne.Persistent interface. Generally, persistent classes are generated from the model as described above, so users do not have to worry about superclass and property implementation details.

The Persistent interface provides access to three persistence-related properties - objectId, persistenceState and objectContext. All three are initialized by the Cayenne runtime framework. Your application code should not attempt to change them. However, it is allowed to read them, which provides valuable runtime information. E.g. ObjectId can be used for a quick equality check of two objects, knowing persistence state would allow highlighting changed objects, etc.

Each persistent object belongs to a single ObjectContext, and can be in one of the following persistence states (as defined in org.apache.cayenne.PersistenceState) :

One of the first things users usually want to do with an ObjectContext is to select some objects from a database:

We’ll discuss queries in some detail in the Queries chapter. The example above is self-explanatory - we create a ObjectSelect that matches all Artist objects present in the database, and then use select to get the result.

Some queries can be quite complex, returning multiple result sets or even updating the database. For such queries, ObjectContext provides the execute() method. While not commonly used, it is nevertheless important in some situations. E.g.:

When multiple such queries need to run together, execute them inside an explicit transaction:

An application might modify selected objects. E.g.:

The first time the object property is changed, the object’s state is automatically set to MODIFIED by Cayenne. Cayenne tracks all in-memory changes until a user calls commitChanges:

At this point, all in-memory changes are analyzed and a minimal set of SQL statements is issued in a single transaction to synchronize the database with the in-memory state. In our example, commitChanges commits just one object, but generally it can be any number of objects.

If, instead of commit, we wanted to reset all changed objects to the previously committed state, we’d call rollbackChanges instead:

newObject method call creates a persistent object and sets its state to NEW:

It only exists in memory until commitChanges is issued. On commit Cayenne might generate a new primary key (unless a user set it explicitly, or a PK was inferred from a relationship) and issue an INSERT SQL statement to permanently store the object.

The deleteObjects method takes one or more Persistent objects and marks them as DELETED:

Additionally, deleteObjects processes all delete rules modeled for the affected objects. This may result in implicitly deleting or modifying extra related objects. Same as insert and update, delete operations are sent to the database only when commitChanges is called. Similarly rollbackChanges will undo the effect of newObject and deleteObjects.

localObject returns a copy of a given persistent object that is local to a given ObjectContext:

Since an application often works with more than one context, localObject is a rather common operation. E.g. to improve performance, a user might utilize a single shared context to select and cache data, and then occasionally transfer some selected objects to another context to modify and commit them:

Often an application needs to inspect mapping metadata. This information is stored in the EntityResolver object, accessible via the ObjectContext:

Here we discussed the most commonly-used subset of the ObjectContext API. There are other useful methods, e.g. those allowing you to inspect registered objects' state in bulk, etc. Check the latest JavaDocs for details.

There is a useful helper class called Cayenne (fully-qualified name org.apache.cayenne.Cayenne) that builds on the ObjectContext API to provide a number of very common operations. E.g. get a primary key (most entities do not model PK as an object property) :

The reverse operation - finding an object given a known PK - is a part of the ObjectContext API. It returns an object already registered in the context without a query, and consults the caches before going to the database:

For more flexibility, you could select by ID with ObjectSelect instead.

Feel free to explore the Cayenne class API for other useful methods.

In all the examples shown so far, an ObjectContext would directly connect to a database to select data or synchronize its state (either via commit or rollback). However, another context can be used in all these scenarios instead of a database. This concept is called ObjectContext "nesting". Nesting is a parent/child relationship between two contexts, where a child is a nested context and selects or commits its objects via a parent.

Nesting is useful to create isolated object editing areas (child contexts) that all need to be committed to an intermediate in-memory store (parent context), or rolled back without affecting changes already recorded in the parent. Think cascading GUI dialogs, or parallel AJAX requests coming to the same session.

In theory, Cayenne supports any number of nesting levels; however, applications should generally stay with one or two levels, as deep hierarchies will almost certainly degrade the performance of the deeply-nested child contexts. This is due to the fact that each context in a nesting chain has to update its own objects during most operations.

To create a nested context, use an instance of CayenneRuntime, passing it the desired parent:

From here, a nested context operates just like a regular context (you can perform queries, create and delete objects, etc…​). The only difference is that commit and rollback operations can either be limited to synchronization with the parent, or cascade all the way to the database:

As described in the CayenneModeler chapter, Cayenne supports mapping of completely generic classes to specific entities. Although for convenience most applications should stick with entity-specific class mappings, the generic feature offers some interesting possibilities, such as creating mappings completely on the fly in a running application.

