In order to have more freedom while accessing repositories it’s possible to create and use custom repository implementation which implements DataSourceAware interface.
For our example let’s assume it will be class implementing TestRepositoryIfc and our implementation will be using JDBC.
To make it work, we need to define TestRepositoryIfc
as a generic interface extending DataSourceAware
interface.
DataSourceAware
interface will provide definition for methods required by Tigase XMPP Server internals to initialize custom repository classes
based on TestRepositoryIfc
.
TestRepositoryIfc.
public interface TestRepositoryIfc<DS extends DataSource> extends DataSourceAware<DS> { // Example method void addItem(BareJID userJid, String item) throws RepositoryException; }
Next we need to prepare our actual implementation of repository - class responsible for execution of SQL statements. In this
class we need to implement all of methods from our interface
and method void setDataSource(DataSource dataSource) which comes from DataSourceAware interface. In this method we need to initialize data source, ie. create prepared statements.
We should annotate our new class with @Repository.Meta
annotation which will allow Tigase XMPP Server to find this class whenever class implementing TestRepositoryIfc
and with support for data source with jdbc URI.
@Repository.Meta(supportedUris = "jdbc:.*") public static class JDBCTestRepository implements TestRepositoryIfc<DataRepository> { private static final String SOME_STATEMENT = "select * from tig_users"; private DataRepository repository; @Override public void setDataSource(DataRepository repository) { // here we need to initialize required prepared statements try { repository.initPreparedStatement(SOME_STATEMENT, SOME_STATEMENT); } catch (SQLException ex) { throw new RuntimeException("Could not initialize repository", ex); } this.repository = repository; } @Override public void addItem(BareJID userJid, String item) throws RepositoryException { try { PreparedStatement stmt = repository.getPreparedStatement(userJid, SOME_STATEMENT); synchronized (stmt) { // do what needs to be done } } catch (SQLException ex) { throw new RepositoryException(ex); } } }
As you can see we defined type of a data source generic parameter for interface TestRepositoryIfc
. With that we make sure that only instance implementing DataRepository
interface will be provided and thanks to that we do not need to cast provided instance of DataSource
to this interface before any access to data source.
With that in place we need to create class which will take care of adding support for multi-database setup. In our case it
will be TestRepositoryMDBean, which will take care of discovery of repository class, initialization and re-injection of data source.
It is required to do so, as it was just mentioned our TestRepositoryMDBean
will be responsible for initialization of JDBCTestRepository
(actually this will be done by MDRepositoryBean
which is extended by TestRepositoryMDBean
.
@Bean(name = "repository", parent = TestComponent.class, active = true) public static class TestRepositoryMDBean extends MDRepositoryBeanWithStatistics<TestRepositoryIfc> implements TestRepositoryIfc { public TestRepositoryMDBean() { super(TestRepositoryIfc.class); } @Override public Class<?> getDefaultBeanClass() { return TestRepositoryConfigBean.class; } @Override public void setDataSource(DataSource dataSource) { // nothing to do here } @Override public void addItem(BareJID userJid, String item) throws RepositoryException { getRepository(userJid.getDomain()).addItem(userJid, item); } @Override protected Class<? extends TestRepositoryIfc> findClassForDataSource(DataSource dataSource) throws DBInitException { return DataSourceHelper.getDefaultClass(TestRepositoryIfc.class, dataSource.getResourceUri()); } public static class TestRepositoryConfigBean extends MDRepositoryConfigBean<TestRepositoryIfc> { } }
Most of this code will be the same in all implementations based on MDRepositoryBeanWithStatistics
. In our case only custom method is void addItem(…) which uses getRepository(String domain) method to retrieve correct repository for a domain.
This retrieval of actual repository instance for a domain will need to be done for every custom method of TestRepositoryIfc
.
Tip
It is also possible to extend MDRepositoryBean
or SDRepositoryBean
instead of MDRepositoryBeanWithStatistics
. However, if you decide to extend abstract repository bean classes without withStatistics
suffix, then no statistics data related to usage of this repository will be gathered. The only change, will be that you will
not need to pass interface class to constructor of a superclass as it is not needed.
Note
As mentioned above, it is also possible to extend SDRepostioryBean
and SDRepositoryBeanWithStatistics
. Methods which you would need to implement are the same is in case of extending MDRepositoryBeanWithStatistics
, however internally SDRepositoryBean
will not have support for using different repository for different domain. In fact SDRepositoryBeanWithStatistics
has only one repository instance and uses only one data source for all domains.
