📊 HiddenMerit Daily · Issue 13

Focus on Database Frontiers, Practical Insights for DBAs May 8, 2026 ｜ 5 Selected Global Breaking News

01｜CETC Kingware Replaces Oracle+PG Hybrid Architecture in Real‑World Tests: TPS Up 35%, Breaking the “One Database Replaces Two” Barrier

In a recent migration validation for a large telecom group’s core network management system, CETC Kingware (formerly Renda Kingware) achieved a 35% increase in core transaction throughput (TPS) after replacing the existing Oracle and PostgreSQL hybrid architecture. The system covers 12 professional domains (transmission, wireless, etc.), involved full migration of 8 TB of historical data, and smoothly switched 11 Oracle instances and 2 PostgreSQL instances, with 100% data accuracy. Under dual high loads – maintaining strong transaction consistency while delivering excellent GIS spatial data processing capability – peak CPU usage dropped from 88% to 72%. Average response latency decreased by 27%, and GIS spatial query time was reduced by 20.8%. Kingware’s kernel‑level dynamic syntax adaptation technology simultaneously supports multiple syntax modes (Oracle, PostgreSQL, MySQL, etc.) in a single database, compressing complex heterogeneous “multi‑database collaboration” architectures into a single domestic platform.

· DBA Perspective: This validation of “one database replacing two heterogeneous databases” directly establishes Kingware’s unique cost and operational advantages in the current database selection landscape. The core challenge for DBAs will no longer be “how to replace A with B”, but “how one database can support multiple dialects and fuse multiple workload types”. You need to at least understand the core differences among mainstream syntaxes, plan workload, data heat, and capacity partitions, and calculate the total cost–benefit of “N‑to‑1” consolidation. · CTO Perspective: This mixed‑replacement case for a telecom operator is essentially a trigger point for the huge cost‑reduction potential of heterogeneous database convergence platforms. For enterprises with long‑standing technical debt, this is a major strategic option – no longer needing to maintain separate operations systems for niche departmental databases. A unified underlying platform compresses management overhead and human resource allocation. · Investor Perspective: A single domestic database replacing an Oracle+PG hybrid architecture demonstrates that domestic vendors now have the comprehensive ability to handle both ultra‑fast OLTP performance and complex GIS spatial data. Kingware’s ability to win core financial‑sector contracts will strengthen, and its valuation expectations in the capital market could be revised upward accordingly.

02｜Dameng’s AI Strategy Surfaces: “Intermediate Language Layer” to Solve Industry AI Implementation Challenges

At the Dameng Database Conference on April 22, Feng Yuan, Deputy General Manager of Dameng, officially released Dameng’s AI strategy. The core concept is to build an “intermediate language layer” between vertical‑domain models and the database, abstracting industry semantics into unified semantic objects and then mapping them to SQL or database operators via rules. In the OLAP domain, AI’s real value lies in deepening the “depth” of analytics – the database not only stores and queries faster but also participates in deeper analytical processes, improving its understanding of data semantics. For OLTP scenarios, Dameng advocates a clear sense of “boundary”: hardcore TP (e.g., fund transactions) should still rely on traditional capabilities, while in general TP scenarios, the bottleneck is not the database but business logic. AI can turn previously hard‑coded logic into dynamic decision‑making. Dameng predicts that in the future, an increasing number of “access subjects” to databases will be AI agents, requiring support for fast start/stop, quick instance creation/destruction, and agent‑adapted access patterns.

· DBA Perspective: Dameng’s “DB × Agent” architecture signals that DBAs will face pressure from billions of agents. Agent access patterns to databases will be high‑frequency, dynamic, small‑batch but persistent and concurrent. Old rules for monitoring slow SQL, lock analysis, and connection pool tuning will be overturned. DBAs must develop new operational skills for “tenant‑level fast elastic start/stop” and “agent‑oriented audit trails”. · CTO Perspective: In the “DB × Agent” era, data architecture needs to expand from passive query storage to upper‑level perception and decision‑making. Once this direction matures, it will effectively compress the time enterprises need to deliver AI pilots. On the eve of the AI strategy explosion, the earlier an enterprise adopts the “semantic middleware” logic, the more likely it is to gain engineering initiative. · Investor Perspective: Dameng’s AI layout stays true to the essence of databases – managing and processing data, which is the underlying need for all AI applications. This gives Dameng a valuation balance of “stability” and “cutting‑edge”. Going forward, track the progress of Dameng’s “intermediate language” engineering implementation in key industries such as finance.

03｜Dameng Q1 Results Again Exceed Expectations: Revenue 410M RMB (+59% YoY), Gross Margin 96.6%

Dameng’s Q1 2026 revenue reached RMB 410 million, a year‑on‑year increase of 59.1%; net profit attributable to shareholders was RMB 151 million, up 54.3% YoY, maintaining high growth from a large base. Benefiting from accelerated Xinchuang procurement in 2025, contract liabilities at the end of the year increased 59.8% YoY, indicating a strong order backlog. Product mix continues to optimise – gross margin climbed to 96.1% for full‑year 2025 and further improved to 96.6% in Q1 2026. The company launched new products in April, including DM9, GDMBASE V4.0, and the DAMENG PAI all‑in‑one machine. Among them, DM9’s single‑node TPC‑C doubled compared to its predecessor, signalling a deep transformation toward an AI database.

