Jump to: navigation, search

EclipseLink/DesignDocs/MultiTenantFeatures

Purpose

EclipseLink provides a number of features that help with the challenges of developing multi-tenant applications. This document will provide an overview of some of the ways EclipseLink can be used to support multi-tenant applications and suggest some enhancements that could be made to further enable those applications.

Some of the challenges involved in developing multitenant applications include:

  1. You must decide whether many tenants share a database or tablespace
  2. Will the schema be flexible? If so, hom much?
  3. How do I differentiate data for tenants that share the same DB or tablespace?

EclipseLink provides features that support various decision paths about the above issues.

Here are some examples of the types of multi-tenant applications and how you could enable them in Eclipselink.

Application Architectures

Static Application

In this application, neither the object model nor the database schema is variable. Tenants, for the most part, use the application as it has been provided. If they want to make use of data that is not part of the initial application they are provided a limited number of predefined database fields and cooresponding fields in the object model.

For Example:

Table Customer contains fields: id, name, attribute1, attribute2
Class Customer contains fields: id, name, attribute1, attribute2

In the above example, the id and name fields have predefined functionality in the application. Fields attribute1, and attribute2 can be configured to store any data the customer wants. Typically these fields store data that is mapped direct from Object to database. (In JPA, a Basic mapping)

This application resembles a basic JPA application and, has at its disposal, all the functionality available in JPA and the extended functionality available in EclipseLink. Basically, the application acquires an EntityManagerFactory for each tenant and that EntityManagerFactory serves that tenant only.

Some features of interest that exist in EclipseLink:

  • Converters - If the attribute1 and attribute2 fields need to be stored as different types in the database and in the application, converters could be used to address this
  • VPD Support - When using an Oracle Database, VPD provides an excellent way of securly segregating data between clients. (i.e. with VPD, it is easy to make the same table contain data for different tenants have automatically have only data for a particular tenant used in queries). VPD users can be chosen through EclipseLink's proxy authentication feature and specified through properties in creation of EntityManager factories.
  • Additional Criteria Support - If VPD is not available, it is possible to segregate data by setting an Additional Criteria for a descriptor. This criteria could be used to ensure only data appropriate to a specific client would be returned.
  • SessionCustomizer/DescriptorCustomizer - these could be used in the configuration of persistence units for each tenant. The Additional Criteria listed above is an example of what could be specified
  • Dynamic Specification of persistence.xml - EclipseLink will allow the persistence.xml for a persistence unit to be specified at runtime. This is an additional way to support different configurations per tenant. In this case, each tenant would have a persistence.xml that configured the persistence unit for their part of the application.

Some features that could be implemented to help:

  • VPD Equivalents for databases other that Oracle - Other vendors provide similar functionality and this could be supported if there was demand.
  • Dynamic specification of mapping xml. If mapping xml file could be specified at EntityManagerFactory creation time that would reduce the need to specify different persistence.xml files and any mapping/descriptor overrides could be specified in that file.

Extended Attributes use Map

To provide slightly more flexibility than the Static Application, a Map could be used for extended attributes.

In this case a model could be designed like this:

Table Customer contains fields: id, name
Table Customer_extensions contains fields: customer_id, attributeName, attributeValue
Class Customer contains fields: id, name, extensions

extensions would be defined using a Map mapping and hold any additional fields the customer wanted to use. For instance, an Employee Number would be stored with key = "EmployeeNumber" and the value = <employee number>. Business logic is used to expose the data in extensions as attributes.

Some features of interest that exist in EclipseLink:

  • Map Mapping Strategies: EclipseLink Supports map mapping strategies with many different types of keys and values.

Metadata Driven

It is possible to write a completely metadata driven application in EclipseLink using our Dynamic JPA feature set. A mapping xml file is sufficient to define the Object model. The idea here is that the application provider would define the metadata for the core of the application in XML and provide a mechanism where the tenant could define additional/changed metadata through an additional XML file.

The level of flexibility here is completely dependant on the level of functionality of the database schema.

  1. For a schema like the one listed in our "Static Application" section, the extensibility would be limited to naming the extended part of the object model correctly (e.g. attribute1 in the Customer object could be called "EmployeeID instead)
  2. For a scenario where each tenant had their own database or schema, DDL could be generated to add to the default database based on the user's metadata changes
  3. "Flex Tables" could be provided in a similar way to the flexible columns listed in the "Static Application" scenario. (i.e. Tables with generic names and a predefined set of columns) These tables could be used to house dynamically generated entities, or as secondary tables to hold additional attributes of existing entities.

Some features of interest that exist in EclipseLink:

  • Dynamic JPA - Allows full object model to be defined in xml or in code based on metadata
  • Secondary tables support - Descriptors can use secondary tables to house additional data. Base table could hold application-developer-defined information. Secondary table could hold tenant defined information.
  • DDL Generation - Could be used to define tables for tenant applications

Some features that could be implemented to help:

  • Schema Alteration - allow alter table statements to be issued based on metadata changes by tenant
  • Selective DDL generation - allow DDL generation of certain tables to be switched off

Partially Metadata Driven

With some small changes to EclispeLink, a hybrid of a Static Application and a Metadata driven application could be developed. In this case, the application developer would define static classes, but allow them to be extended by our dynamic JPA feature set.

For example, the application developer would develop an application that contained a Customer class that contains an id and name field but also allow it to be extended by the tenant by allowing an xml file defining extended mappings to be developed.

This kind of application would depend on the same database schema restrictions as the metadata driven application, and allow tenants running on the basic functionality to use only the static classes.

Some features that could be implemented to help:

Application configuration

There are many different options for configuration of the persitence units for use in multi-tenant applications.

Single Persistence Unit

A single persistence unit could be provided that maps all the attributes. This strategy is useful in the cases where both the schema and the object model remain the same.

Deployment per tenant

A full persistence unit could be configured per tenant. That persistence unit could contain different persistence.xml, mapping files, classes and resources. This could be used with any of the above strategies. This is on the opposite extreme from a Single Persistence unit deployment. It allows complete configuration for each tenant, of both the data and object model and could be used with any of the application designs listed above.

Persistence xml per tenant

In this case, each tenant has a persistence.xml file which creates them a separate persistence unit. The classes used by each tenant remain the same. EclipseLink supports dynamic selection of persistence.xml so as long as the persistence xml for each tenant could be provided, this could all be configured in one deployment. This option allows different session-level configuration options for each tenant and could be used as a mechanism for replacing mapping information as well. This type of a configuration is a candidate for any of the application designs indicated above.

eclipselink-orm.xml per tenant

This requires EclipseLink to be enhanced to allow dynamic specification of orm xml

It is possible to add a persistence unit property that allows orm.xml files to be provided by on a per-tenant basis. In this case, everything about the persistence unit would be the same except a single eclipselink-orm.xml file which defines the mappings that are particular to a single tenant. This type of a configuration is a candidate for any of the application designs indicated above.

Metadata in the db

This requires a feature that allows metadata to be stored in the database. This feature would likely have to be implemented both within EclipseLink and within the application server it runs on.

An easy way to store multiple configurations is in some kind of a database. (either a relational db, or a SCM system) Ideally, configuration information could be stored in that way and EclipseLink could be configured to find that information.