The same behavior is presented by MDRepositoryBeanWithStatistics
if only single default
instance of repository is configured. However, MDRepositoryBeanWithStatistics
gives better flexibility and due to that usage of SDRepositoryBean
and SDRepositoryBeanWithStatistics
is discouraged.
While this is more difficult to implement than in previous version, it gives you support for multi database setup and provides you with statistics of database query times which may be used for diagnosis.
As you can also see, we’ve annotated TestRepositoryMDBean with @Bean annotation which will force Tigase Kernel to load it every time TestComponent will be loaded. This way it is possible to inject instance of this class as a dependency to any bean used by this component (ie. component, module, etc.) by just creating a field and annotating it:
@Inject private TestRepositoryIfc testRepository;
Tip
In testRepository field instance of TestRepositoryMDBean will be injected.
Note
If the class in which we intend to use our repository is deeply nested within Kernel dependencies and we want to leverage
automatic schema versioning we have to implement tigase.kernel.beans.RegistrarBean
in our class!
Our class TestRepositoryMDBean
is annotated with @Bean
which sets its name as repository
and sets parent as TestComponent
. Instance of this component was configured by use under name of test
in Tigase XMPP Server configuration file.
As a result, all configuration related to our repositories should be placed in repository
section placed inside test
section.
Example.
test(class: TestComponent) { repository () { // repository related configuration } }
As mentioned above, if we use MDRepositoryBeanWithStatistics
as our base class for TestRepositoryMDBean
, then we may have different data sources used for different domains.
By default, if we will not configure it otherwise, MDRepositoryBeanWithStatistics
will create only single repository instance named default
.
It will be used for all domains and it will, by default, use data source named the same as repository instance - it will use
data source named default
.
This defaults are equal to following configuration entered in the config file:
test(class: TestComponent) { repository () { default () { dataSourceName = 'default' } } }
It is possible to make any repository use different data source than data source configured under the same name as repository
instance.
To do so, you need to set dataSourceName
property of repository instance to the name of data source which it should use.
Example setting repository default
to use data source named test
.
test(class: TestComponent) { repository () { default () { dataSourceName = 'test' } } }
To configure repository instance to be used for particular domain, you need to define repository with the same name as domain for which it should be used. It will, by default, use data source with name equal domain name.
Separate repository for example.com
using data source named example.com
.
dataSource () { // configuration of data sources here is not complete default () { uri = "jdbc:derby:/database" } 'example.com' () { uri = "jdbc:derby/example" } } test(class: TestComponent) { repository () { default () { } 'example.com' () { } } }
Separate repository for example.com
using data source named test
.
dataSource () { // configuration of data sources here is not complete default () { uri = "jdbc:derby:/database" } 'test' () { uri = "jdbc:derby/example" } } test(class: TestComponent) { repository () { default () { } 'example.com' () { dataSourceName = 'test' } } }
Note
In both examples presented above, for domains other than example.com
, repository instance named default
will be used and it will use data source named default
.
It’s also possible to enable repository versioning capabilities when creating custom implementation. There are a couple of parts/steps to fully take advantage of this mechanism.
Each DataSource
has a table tig_schema_versions
which contains information about component schema version installed in the database associated with particular DataSource.
First of all, repository implementation should implement tigase.db.util.RepositoryVersionAware
interface (all it’s methods are defined by default) and annotate it with tigase.db.Repository.SchemaId
. For example
.Repository annoted with SchemaId
and implementing RepositoryVersionAware
@Repository.SchemaId(id = "test-component", name = "Test Component") public static class TestRepositoryMDBean extends MDRepositoryBeanWithStatistics<TestRepositoryIfc> implements TestRepositoryIfc { … }
This action alone will result in performing the check during Tigase XMPP Server startup and initialisation of repository whether
tables, indexes, stored procedures and other elements are present in the configured data source in the required version.