Generic objects are first-class citizens in Cayenne, and all common persistent operations apply to them as well. There are some peculiarities, however, described below.

When creating a generic object, either use ObjectContext.newObject(String), that takes an entity name, or provide your object with an explicit ObjectId:

ObjectSelect for a generic object should be created by passing the entity name String, instead of just a Java class:

Use Persistent API to access and modify properties of a generic object:

This is how an application can obtain the entity name of a generic object:

Considering how much attention is given to managing transactions in most other ORMs, transactions have been conspicuously absent from the ObjectContext discussion till now. The reason is that transactions are seamless in Cayenne in all but a few special cases. ObjectContext is an in-memory container of objects that is disconnected from the database, except when it needs to run an operation. So it does not care about any surrounding transaction scope. Sure enough, all database operations are transactional, so when an application does a commit, all SQL execution is wrapped in a database transaction. But this is done behind the scenes and is rarely a concern to the application code.

Two cases where transactions need to be taken into consideration are container- and application-managed transactions.

If you are using Spring, EJB or another environment that manages transactions, you’ll likely need to switch the Cayenne runtime into "external transactions mode". This is done by setting the DI configuration property defined in RuntimeProperties.EXTERNAL_TX_PROPERTY (see Appendix A). If this property is set to "true", Cayenne assumes that JDBC Connections obtained by runtime, whenever that might happen, are all coming from a transactional DataSource managed by the container. In this case, Cayenne does not attempt to commit or roll back the connections, leaving it up to the container to do that when appropriate.

In the second scenario, an application might need to define its own transaction scope that spans more than one Cayenne operation. E.g. two sequential commits that need to be rolled back together in case of failure. This can be done via the CayenneRuntime.performInTransaction method:

When inside a transaction, current thread Transaction object can be accessed via a static method:

You can control transaction isolation level and propagation logic using TransactionDescriptor.

Before discussing how to build expressions, it is important to understand one group of expressions widely used in Cayenne - path expressions. There are two types of path expressions - object and database, used for navigating graphs of connected objects or joined DB tables, respectively. Object paths are much more commonly used, as, after all, Cayenne is supposed to provide a degree of isolation of the object model from the database. However, database paths are helpful in certain situations. The general structure of path expressions is the following:

An object path expression represents a chain of property names rooted in a certain (unspecified during expression creation) object and "navigating" to its related value. E.g. a path expression "artist.name" might be a property path starting from a Painting object, pointing to the related Artist object, and then to its name attribute. A few more examples:

Similarly a database path expression is a dot-separated path through DB table joins and columns. In Cayenne joins are mapped as DbRelationships with some symbolic names (the closest concept to DbRelationship name in the DB world is a named foreign key constraint. But DbRelationship names are usually chosen arbitrarily, without regard to constraints naming or even constraints presence). A database path therefore might look like this - db:dbrelationshipX.dbrelationshipY.COLUMN_Z". More specific examples:

Cayenne supports "aliases" in path expressions. An alias is appended to a relationship segment after the # character:

The main use of aliases is to allow users to control how SQL joins are generated if the same path is encountered more than once in any given Expression. Each alias for any given path would result in a separate join. Without aliases, a single join will be used for a group of matching paths.

While in most cases users are likely to rely on the API from the following section for expression creation, we’ll start by showing String expressions, as this will help you understand the semantics. A Cayenne expression can be represented as a String, which can be converted to an expression object using the ExpressionFactory.exp static method. Here is an example:

This particular expression may be used to match Paintings whose names start with "A" and whose price is less than $1000. While this example is pretty self-explanatory, there are a few points worth mentioning. "name" and "price" here are object paths discussed earlier. As always, paths themselves are not attached to a specific root entity and can be applied to any entity that has similarly named attributes or relationships. So, when we say that this expression "may be used to match Paintings", we are implying that there may be other entities for which this expression is valid. Now the expression details…​

Character constants that are not paths or numeric values should be enclosed in single or double quotes. Two of the expressions below are equivalent:

Case sensitivity. Expression operators are case sensitive and are usually lowercase. Complex words follow the Java camel-case style:

Grouping with parenthesis:

