📊 HiddenMerit Daily · Issue 53

Focus on Database Frontiers, Practical Insights for DBAs July 2, 2026 ｜ 5 Selected Global Breaking News

01｜OceanBase Releases “Lakehouse‑Integrated” AI Database: Yang Bing Says China Has Opportunity to Define Next‑Gen Database Paradigm

On June 29, OceanBase officially released its Lakehouse‑Integrated AI Database for the AI era, proposing “Lakehouse Integration” as the core architecture – unifying the openness and massive storage capabilities of data lakes, the transaction processing and analytical capabilities of databases, and multi‑modal data processing capabilities into a single strongly‑consistent data foundation, enabling agents to obtain complete business context in one go.

Core Judgement: OceanBase CEO Yang Bing believes that the bottleneck for AI implementation is shifting from “whether the model can think” to “whether it understands the business.” Model capabilities are converging rapidly, and business differentiation is shifting toward the data layer. Data consumers are shifting from “humans” to agents, and the form of data is shifting from structured to multi‑modal. Databases must be rebuilt from the architectural layer rather than patched. In a People’s Daily opinion piece, Yang Bing stated: “We have the opportunity to stand on the same starting line as our overseas counterparts and define the paradigm of AI databases in our own way. For China’s foundational software industry, this is a valuable opportunity worth cherishing.”

Product System Trio:

• Lakebase (Underlying Engine): Enables structured, unstructured, and vector data to be managed, processed, retrieved, and invoked within a unified architecture.
• DataStudio (Data Governance Layer): Covers data access, processing, semantic modelling, and agent collaboration.
• DataPilot (Business Entry Point): Business users complete analytical reports and data dashboards through natural language.

Five Key Technical Features: Lakehouse integration, multi‑modal tables and AI columns, agent‑friendly, open ecosystem (supports S3 and Iceberg), TCO reduced by 30%‑50%.

Validation: Already validated in scenarios such as Ant Afu and Lingguang. Lingguang provides a “one‑sentence app generation” capability to the public and has already hosted 30 million flash applications, validating the feasibility of the lakehouse‑integrated architecture in scenarios with tens of millions of agents.

• DBA Perspective: OceanBase’s “lakehouse‑integrated” approach directly addresses the biggest headache for DBAs in the AI era – the operational complexity and data consistency issues caused by data scattered across multiple systems (transaction DB + data warehouse + vector DB + data lake). The 30%‑50% TCO reduction provides a quantitative basis for DBAs’ cost arguments in technology selection. The design of multi‑modal tables and AI columns also means DBAs need to learn new skills in multi‑modal data modelling and vector index maintenance. The judgement that agents are becoming the new users of databases – Gartner predicts that by 2028, approximately one‑third of enterprise software interactions will be completed by agents – means the DBA’s focus will shift from “serving human users” to “serving AI agents.”

• CTO Perspective: Yang Bing’s judgement of moving “from rule followers to rule definers” is a key signal of the maturity of China’s foundational software industry. The Lakebase + DataStudio + DataPilot trio covers the entire chain from data engine to data governance to business entry point, providing CTOs with an “All‑in‑One” option for AI data architecture planning. It has already been validated in real scenarios such as Ant Afu, handling tens of millions of agents, reducing selection risk.

02｜CETC Kingware, Hygon, and Kylin Reach Strategic Cooperation: “Chip + OS + Database” Iron Triangle Takes Shape

On June 30, CETC Kingware, Hygon Information, and Kylin Software signed a strategic cooperation agreement in Beijing. The three parties will conduct deep technical integration and innovation around the three core foundations – database, chip, and operating system – jointly building an autonomous and controllable AI integrated full‑stack solution, providing a technical path for the digital and intelligent transformation of key industries such as finance, energy, healthcare, and transportation.

Roles of Each Party:

• CETC Kingware: AI data foundation, with a built‑in native AI vector engine, enabling business queries and semantic hybrid retrieval through a single SQL statement.
• Hygon Information: Underlying compute power support, with a CPU+DCU dual‑chip collaborative architecture, handling both general business and AI inference.
• Kylin Software: Full‑stack scheduling hub, with a Panshi kernel + AI unified scheduling system, and a built‑in intelligent O&M assistant.