By default, required version matches the implementation version (obtained via call to java.lang.Package.getImplementationVersion()
), however it’s possible to specify required version manually, either:
- by utilizing
tigase.db.util.RepositoryVersionAware.SchemaVersion
annotation:
@Repository.SchemaId(id = "test_component", name = "Test Component") @RepositoryVersionAware.SchemaVersion(version = "0.0.1") public static class TestRepositoryMDBean extends MDRepositoryBeanWithStatistics<TestRepositoryIfc> implements TestRepositoryIfc { … }
- or by overriding
tigase.db.util.RepositoryVersionAware.getVersion
method:
@Override public Version getVersion() { return "0.0.1"; }
To detect that version information in database is inadequate following logic will take place:
- if there is no version information in the database the service will be stopped completely prompting to install the schema
(either via
update-schema
orinstall-schema
depending on user preference); -
if there is an information about loaded component schema version in the repository and the base part of the required schema version (i.e. taking into account only major.minor.bugfix part) is different from the one present in the repository then:
- if the required version of the component schema is final (i.e. non
SNAPSHOT
) the server will shutdown and print in the log file (namelylogs/tigase-console.log
) terminal error forcing the user to upgrade the schema; - if the required version of the component schema is non-final (i.e. having
SNAPSHOT
part) then there will be a warning printed in the log file (namelylogs/tigase-console.log
) prompting user to run the upgrade procedure due to possible changes in the schema but the server will not stop;
- if the required version of the component schema is final (i.e. non
Upgrade of the loaded schema in the database will be performed by executing:
./scripts/tigase.sh upgrade-schema etc/tigase.conf
The above command will load current configuration, information about all configured data sources and enabled components, and then perform upgrade of the schema of each configured component in the appropriate data source.
Depending on the type of the database (or specified annotation), how the upgrade procedure is handled internally is slightly different.
For all relational databases (MySQL, PostgreSQL, MS SQL Server, etc…) we highly recommend storing complete database schema
in external files with following naming convention: <database_type>-<component_name>-<version>.sql
, for example complete schema for our Test component version 0.0.5 intended for MySQL would be stored in file named mysql-test-0.0.5.sql
. What’s more - schema files must be stored under database/
subdirectory in Tigase XMPP Server installation directory.
Note
this can be controlled with external
property of Repository.SchemaId
annotation, which defaults to "true", if set to false
then handling will be done as described in ???
For example:
database/mysql-test-0.0.1.sql
database/mysql-test-0.0.2.sql
database/mysql-test-0.0.3.sql
database/mysql-test-0.0.4.sql
database/mysql-test-0.0.5.sql
During the upgrade process all required schema files will be loaded in the ascending version order. Version range will depend on the conditions and will follow simple rules:
- Start of the range will start at the next version to the one currently loaded in the database (e.g. if the current version
loaded to the database is
0.0.3
and we are deploying component version0.0.5
then SchemaLoader will try to load schema from files:database/mysql-test-0.0.4.sql
anddatabase/mysql-test-0.0.5.sql
) - If we are trying to deploy a SNAPSTHOT version of the component then schema file matching that version will always be included in the list of files to be loaded
(e.g. if we are trying to deploy a nightly build with component version
0.0.5-SNAPSHOT
and currently loaded schema version in the database is0.0.5
then SchemaLoader will includedatabase/mysql-test-0.0.5.sql
in the list of files to be loaded)
It’s also possible to skip above filtering logic and force loading all schema files for particular component/database from
database/
directory by appending --forceReloadAllSchemaFiles=true
parameter to the upgrade-schema
/install-schema
command.
If there is a need to handle database schema internally (for example for cases like NoSQL databases or simply there is such
preference) then it’s possible to do so by setting external
attribute of Repository.SchemaId
annotation to false
:
@Repository.SchemaId(id = "test_component", name = "Test Component", external = false)
In such case, updateSchema
method from tigase.db.util.RepositoryVersionAware
interface should be implemented to handle installation/updating of the schema. It takes two arguments:
Optional<Version> oldVersion
- indicating current version of the schema loaded to the database (if it’s present)Version newVersion
- indicating required version (either version of component or specific version of the repository)
Each versioned schema file should consist at the end code responsible for setting appropriate version of the loaded schema in the form of Stored Procedure call with the name of the component and the version as parameters:
- Postgresql
-- QUERY START: select TigSetComponentVersion('test_component', '0.0.5'); -- QUERY END:
- MsSQL Server
-- QUERY START: exec TigSetComponentVersion 'test_component', '0.0.5'; -- QUERY END: GO
- MySQL
-- QUERY START: call TigSetComponentVersion('test_component', '0.0.5'); -- QUERY END:
- Derby
-- QUERY START: call TigSetComponentVersion('test_component', '0.0.5'); -- QUERY END:
In case of schema handled internally, after successful load (i.e. execution of the implemented tigase.db.util.RepositoryVersionAware.updateSchema
method returning tigase.db.util.SchemaLoader.Result.ok
) the version in the database will be set to the current version of the component.
This allows (in case of schema handled externally) to load it by hand by directly importing .sql
files into database.