Path prefixes. Object expressions are unquoted strings, optionally prefixed by obj: (usually they are not prefixed at all). Database expressions are always prefixed with db:. A special kind of prefix, not discussed yet, is enum: that prefixes an enumeration constant:

Binary conditions are expressions that contain a path on the left, a value on the right, and some operation between them, such as equals like, etc. They can be used as qualifiers in SelectQueries:

Parameters. Expressions can contain named parameters (names that start with "$") that can be substituted with values either by name or by position. Parameterized expressions let you create reusable expression templates. Also, if an expression contains a complex object that doesn’t have a simple String representation (e.g. a Date, a Persistent, an ObjectId), parameterizing the expression is the only way to represent it as String. Here are examples of both positional and named parameter bindings:

Positional binding is usually shorter. You can pass positional bindings to the exp(..) factory method (its second argument is a params vararg):

In parameterized expressions with a LIKE clause, SQL wildcards must be part of the values in the Map and not the expression string itself:

When matching on a relationship, the value parameter must be either a Persistent object, an org.apache.cayenne.ObjectId, or a numeric ID value (for single column IDs). E.g.:

When you use positional binding, Cayenne expects values for all parameters to be present. Binding by name offers extra flexibility: sub-expressions with uninitialized parameters are automatically pruned from the expression. So, e.g., if certain parts of the expression criteria are not provided to the application, you can still build a valid expression

Null handling. Handling of Java nulls as operands is no different handling from normal values. Instead of using special conditional operators, like SQL does (IS NULL, IS NOT NULL), "=" and "!=" expressions are used directly with null values. It is up to Cayenne to translate expressions with nulls to the valid SQL.

Creating expressions from Strings is a powerful and dynamic approach, however a safer alternative is to use the Java API. It provides compile-time checking of the expression’s validity. The API in question is provided by the ExpressionFactory class (that we’ve seen already), the Property class and the Expression class itself. ExpressionFactory contains a number of self-explanatory static methods that can be used to build expressions. E.g.:

As discussed earlier, Cayenne supports aliases in path Expressions, so you can control how SQL joins are generated if the same path is encountered more than once in the same Expression. Two ExpressionFactory.matchAllExp(..) methods (one taking a Collection of values, another - a vararg) let you implicitly generate aliases to "split" match paths into individual joins if needed:

The "Path" argument to both of these methods can use a split character (a pipe symbol '|') instead of a dot to indicate that the relationship following a path should be split into a separate set of joins, one per collection value. There can only be one split at most in any given path. The split must always precede a relationship. E.g. "|exhibits.paintings", "exhibits|paintings", etc. Internally, Cayenne generates distinct aliases for each of the split expressions, forcing separate joins.

While ExpressionFactory is pretty powerful, there’s an even easier way to create an expression using static Property objects generated by Cayenne for each persistent class. Some examples:

Property objects provide the API mostly analogous to ExpressionFactory, though it is significantly shorter and is aware of the value types. It provides compile-time checks of both property names and types of arguments in conditions. We will use Property-based API in further examples.

When used in a query, an expression is converted to a SQL WHERE or ORDER BY clause by Cayenne during query execution. Thus the actual evaluation against the data is done by the database engine. However, the same expressions can also be used for accessing object properties, calculating values, and in-memory filtering.

Checking whether an object satisfies an expression:

Reading property value:

Filtering a list of objects:

An ObjectSelect or a ColumnSelect can be created from a String. This is an alternative to the fluent API. E.g.:

The parser produces the same query object that the fluent API would, so everything said about ObjectSelect above applies. The parsed query can be customized further with the fluent API before it is executed:

SQL is very powerful and allows to manipulate data in ways that can not always be described as a graph of related entities. Cayenne acknowledges this fact and provides a facility to execute SQL, sometimes allowing to map results back to persistent objects. SQLSelect and SQLExec are a pair of queries that allow to run native SQL. SQLSelect can be used (as the name suggests) to select custom data in form of entities, separate columns, collection of DataRow or Object[]. SQLExec is designed to execute any SQL (e.g. updates, deletes, DDLs, etc.).

Both queries support advanced SQL templating, with variable substitution and special directives as described in the next chapter. Here we’ll just provide a few simple examples:

And here is an example of how to use SQLExec:

A powerful feature of SQLSelect and SQLExec is that SQL string is treated by Cayenne as a dynamic template. Before creating a PreparedStatement, the String is evaluated, resolving its dynamic parts. The two main scripting elements are "variables" (that look like $var) and "directives" (that look like #directive(p1 p2 p3)). In the discussion below we’ll use both selecting and updating examples, as scripting works the same way for both SQLSelect and SQLExec.