· DBA Perspective: Dameng’s frequent quarterly earnings reports show that order‑driven effects are extremely pronounced in the domestic database industry. The most direct impact for DBAs: Dameng and its competitors will accelerate their recruitment of technical talent, and the job market for consultant‑type DBAs will significantly strengthen over the next 18 months. · CTO Perspective: Several years of nearly double‑digit growth combined with high absolute gross margins reveal Dameng’s pricing power and product strength in key Xinchuang replacement sectors. When selecting a database path, CTOs must balance short‑term project budgets with long‑term technological sustainability. · Investor Perspective: Consecutive earnings beats position Dameng as a growth leader in the Xinchuang sector over the next 2–3 years. DM9’s doubling of single‑node TPC‑C creates a differentiated competitive advantage over other domestic products. Investors should continue to track the progress of Dameng’s marquee customer wins in the financial vertical.

04｜MongoDB 8.3 GA: Built for AI Speed, AI Agent Memory Engine Takes Centre Stage

On May 7, at the .local conference in London, MongoDB officially announced the GA of MongoDB 8.3, positioning it as “the database built for AI speed”. Compared to version 8.0, 8.3 delivers +35% write throughput, +45% read throughput, and +15% ACID transaction throughput without any application code changes. On the AI features front, Atlas Vector Search adds a public preview of Automated Voyage AI Embeddings, automatically generating embeddings when data is written or updated, providing AI agents with accurate, real‑time context. LangGraph.js long‑term memory storage is now GA, enabling JavaScript and TypeScript developers to persist user preferences and interaction histories across conversations.

· DBA Perspective: The demand for persistent memory for AI agents is becoming a major new workload for mainstream databases – DBAs need to manage vector generation pipelines and storage costs while balancing performance and recall accuracy. DBAs must now understand: does embedding generation latency impact write throughput? How do vector and scalar filters coordinate in hybrid search? What are the scaling solutions for memory storage under massive agent concurrency? · CTO Perspective: MongoDB 8.3 resolves the inherent tension between long‑term agent memory and high‑performance search by offering a unified data platform that reduces operational and coordination overhead from stacking multiple components. For technical teams building customised agents, it provides a standard reference for the DevOps stack and clear reference value for agent development projects in 2026. · Investor Perspective: MongoDB demonstrates the elasticity of the non‑relational database model in the AI era – a triple platform of vector search + model inference + long‑term memory can effectively offset competition from large cloud databases and open‑source in‑memory stores. Locking in developer experience and downstream licensing revenue remains the core driver of mid‑term valuation recovery.

05｜PostgreSQL Ecosystem Stays Active: AI DBA Assistant Debuts, GraphRAG Extension Upgraded

On May 6, pgEdge officially released the AI DBA Workbench, an AI‑powered monitoring and management tool dedicated to Postgres teams, providing a 24/7 database management co‑pilot. pg_sorted_heap 0.13.0 was released concurrently, adding a stable GraphRAG capability that performs graph‑like fact retrieval and re‑ranking directly inside PostgreSQL. It supports ANN seed retrieval then expands relationship hops and re‑ranks candidate sets precisely, keeping the retrieval process inside PG without needing an external vector search sidecar. The PostgreSQL community is currently highly focused on stabilising PG 19, targeting a formal release in September 2026. European teams such as CYBERTEC have already contributed substantial kernel code.

· DBA Perspective: The AI DBA assistant will provide 7×24 real‑time help with diagnostics, anomaly prediction, and change assessment. Meanwhile, pg_sorted_heap’s GraphRAG capability opens new possibilities for “JSON vector + graph query” within a single kernel. If your application logic involves knowledge expansion of 1–2 hops, you may rely on PG without additional graph databases, greatly simplifying your tech stack. · CTO Perspective: Although pg_sorted_heap 0.13.0’s GraphRAG capability is still early, it could reshape the RAG technology landscape – performing scalar filtering, vector similarity, and simple graph hop expansion within a single database. CTOs should no longer default to stacking micro‑services on top of separate graph and vector databases, but actively explore the engineering ceiling of PostgreSQL as a unified foundation. · Investor Perspective: pgEdge raising funding and launching an AI DBA product, combined with the community’s deep exploration of GraphRAG and vector search, shows that the PostgreSQL ecosystem retains strong vitality in the AI era. Investing in PostgreSQL toolchains, operations automation, and migration services companies is a sound and cautious niche strategy.

📌 Issue Summary

News Core Keywords DBA Actions CTO/Decision‑Maker Focus Investor Perspective CETC Kingware replaces Oracle+PG hybrid TPS +35%, one DB for two, heterogeneous syntax fusion Learn multi‑syntax dynamic adaptation, plan workload partitions for N‑to‑1 consolidation Assess management & labour compression from unified platform Validates domestic DBs’ comprehensive ability; strong signal for financial core contract wins Dameng AI strategy surfaces Intermediate language layer, DB × Agent, general TP intelligence Build skills for tenant‑level fast elastic start/stop and agent audit trails Consider AI planning efficiency; integrate “semantic middleware” into data architecture evolution AI strategy stays in database’s core domain; track “intermediate language” engineering progress Dameng Q1 results beat again Revenue +59%, GPM 96.6%, DM9 single‑node TPC‑C doubles Top vendor hires aggressively; accumulate domestic DB skills regionally Balance short‑term budgets with long‑term sustainability when selecting Xinchuang paths Consecutive beats set Dameng as growth leader; watch financial‑sector flagship wins MongoDB 8.3 GA AI speed, automated embeddings, agent long‑term memory Manage vector pipelines and scalability; evaluate performance vs. recall trade‑offs Reduce multi‑component O&M load; set AI agent platform baseline Triple‑platform offsets open‑source competition; developer experience drives mid‑term valuation PostgreSQL ecosystem active AI DBA co‑pilot, GraphRAG stable, PG 19 stabilisation Follow AI‑assisted diagnosis trends; assess GraphRAG for single‑source knowledge reasoning Explore PostgreSQL unified base’s engineering ceiling to reduce micro‑service debt PG toolchain and ecosystem enjoy open‑source dividends; form moat for downstream startups

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