Models, conversion, and schema

Persistable IDs and equality

AbstractCassandraPersistable<PK> implements Spring Data Persistable and Serializable. A subclass supplies getId() and setId().

• isNew() is true while the ID is null.
• Two objects are equal only when their effective user classes and non-null IDs match.
• With an ID, the hash is the ID hash; without one, it uses identity hash.
• Assigning an ID can therefore change the hash code.

Do not put a transient entity into a HashSet or use it as a HashMap key and then assign its ID. Review mutable-entity equality as part of the domain model.

Auditing changes new-state detection

AbstractCassandraAuditable<U, PK> maps created-by, created-at, last-modified-by, and last-modified-at fields. It overrides the parent ID-based rule and returns _createdAt == null from isNew().

@EnableCassandraAuditing
@Table("users")
class UserEntity : AbstractCassandraAuditable<String, UUID>() {
    @PrimaryKey
    private var pk: UUID? = null

    override fun getId(): UUID? = pk
    override fun setId(id: UUID) { pk = id }
}

Without active auditing and an AuditorAware, createdAt is not populated and an entity with an ID may still appear new. The source maps the last-modified user with the column spelling lastModified_by; verify it against an existing schema convention.

Spring Data owns converters

The module does not auto-register application converters. Register custom types in Spring Data Cassandra configuration. The test CurrencyConverter and type-mapping fixtures demonstrate this boundary.

For UDTs, collections, temporal values, and enums, decide whether the driver codec or Spring Data converter owns conversion. Defining conflicting conversion at both levels can make repository and raw-session results disagree.

What SchemaGenerator does

SchemaGenerator.createTableAndTypes reads entity metadata from the Spring Data mapping context and then:

1. discovers UDTs required by the entity or its properties;
2. executes UDT create specifications with IF NOT EXISTS;
3. creates the table only when current keyspace metadata has no such table.

potentiallyCreateTableFor checks table existence only. It does not compare columns, primary keys, clustering order, or table options.

Truncation is destructive

SchemaGenerator.truncate<T> runs Spring Data truncate when the table exists in current keyspace metadata. It is a no-op for an absent table, but removes every row from a present table.

This is useful for test-fixture setup, not an application startup or request path. Use reviewable CQL with history, ordering, and rollback strategy for production migration.

Criteria helper

Criteria.eq is an infix alias for the backtick-heavy Criteria.is(value) call.

val query = Query.query(Criteria.where("tenant_id") eq tenantId)

It does not bypass Cassandra query-model rules. Partition, clustering, and index design remain the caller’s responsibility.

Sources

• AbstractCassandraPersistable.kt
• AbstractCassandraAuditable.kt
• SchemaGenerator.kt
• AbstractCassandraModelTest.kt
• SchemaGeneratorTest.kt