MappedSelect and MappedExec is a queries that are just a reference to another queries stored in the DataMap. The actual stored query can be ObjectSelect, SQLTemplate or ProcedureQuery. Difference between MappedSelect and MappedExec is (as reflected in their names) whether underlying query intended to select data or just to perform some generic SQL code.

An ObjectSelect stored in the DataMap is defined by a String in the Object Query Language, so it can be any query that the language supports, including the column queries and the queries rooted in a DbEntity. Its "$name" parameters are bound by name when the query is executed. A condition with a parameter that has no value is dropped from the query, so the same mapped query can be used both with and without its optional parameters.

Here is example of how to use MappedSelect:

And here is example of MappedExec:

Stored procedures are mapped as separate objects in CayenneModeler. ProcedureCall provides a way to execute them with a certain set of parameters. This query is a "fluent" version of older ProcedureQuery. Just like with SQLTemplate, the outcome of a procedure can be anything - a single result set, multiple result sets, some data modification (returned as an update count), or a combination of these. So use root class to get a single result set, and use only procedure name for anything else:

QueryResult is a sealed interface with SelectResult, UpdateResult, ResultIterator and OutParametersResult variants, listed in the order the procedure produced them, so the items can be accessed by index or scanned with a pattern-matching switch. A stored procedure can return data back to the application as result sets or via OUT parameters. The values of OUT and INOUT parameters are reported as a separate OutParametersResult item of the result, a map keyed by parameter name:

There maybe a situation when a stored procedure handles its own transactions, but an application is configured to use Cayenne-managed transactions. This is obviously conflicting and undesirable behavior. In this case ProcedureQueries should be executed explicitly wrapped in an "external" Transaction. This is one of the few cases when a user should worry about transactions at all. See Transactions section for more details.

If a user needs some extra functionality not addressed by the existing set of Cayenne queries, he can write his own. The only requirement is to implement org.apache.cayenne.query.Query interface. E.g. to do something directly in the JDBC layer, you might implement Query like this:

To delegate the actual query execution to a standard Cayenne query, you may subclass IndirectQuery:

In fact many internal Cayenne queries are IndirectQueries, delegating to ObjectSelect or SQLTemplate after some preprocessing.

Cayenne defines the following 8 types of lifecycle events for which callbacks can be registered:

Apache Cayenne provides two main methods to set up callbacks for Persistent objects: using annotated methods or configuring them via Cayenne Modeler. As a rule callback methods do not have any knowledge of the outside application, and can only access the state of the object itself and possibly the state of other persistent objects via object’s own ObjectContext.

A listener is simply some application class that has one or more annotated callback methods. A callback method signature should be void someMethod(SomePersistentType object). It can be public, private, protected or use default access:

Notice that the example above contains an annotation on the callback method that defines the type of the event this method should be called for. Before we go into annotation details, we’ll show how to create and register a listener with Cayenne. It is always a user responsibility to register desired application listeners, usually right after CayenneRuntime is started. Here is an example:

First let’s define 2 simple listeners.

Ignore the annotations for a minute. The important point here is that the listeners are arbitrary classes unmapped and unknown to Cayenne, that contain some callback methods. Now let’s register them with runtime:

Listeners in this example are very simple. However they don’t have to be. Unlike Persistent objects, normally listeners initialization is managed by the application code, not Cayenne, so listeners may have knowledge of various application services, operation transactional context, etc. Besides a single listener can apply to multiple entities. As a consequence their callbacks can do more than just access a single ObjectContext.

