Database Technology

Radim Marek argues that PostgreSQL's MVCC criticisms, including table bloat and vacuum overhead, stem from a 40-year-old architectural decision rather than implementation bugs. He validates these claims by reproducing the issues on a live PostgreSQL 19 beta, showing that every drawback traces back to a specific design choice. The analysis confirms that while the tradeoffs are real, they are inherent to the concurrency model rather than accidental flaws.

• Bloat and vacuum struggles stem from MVCC's concurrency model, not code defects.
• Reproduced write amplification and size doubling on PostgreSQL 19 beta.
• Industry critiques from Uber and Andy Pavlo's group are technically grounded.
• Treat MVCC limitations as architectural tradeoffs to manage, not bugs to fix.

Fujitsu's development team played a key role in shaping PostgreSQL 19 through sustained code contributions and community engagement. Their efforts specifically targeted enhancements to logical replication, adding new capabilities to the database core. This release highlights their growing influence on open-source database innovation and technical direction.

• Fujitsu's contributions directly influenced PostgreSQL 19's logical replication features.
• Sustained code investment signals long-term commitment to open-source database growth.
• Community recognition underscores the technical quality of Fujitsu's engineering efforts.
• Logical replication improvements may reduce sync latency for distributed architectures.

A sponsored feature in The Register argues that while AI architectures have evolved rapidly, underlying storage systems have failed to keep pace with the demand for high-throughput data access. This bottleneck results in GPUs remaining underutilized because they cannot ingest data fast enough from legacy storage infrastructures. The piece highlights a critical mismatch between compute capabilities and data delivery speeds in modern AI workloads.

• Storage throughput is the primary bottleneck preventing GPUs from achieving full utilization.
• Legacy storage architectures are ill-suited for the massive, concurrent read demands of AI training.
• Infrastructure upgrades must prioritize data movement speed alongside compute power.
• Bottlenecks in storage directly impact the efficiency and cost-effectiveness of AI deployments.

The Amazon RDS console has added an auto-migration capability that allows users to move self-managed PostgreSQL databases on EC2 to Amazon RDS for PostgreSQL or Aurora PostgreSQL-Compatible Edition. This feature leverages AWS Database Migration Service (AWS DMS) to handle homogeneous data migrations directly from the UI. It simplifies the process by removing the need for manual DMS task configuration for standard migration scenarios.

• Simplifies migration from EC2-hosted PostgreSQL to RDS or Aurora via the RDS console UI.
• Powered by AWS DMS homogeneous migration capabilities under the hood.
• Reduces manual setup overhead for standard database lift-and-shift operations.
• Supports both Amazon RDS for PostgreSQL and Aurora PostgreSQL-Compatible Editions.

PostgreSQL 16 introduces the gss_accept_delegation GUC, allowing servers to accept delegated Kerberos credentials to authenticate as users against downstream services. This feature is disabled by default due to the significant security trade-offs involved in credential delegation. It provides a mechanism for multi-hop authentication scenarios where the database needs to act on behalf of the client.

• Enable gss_accept_delegation only if multi-hop Kerberos auth is required
• Default is off; enabling it expands the trust boundary for credentials
• Review KDC policies and service principal permissions before activating
• Useful for database acting as a proxy to other secure backend services