The three parties stated that they have already implemented benchmark practices in finance (intelligent risk control), energy (intelligent load scheduling), healthcare (multi‑modal retrieval of electronic medical records), and transportation (track sensor data prediction).

• DBA Perspective: The strategic cooperation between Kingware, Hygon, and Kylin is a landmark event in the evolution of the Xinchuang ecosystem from “single‑point replacement” to “full‑stack collaboration.” The deep integration of the three core foundations – chip, operating system, and database – means that the DBA’s scope of responsibility in Xinchuang projects will expand from “managing databases” to “evaluating full‑stack compatibility.” Kingware’s built‑in native AI vector engine also indicates that DBAs need to pay attention to Kingware’s evolution in the multi‑modal convergence direction.

• CTO Perspective: The deep strategic cooperation among the three core domestic vendors – chip, OS, and database – provides CTOs with a “verified full‑stack solution” for Xinchuang selection. The three parties have already established benchmark cases in finance, energy, healthcare, and transportation, reducing selection risk.

03｜MongoDB Atlas Natively Integrates Reranking: Retrieval Quality Improves by 30%, AI Architecture Simplified

At the end of June, MongoDB announced the launch of Native Reranking for Atlas, currently in public preview, powered by Voyage AI and running directly within the MongoDB aggregation pipeline, improving retrieval quality by up to 30%.

Core Value: Reranking technology typically requires introducing additional vendors, APIs, and orchestration layers, increasing system complexity and governance costs. By embedding reranking directly into the database, MongoDB reduces developer operational overhead (less code, no need to build retry logic and error handling) and lowers token costs for scaling AI.

Analyst Views:

• Stephanie Walter of HyperFRAME Research noted: “Every additional AI service adds another node that needs to be managed, secured, monitored, and paid for. Native reranking reduces failure points that can degrade retrieval quality.”
• Ashish Chaturvedi of HFS Research believes that better retrieval is “infrastructure for building trust” – enterprises will not delegate major decisions to AI agents unless they can trust the quality of the information retrieved by the system.

Potential Risk: Analysts warn that native integration could lead to vendor lock‑in, increasing the cost of switching platforms later. Its value also depends on whether MongoDB serves as the organisation’s primary data repository.

• DBA Perspective: MongoDB’s native integration of reranking into the database is a direct confirmation that “databases are becoming full‑chain AI infrastructure.” In the past, DBAs had to maintain complex orchestration logic between the database and independent reranking services; now MongoDB handles it within the aggregation pipeline. For DBAs, this means reduced operational burden, but also requires learning the principles and tuning methods of reranking.

• CTO Perspective: The value of MongoDB’s native reranking lies in simplifying AI architecture. Every independent AI service removed means one less node to manage, secure, and pay for. For CTOs scaling AI applications, this is a worthwhile architectural simplification option to evaluate.

04｜Postgres Pro Enterprise 18.4.1 Released: Standby Read Extensions and Distributed High Availability Upgrades

On July 1, Postgres Professional released Postgres Pro Enterprise 18.4.1, a commercial distribution based on PostgreSQL, with performance, high availability, and security enhancements for large enterprise systems.

Core Enhancements:

• Standby Temporary Object Support: Supports read‑write operations on temporary tables on standby servers, enabling complex report generation on standby without creating independent database copies.

• Proxima Extension Upgrade: Load balancer adds adaptive algorithms, allocating queries based on memory consumption and disk I/O.

• BiHA Multi‑Level Disaster Recovery: Implements multi‑level disaster recovery, supporting cross‑data‑center cascading replication and automatic reconstruction.

• Global Index Enhancements: Supports ON CONFLICT, CONCURRENTLY rebuild, AUTOVACUUM, and other features.

• Management Tools: Adds logical replication slot management, temporary table statistics, and data integrity verification tools.

• DBA Perspective: Postgres Pro Enterprise 18.4.1’s support for standby temporary table read‑write is a major benefit for DBAs in read‑write separation architectures. In the past, generating reports on standby was limited to read‑only; now complex ETL and temporary table operations can be performed on standby, reducing primary load. Proxima adaptive load balancing and BiHA multi‑level disaster recovery also provide DBAs with more refined high‑availability management tools.

• CTO Perspective: The update direction of Postgres Pro Enterprise reflects the core demands of large enterprises for PostgreSQL – horizontal read scaling, geographically distributed disaster recovery, and global index support. Postgres Pro, as a representative commercial distribution of the Russian PostgreSQL ecosystem, is worth watching for its technology roadmap.