Now let’s discuss the annotations. There are eight annotations exactly matching the names of eight lifecycle events. A callback method in a listener should be annotated with at least one, but possibly with more than one of them. Annotation itself defines what event the callback should react to. Annotation parameters are essentially an entity filter, defining a subset of ObjEntities whose events we are interested in:

As shown above, "value" (the implicit annotation parameter) can contain one or more entity classes. Only these entities' events will result in callback invocation. There’s also another way to match entities - via custom annotations. This allows to match any number of entities without even knowing what they are. Here is an example. We’ll first define a custom annotation:

Now we can define a listener that will react to events from ObjEntities annotated with this annotation:

As you see we don’t have any entities yet, still we can define a listener that does something useful. Now let’s annotate some entities:

A final touch in the listeners design is preserving the state of the listener within a single select or commit, so that events generated by multiple objects can be collected and processed all together. To do that you will need to implement a DataChannelSyncFilter (and/or DataChannelQueryFilter), and add some callback methods to it. They will store their state in a ThreadLocal variable of the filter. Here is an example filter that does something pretty meaningless - counts how many total objects were committed. However it demonstrates the important pattern of aggregating multiple events and presenting a combined result:

Now since this is both a filter and a listener, it needs to be registered as such:

To receive commit notifications, follow three steps.

To audit only entities marked with @CommitLog:

By default, listeners are called within the database transaction, so a listener can write additional rows as part of the same commit. If your listener performs its own transaction (for example writing to a different database), call excludeCommitLogFromTransaction() to move the notification outside the main transaction:

For full control over which entities are included and how their metadata is computed, provide a custom CommitLogEntityFactory:

Prefetching is a technique that allows to bring back in one query not only the queried objects, but also objects related to them. In other words it is a controlled eager relationship resolving mechanism. Prefetching is discussed in the "Performance Tuning" chapter, as it is a powerful performance optimization method. However another common application of prefetching is to refresh stale object relationships, so more generally it can be viewed as a technique for managing subsets of the object graph.

Prefetching example:

All types of relationships can be prefetched - to-one, to-many, flattened. A prefetch can span multiple relationships:

A query can have multiple prefetches:

If a query is fetching DataRows, all "disjoint" prefetches are ignored, only "joint" prefetches are executed (see prefetching semantics discussion below for what disjoint and joint prefetches mean).

A strategy to prefetch relationships is defined by prefetch "semantics". Depending on semantics, Cayenne would generate different types of queries. The end result is the same - query root objects with related objects fully resolved. However semantics can affect performance, in some cases significantly. There are 3 types of prefetch semantics defined as constants in org.apache.cayenne.query.PrefetchTreeNode:

Disjoint prefetch semantics results in Cayenne generating one SQL statement for the main objects, and a separate statement for each prefetch path (hence "disjoint" - related objects are not fetched with the main query). Each additional SQL statement uses a qualifier of the main query plus a set of joins traversing the prefetch path between the main and related entity.

This strategy has an advantage of efficient JVM memory use, and faster overall result processing by Cayenne, but it requires (1+N) SQL statements to be executed, where N is the number of prefetched relationships.

Disjoint-by-ID prefetch semantics is a variation of disjoint prefetch where related objects are matched against a set of IDs derived from the fetched main objects (or intermediate objects in a multi-step prefetch). Cayenne limits the size of the generated WHERE clause, as most DBs can’t parse arbitrary large SQL. So prefetch queries are broken into smaller queries. The size of is controlled by the DI property RuntimeProperties.MAX_ID_QUALIFIER_SIZE_PROPERTY (the default number of conditions in the generated WHERE clause is 10000). Cayenne will generate (1 + N * M) SQL statements for each query using disjoint-by-ID prefetches, where N is the number of relationships to prefetch, and M is the number of queries for a given prefetch that is dependent on the number of objects in the result (ideally M = 1).

The advantage of this type of prefetch is that matching database rows by ID may be much faster than matching the qualifier of the original query. Moreover this is the only type of prefetch that can handle SelectQueries with fetch limit. Both joint and regular disjoint prefetches may produce invalid results or generate inefficient fetch-the-entire table SQL when fetch limit is in effect.

The disadvantage is that query SQL can get unwieldy for large result sets, as each object will have to have its own condition in the WHERE clause of the generated SQL.

Joint prefetch semantics results in a single SQL statement for root objects and any number of jointly prefetched paths. Cayenne processes in memory a cartesian product of the entities involved, converting it to an object tree. It uses OUTER joins to connect prefetched entities.

Joint is the most efficient prefetch type of the three as far as generated SQL goes. There’s always just 1 SQL query generated. Its downsides are the potentially increased amount of data that needs to get across the network between the application server and the database, and more data processing that needs to be done on the Cayenne side.

ObjectSelect query supports all three types of semantics. You can mix and match them in the same query for different prefetches.

