Relational Database Metadata Versioning for DDL Snapshot Isolation

The patent segments the monolithic schema modification lock into multiple granular locks organized in a lock graph structure. Instead of requiring a single SCH-M lock on all schema objects, the system creates individual locks for each schema object and organizes them in a graph where nodes represent locks and edges represent conflict relationships. This allows concurrent DDL operations on different schema objects to proceed simultaneously without blocking each other, while still maintaining consistency through the lock graph's conflict detection mechanism.

The patent introduces a lock graph as an intermediary data structure that mediates between multiple DDL operations and the underlying schema objects. The lock graph tracks lock requests, determines conflicts, and coordinates grant decisions without requiring a single exclusive lock. This intermediary enables the system to maintain schema consistency while allowing multiple DDL operations to progress concurrently by resolving conflicts at the lock graph level rather than blocking at the schema object level.

By segmenting the lock requirement into individual schema object locks rather than a global SCH-M lock, the system allows DDL operations to acquire locks on specific objects independently. This segmentation enables overlapping execution of DDL operations and queries on different objects, dramatically reducing the total time the system remains unavailable while maintaining consistency through the lock graph's conflict detection.

The system performs preliminary lock acquisition attempts on individual schema objects before committing the DDL operation. The lock graph tracks these preliminary lock requests and determines which can be granted immediately without blocking. This preliminary action allows multiple DDL operations to make progress concurrently rather than waiting for all queries to complete before any DDL can proceed.

The patent segments the schema modification process into independent lock acquisitions on individual schema objects. This segmentation allows DDL operations to proceed concurrently on different objects without requiring the primary replica to hold a global exclusive lock for the entire duration. As a result, log replication can continue for non-conflicting operations, reducing log accumulation on the primary and improving replication speed to secondary replicas.

The system performs preliminary lock acquisition and conflict detection before the DDL operation commits and generates replication logs. By using the lock graph to determine which DDL operations can proceed without conflicts in advance, the system minimizes the duration of exclusive locks held on schema objects. This preliminary action reduces the window during which log replication must wait, thereby improving replication throughput and reducing log accumulation.

The patent segments schema consistency into object-level units tracked by the lock graph. Instead of invalidating all snapshots when any schema change occurs, the system maintains version information at the individual schema object level. This segmentation allows queries to continue using valid snapshots for the schema objects they access, even when other objects are being modified, thereby reducing query failures and retries while maintaining consistency for affected objects.

The system creates and maintains copies of schema object versions in the lock graph structure. When a DDL operation modifies a schema object, the lock graph preserves the previous version information and creates new version entries. This copying mechanism allows ongoing queries to continue referencing valid historical versions of schema objects, preventing query compilation failures while still allowing the schema to evolve. Queries only need to be retried when their specific schema objects are modified, not when any schema change occurs.