meta data

We’ve recently encountered this interesting use-case on the jOOQ user group. How do you discover all primary keys of your schema via the jOOQ API? There are two ways:

Using the generated meta data
Using runtime meta data

Let’s see how it works:

Using the generated meta data

This is straightforward. If you’re using Java 8, you can run the following program

GENERATED_SCHEMA
    .getTables()
    .stream()
    .map(t -> t.getPrimaryKey())
    .filter(Objects::nonNull)
    .forEach(System.out::println);

Whereas with Java 7 or less, you’d write

for (Table<?> t : GENERATED_SCHEMA.getTables()) {
    UniqueKey<?> key = t.getPrimaryKey();

    if (key != null)
        System.out.println(key);
}

Using runtime meta data

If you’re not using the code generator, the same can be achieved with:

DSL.using(configuration)
   .meta()
   .getTables()
   .[ same as above ]

That’s it! Another interesting use case of querying the meta information via the jOOQ API can be seen here.

An important part of jOOQ is jooq-meta, the database schema navigation module. This is used by the code generator to discover relevant schema objects. I was asked several times why I rolled my own instead of using other libraries, such as SchemaCrawler or SchemaSpy, and indeed it’s a pity I cannot rely on other stable third party products. Here are some thoughts on database schema navigation:

Standards

The SQL-92 standard defines how RDBMS should implement an INFORMATION_SCHEMA containing their dictionary tables. And indeed, some RDBMS do implement parts of the standard specification. These RDBMS ship with some implementation of the standard.

Close to the standard

HSQLDB: very close to the true standard
Postgres: close to the standard, with some tweaks (also has proprietary dictionary tables)
SQL Server: close to the standard but quite incomplete (also has proprietary dictionary tables)

Liberal interpretation of the standard

H2 (some backwards-incompatible changes, recently)
MySQL (only since 5.0, also has proprietary dictionary tables)

Other RDBMS provide their own idea of dictionary tables. This is something very tricky for schema navigation tools like jOOQ, to get a hold of. The dictionary table landscape can be described like this (my biased opinion):

Neat and well-documented dictionary tables

DB2: These dictionary tables somehow look like the standard, with different names. They feel intuitive.
Oracle: In my opinion has a better set of dictionary views than the ones proposed by the standard. Very easy to understand and well-documented all over the Internet
SQLite: There are no dictionary tables, but the SQLite stored procedures are very simple to use. It’s a simple database, after all

Hard to understand, not well-documented dictionary tables

Derby: Created the notion of conglomerates instead of using normal database-speak, such as relations, keys, etc.
MySQL: the old mysql schema was quite a pain. Fortunately, this is no longer true with MySQL 5.0
Ingres: Well… Ingres is an old database. Usability was not one of the main things in the 70’s…
Sybase SQL Anywhere: Lots of objects that have to be joined in complicated relations. Documentation is scarce
Sybase ASE: Even more difficult than SQL Anywhere. Some data can only be obtained with “tricks”

JDBC abstraction

The variety of dictionary tables seems to scream for standard abstraction. While the SQL-92 standard could in fact be implemented on most of these RDBMS, JDBC abstraction is even better. JDBC knows of the DatabaseMetaData object and allows for navigating database schemata easily. Unfortunately, every now and then, this API will throw a SQLFeatureNotSupportedException. There is no general rule about which JDBC driver implements how much of this API and when a workaround is needed. For jOOQ code generation, these facts make this API quite useless.

Other tools

There are some other tools in the open source world, as mentioned previously. Here are some drawbacks of using those tools in jOOQ:

Both tools that I know of are licensed with LGPL, which is not nicely compatible with jOOQ’s Apache 2 license.
Both tools navigate the entity-relationships very well, but seem to lack support for many non-standard constructs, such as UDT’s, advanced stored procedure usage (e.g. returning cursors, UDT’s, etc), ARRAY’s
SchemaCrawler supports only 8 RDBMS, jOOQ has 12 now
Both tools are rather inactive. See here and here

For more information, visit their sites:

jooq-meta

Because of the above reasons, jOOQ ships with its own database schema navigation: jooq-meta. This module can be used independently as an alternative to JDBC’s DatabaseMetaData, SchemaCrawler or SchemaSpy. jooq-meta uses jOOQ-crafted queries to navigate database meta-data, hence it is also part of the integration test suite. As an example, see how the Ingres foreign key relationships are navigated with jooq-meta:

Result<Record> result = create()
    .select(
        IirefConstraints.REF_CONSTRAINT_NAME.trim(),
        IirefConstraints.UNIQUE_CONSTRAINT_NAME.trim(),
        IirefConstraints.REF_TABLE_NAME.trim(),
        IiindexColumns.COLUMN_NAME.trim())
    .from(IICONSTRAINTS)
    .join(IIREF_CONSTRAINTS)
    .on(Iiconstraints.CONSTRAINT_NAME.equal(IirefConstraints.REF_CONSTRAINT_NAME))
    .and(Iiconstraints.SCHEMA_NAME.equal(IirefConstraints.REF_SCHEMA_NAME))
    .join(IICONSTRAINT_INDEXES)
    .on(Iiconstraints.CONSTRAINT_NAME.equal(IiconstraintIndexes.CONSTRAINT_NAME))
    .and(Iiconstraints.SCHEMA_NAME.equal(IiconstraintIndexes.SCHEMA_NAME))
    .join(IIINDEXES)
    .on(IiconstraintIndexes.INDEX_NAME.equal(Iiindexes.INDEX_NAME))
    .and(IiconstraintIndexes.SCHEMA_NAME.equal(Iiindexes.INDEX_OWNER))
    .join(IIINDEX_COLUMNS)
    .on(Iiindexes.INDEX_NAME.equal(IiindexColumns.INDEX_NAME))
    .and(Iiindexes.INDEX_OWNER.equal(IiindexColumns.INDEX_OWNER))
    .where(Iiconstraints.SCHEMA_NAME.equal(getSchemaName()))
    .and(Iiconstraints.CONSTRAINT_TYPE.equal("R"))
    .orderBy(
        IirefConstraints.REF_TABLE_NAME.asc(),
        IirefConstraints.REF_CONSTRAINT_NAME.asc(),
        IiindexColumns.KEY_SEQUENCE.asc())
    .fetch();

Conclusion

Once more it can be said that the world of RDBMS is very heterogeneous. Database abstraction in Java is established only to a certain degree in technologies such as JDBC, Hibernate/JPA, and third party libraries such as SchemaCrawler, SchemaSpy, and jooq-meta.

Java, SQL and jOOQ.