SQLSelect query supports "JOINT" and "DISJOINT_BY_ID". It does not work with "DISJOINT", as the query does not provide enough information to Cayenne to build dependent prefetch queries. So "DISJOINT" will result in exception. "JOINT" prefetching requires a bit of effort shaping the SQL to include the right columns in the result and label them properly to be convertible into object properties. The main rules to follow are:

E.g.:

Converting result set data to Persistent objects and registering these objects in the ObjectContext can be an expensive operation comparable to the time spent running the query (and frequently exceeding it). Internally Cayenne builds the result as a list of DataRows, that are later converted to objects. Skipping the last step and using data in the form of DataRows can significantly increase performance.

DataRow is a simply a map of values keyed by their DB column name. It is a ubiquitous representation of DB data used internally by Cayenne. And it can be quite usable as is in the application in many cases. So performance sensitive selects should consider DataRows - it saves memory and CPU cycles. All selecting queries support DataRows option, e.g.:

Individual DataRows may be converted to Persistent objects as needed. So e.g. you may implement some in-memory filtering, only converting a subset of fetched objects:

While contemporary hardware may easily allow applications to fetch hundreds of thousands or even millions of objects into memory, it doesn’t mean this is always a good idea to do so. You can optimize processing of very large result sets with two techniques discussed in this and the following chapter - iterated and paginated queries.

Iterated query is not actually a special query. Any selecting query can be executed in iterated mode by an ObjectContext. ObjectContext creates an object called ResultIterator that is backed by an open ResultSet. Iterator provides constant memory performance for arbitrarily large ResultSets. This is true at least on the Cayenne end, as JDBC driver may still decide to bring the entire ResultSet into the JVM memory.

Data is read from ResultIterator one row/object at a time until it is exhausted. There are two styles of accessing ResultIterator - direct access which requires explicit closing to avoid JDBC resources leak, or a callback that lets Cayenne handle resource management. In both cases iteration can be performed using "for" loop, as ResultIterator is "Iterable".

Direct access. Here common sense tells us that ResultIterators instances should be processed and closed as soon as possible to release the DB connection. E.g. storing open iterators between HTTP requests for unpredictable length of time would quickly exhaust the connection pool.

Same thing with a callback:

Another example is a batch iterator that allows to process more than one object in each iteration. This is a common scenario in various data processing jobs - read a batch of objects, process them, commit the results, and then repeat. This allows to further optimize processing (e.g. by avoiding frequent commits).

Enabling query pagination allows to load very large result sets in a Java app with very little memory overhead (much smaller than even the DataRows option discussed above). Moreover it is completely transparent to the application - a user gets what appears to be a list of Persistent objects - there’s no iterator to close or DataRows to convert to objects:

Having said that, DataRows option can be combined with pagination, providing the best of both worlds:

The way pagination works internally, it first fetches a list of IDs for the root entity of the query. This is very fast and initially takes very little memory. Then when an object is requested at an arbitrary index in the list, this object and adjacent objects (a "page" of objects that is determined by the query pageSize parameter) are fetched together by ID. Subsequent requests to the objects of this "page" are served from memory.

An obvious limitation of pagination is that if you eventually access all objects in the list, the memory use will end up being the same as with no pagination. However it is still a very useful approach. With some lists (e.g. multi-page search results) only a few top objects are normally accessed. At the same time pagination allows to estimate the full list size without fetching all the objects. And again - it is completely transparent and looks like a normal query.

By default when a single ObjectContext commits its changes, all other contexts in the same runtime receive an event that contains all the committed changes. This allows them to update their cached object state to match the latest committed data. There are however many problems with this ostensibly helpful feature. In short - it works well in environments with few contexts and in unclustered scenarios, such as single user desktop applications, or simple webapps with only a few users. More specifically:

So we’ve made a good case for disabling synchronization in most webapps. To do that, set to "false" the following DI property - RuntimeProperties.CONTEXTS_SYNC_PROPERTY, using one of the standard Cayenne DI approaches. E.g. from command line:

Or by changing the standard properties Map in a custom extensions module:

Cayenne runtime is built around a small powerful dependency injection (DI) container. Just like other popular DI technologies, such as Spring or Guice, Cayenne DI container manages sets of interdependent objects and allows users to configure them. These objects are regular Java objects. We are calling them "services" in this document to distinguish from all other objects that are not configured in the container and are not managed. DI container is responsible for service instantiation, injecting correct dependencies, maintaining service instances scope, and dispatching scope events to services.