05｜Weekly Security Vulnerabilities Focus: Redis Lua Script RCE (CVE-2026-23631) PoC Publicly Available, Splunk Pre‑Auth RCE Chain Requires Urgent Attention

Multiple high‑risk database‑related security vulnerabilities continued to develop this week:

Redis Lua Script Remote Code Execution (CVE-2026-23631) : CVSS 8.1, affecting multiple Redis versions 7.2.x through 8.6.x. An authenticated attacker can exploit the master‑slave data synchronisation mechanism to trigger a memory management flaw on a replica server with read‑only mode disabled, achieving RCE through Lua script operations. PoC and EXP have been made public. Immediate upgrade to versions 7.2.14, 7.4.9, 8.2.6, 8.4.3, or 8.6.3 or higher is recommended. Special Note: Ensure all replica nodes are configured with replica-read-only yes.

Splunk Enterprise Pre‑Auth RCE Vulnerability Chain (CVE-2026-20253) : CVSS 9.8, affecting Splunk Enterprise version 10 and later. The vulnerability stems from the newly introduced PostgreSQL Sidecar Service component in newer Splunk versions, which is enabled by default in AWS cloud deployments. Attackers can exploit directory traversal and PostgreSQL connection string injection to achieve arbitrary file writing and RCE. Splunk Enterprise users on AWS should prioritise patching.

Aix‑DB Unauthorised SQL Queries (CVE-2026-8335) : Affects Aix‑DB 1.2.4 and earlier. The “/llm/process_llm_out” endpoint lacks authentication checks, allowing unauthenticated clients to execute arbitrary SELECT SQL queries.

• DBA Perspective: The PoC/EXP for the Redis vulnerability has been made public. It is recommended to urgently check Redis versions and upgrade. Special attention should be paid to the replica-read-only configuration on replica servers. The particularity of the Splunk vulnerability is that the attack surface is the “database middleware layer” – an authentication flaw in the PostgreSQL Sidecar Service is used as a springboard to achieve RCE. This pattern reminds DBAs that auxiliary services around databases (Sidecars, proxies, migration tools) are becoming new attack entry points. The Aix‑DB vulnerability also shows that the security maturity of emerging AI‑related database components remains concerning.

• CTO Perspective: Both the Redis and Splunk high‑risk vulnerabilities have publicly available PoC/EXP. Security teams should treat them as emergency response items. The Splunk vulnerability is enabled by default in AWS cloud environments, significantly impacting enterprises using Splunk in cloud deployments. It is recommended to include “database peripheral components” in security scanning scope.

📚 SQL Little Knowledge Point

This Issue’s Knowledge Point: What is “Reranking”?

Reranking is a critical component in RAG (Retrieval‑Augmented Generation) systems, positioned between “initial retrieval” and “generating the answer.” Its role is: after the vector retrieval returns a batch of candidate documents, a more refined model re‑ranks them, placing the most relevant documents at the front, thereby improving the quality of the final generated answer.

Why is Reranking Needed?

Initial retrieval (e.g., vector similarity search) prioritises speed, typically using lightweight models to quickly recall Top‑K candidates from billions of records. However, “fast recall” results are not always perfectly accurate – less relevant documents may be ranked ahead of more relevant ones. Reranking uses a more computationally intensive but more accurate model to re‑evaluate the candidate set, ensuring the most relevant documents are prioritised for the large model.

The Value of MongoDB Atlas Native Reranking:

In traditional architectures, reranking requires introducing an independent third‑party service (e.g., a reranking API), adding network calls, data movement, and orchestration complexity. By embedding reranking directly into the database aggregation pipeline, developers no longer need to maintain complex synchronisation and retry logic between the database and the reranking service. According to official data, this feature can improve retrieval quality by up to 30%.

Impact on DBAs:

• Reduced operational burden: one less independent service node to manage, monitor, and pay for.
• However, DBAs need to learn the configuration and tuning methods for reranking.
• Be aware of vendor lock‑in risk: natively integrated features are often difficult to migrate to other platforms.

HiddenMerit Team Production Slogan: 绩优隐于内，金石启新程 | Hidden deep. Merit bold. Forge ahead.