The services are configured in special Java classes called "modules". Each module defines binding of service interfaces to implementation instances, implementation types or providers of implementation instances. There are no XML configuration files, and all the bindings are type-safe. The container supports injection into instance variables and constructor parameters based on the @Inject annotation. This mechanism is very close to Google Guice.

The discussion later in this chapter demonstrates a standalone DI container. But keep in mind that Cayenne already has a built-in Injector, and a set of default modules. A Cayenne user would normally only use the API below to write custom extension modules that will be loaded in that existing container when creating CayenneRuntime. See "Starting and Stopping CayenneRuntime" chapter for an example of passing an extension module to Cayenne.

Cayenne DI probably has ~80% of the features expected in a DI container and has no dependency on the rest of Cayenne, so in theory can be used as an application-wide DI engine. But its primary purpose is still to serve Cayenne. Hence there are no plans to expand it beyond Cayenne needs. It is an ideal "embedded" DI that does not interfere with Spring, Guice or any other such framework present elsewhere in the application.

The previous section discussed how Cayenne DI works in general terms. Since Cayenne users will mostly be dealing with an existing Injector provided by CayenneRuntime, it is important to understand how to build custom extensions to a preconfigured container. As shown in "Starting and Stopping CayenneRuntime" chapter, custom extensions are done by writing an application DI module (or multiple modules) that configures service overrides. This section shows all the configuration possibilities in detail, including changing properties of the existing services, contributing services to standard service lists and maps, and overriding service implementations. All the code examples later in this section are assumed to be placed in an application module "configure" method:

AI coding agents such as Claude Code, Cursor, etc. can read and write Cayenne projects like any other Java codebase. For day-to-day API work — writing queries, manipulating an ObjectContext, etc. — you may not need anything Cayenne-specific. Prompt the agent the way you would for any other project. But two additional elements make the agent Cayenne-aware and allow to implement full ORM workflows:

Both are independent but work best together. This chapter walks you through installing both and explains the available MCP tools.

(Claude Code specific) Install the apache-cayenne plugin with Cayenne skills under Claude Code:

The plugin adds skills for DataMap editing, class generation, reverse engineering, query writing, and runtime setup. The skills work best with the MCP server present (see the next chapter).

The MCP server is a self-contained, stdio-based process that the AI agent launches on demand. It works with most agents (not only Claude). The MCP server is bundled with the platform-specific CayenneModeler and requires Java 21 or later on the system PATH. So to get it, first download the Modeler at https://cayenne.apache.org/download/ .

MCP runnable jar is CayenneMCPServer.jar. Its location is OS-dependent (you will need this location in the next step):

Mac:

Windows:

Linux / cross-platform:

When installing the MCP, replace /path/to/CayenneMCPServer.jar mentioned in the instructions with the actual jar path:

Cache invalidation module is an extension that allows to define cache invalidation policy programmatically.

Crypto module allows encrypt and decrypt values stored in DB transparently to your Java app.

Allows to integrate any JCache (JSR 107) compatible caching provider with Cayenne.

Enables usage of full featured Apache Velocity templates in SQLSelect / SQLExec queries.

The full plugin Maven name is org.apache.cayenne.plugins:cayenne-maven-plugin. It can be executed as mvn cayenne:<goal>.

Cayenne Gradle plugin provides tasks similar to Maven plugin. It also provides cayenne extension that has some useful utility methods. Here is example of how to include Cayenne plugin into your project:

Ant tasks are the same as Maven plugin goals described previously, namely "cgen", "cdbgen", "cdbimport". Configuration parameters are also similar (except Maven can guess many defaults that Ant can’t). To include Ant tasks in the project, use the following Antlib:

The first thing you usually want to control during reverse engineering is what exactly should be loaded from database and what not. One of the most common cases is excluding system tables, as you usually don’t want to map them.

Briefly, you are able to include/exclude tables, columns and procedures and do it at several levels: default, catalog, schema. Although everything defined at the top level (default rules) will be applied for the nested elements, all rules from the most specific areas will override general rules (i.e. rules from schemas override rules from catalogs and even more override default rules).

The following use-cases will provide you a better understanding of how filtering works and how you could use it.

In databases relations are defined via foreign keys and there are a lot of different politics according to the level of relationships and ways how those relationships could be modeled in database. Anyway, cdbimport is able to recognize basic patterns of relationships, such as OneToMany, OneToOne and ManyToMany.