Segment 1: HiddenMerit Morning Post · Issue 93

Focus on Database Frontiers, Practical Insights for DBAs September 3, 2026 ｜ 5 Selected Global Breaking News

01｜VLDB 2026 Opens in Boston: Agentic Data Systems Take Centre Stage, Multiple Papers Focus on AI‑Database Integration

From August 31 to September 4, the 52nd International Conference on Very Large Databases (VLDB 2026) is being held in Boston, USA. This year’s conference has listed “Data Systems for the AI Agent Era” as one of its core themes, with dedicated forums and workshops on Agentic Data Systems and database‑large model collaboration.

Multiple papers presented at the conference reflect a paradigm shift in database research: from “how to store and query fast” to “how to make data understandable and usable by AI agents.” In Tencent Cloud Database’s DBTalk “Three Consecutive Top Conference Hits” event, Professor Zhang Feng from Renmin University, Researcher Li Huan from Zhejiang University, and Researcher Yang Chengcheng from East China Normal University offered their latest observations from three directions: system evolution, cloud‑native data infrastructure, and testing and evaluation.

The three presentations converged on the same question: When workloads are being rewritten by AI, how should databases be redesigned? Alibaba Cloud’s Zhou Jingren will deliver a talk on September 3, exploring the efficient and reliable systems required for large‑scale foundation model training, post‑training, and inference.

DBA Perspective: VLDB 2026 sends a clear signal – database research is shifting from “performance optimisation” to “AI‑Ready architectures.” For DBAs, this means that skill development over the next 3‑5 years needs to expand from “tuning SQL parameters” to “understanding AI agent data access patterns, multi‑modal data management, and the operational logic of heterogeneous hardware (GPU/RDMA).”

02｜Modb September Ranking: OceanBase Holds Top Spot with 884 Points, Dameng and Kingware Hold Top Three

The Modb September database popularity ranking shows OceanBase firmly in first place with 884.27 points, Dameng in second with 726.08 points, and Kingware holding third with 685.02 points.

Monthly Highlights:

PingCAP Database (formerly TiDB) ranks sixth with 472.56 points, just 1.34 points behind fifth‑place GoldenDB (473.90 points) – competition entering “decimal‑level” intensity.
PolarDB returned to fourth place after dropping to sixth last month.
GaussDB ranks seventh, up two positions from six months ago – one of the biggest gainers on the ranking.
TDengine ranks tenth, the only time‑series database in the top ten, having held this position for two years.

Market Landscape: The Frost & Sullivan report shows that during 2025H2‑2026H1, the China distributed transaction database market reached approximately RMB 5.95 billion. OceanBase led with 20.2% market share, followed by Alibaba Cloud (17.9%), Tencent Cloud (16.5%), Huawei Cloud (14.3%), and GoldenDB (10.1%), with the top five accounting for 78.9% .

DBA Perspective: The 1.34‑point gap between PingCAP Database and GoldenDB shows that competition in the distributed relational database track has entered a “white‑hot” phase. When making Xinchuang selections, DBAs cannot rely solely on rankings but must also consider business scenario adaptability and ecosystem maturity.

03｜EDB Postgres AI Summit Seoul 2026 Held Today: Korean PostgreSQL Ecosystem Accelerates

On September 3, EDB is hosting Postgres AI Summit Seoul 2026 in Seoul, themed “Change the Game.” Over 300 C‑level and IT decision‑makers from finance, manufacturing, IT, and retail sectors are expected to attend.

Core Agenda:

IBK Enterprise Bank sharing “the secret of converting 15 core systems simultaneously.”
Korea’s largest semiconductor company sharing DBaaS platform MPP DB diversification and next‑gen data warehouse “WarehousePG” adoption experience.
Kyobo Bookstore sharing enterprise DB modernisation and open‑source migration cases.
Shopcast sharing “Agentic Lakehouse” construction cases combining AI and data lakehouse.

EDB Korea stated that the PostgreSQL ecosystem has grown from single‑point products to a complete industry ecosystem covering “introduction consulting → conversion → operations → AI applications.” EDB positions PostgreSQL as the best choice for the “post‑Oracle era,” and its “Oracle compatibility 95%+” is a core reference indicator for DBAs performing Oracle migrations.

DBA Perspective: The maturation path of the Korean PostgreSQL ecosystem offers a valuable reference for domestic DBAs. From financial core systems (IBK converting 15 core systems simultaneously) to manufacturing (semiconductor companies adopting WarehousePG) and retail (Kyobo Bookstore), PostgreSQL’s industry coverage in Korea is now quite comprehensive. For DBAs, PostgreSQL skills are moving from “nice‑to‑have” to “must‑have.”

04｜MongoDB .Local Seoul 2026: AI Search and MCP Support Become Entry Points into the Korean Market

On September 1, MongoDB held its .Local Seoul 2026 media briefing in Seoul. MongoDB Korea has defined its two pillars for the Korean business as “Modernisation” and “AI Innovation” – transforming traditional relational databases and siloed data architectures into flexible data platforms, and building on these to support AI applications and agent construction.

Core Advantages: MongoDB Atlas supports data read/write, full‑text search (keyword search), vector search (semantic search), embedding (data vectorisation), and reranking (search result re‑ranking) all on a single platform. Customers no longer need to move and synchronise data across multiple search systems and vector databases – the latest data can be directly used for AI search.

MCP Support: The Atlas Managed MCP Server allows enterprises to connect MCP‑compatible AI tools (ChatGPT, Claude, Codex, Cursor, etc.) to MongoDB without operating their own servers. Service accounts, user authentication, and auditing features enable enterprises to manage agent access to data.

DBA Perspective: MongoDB’s “single platform for read/write + full‑text + vector + reranking” architecture is simplifying the data stack for AI applications. For DBAs, this means the traditional multi‑component stack of “relational DB + ES + vector DB” is being replaced by a single platform. DBAs need to learn MongoDB Atlas hybrid search configuration and MCP Server permission management.

05｜Economic Daily: China Database Market to Reach RMB 97.974 Billion by 2028, Technology Evolution Toward “AI Empowerment”

Economic Daily, citing CAICT’s “Database Development Research Report (2026),” reported that China’s database market reached $9.49 billion (approximately RMB 67.796 billion) in 2025 and is projected to reach RMB 97.974 billion by 2028.

Three Key Characteristics:

Shift from quantitative expansion to quality improvement: Product offerings are becoming increasingly diverse, with leading enterprises’ technical expertise and service capabilities continuing to strengthen.
Technology evolution toward “AI empowerment”: Databases are moving from “passively carrying data” to “actively understanding data, organising knowledge, and assisting decision‑making,” significantly improving the accuracy, security, and continuity of industry large model operations.
Industry applications entering core system breakthrough phases: Industries such as finance, telecommunications, and energy are advancing full‑stack upgrades with centralised and distributed architectures running in parallel, tailored to different workloads including transaction processing, real‑time analytics, and time‑series data.

The report also notes that China’s database industry still faces challenges in fundamental theoretical research, inconsistent interface/migration tool standards, and insufficient international influence in the open‑source ecosystem.

DBA Perspective: The continued growth of the market and the AI empowerment trend mean that DBA skills must upgrade from “basic domestic database operations” to “multi‑modal data processing and mixed‑load tuning under AI‑native architectures.” Sustained policy support will also accelerate large‑scale Xinchuang implementation in key industries such as finance, telecommunications, and energy.

💡 SQL Little Knowledge Point

This Issue’s Knowledge Point: What is an “Agentic Data System”?

VLDB 2026 has listed “Data Systems for the AI Agent Era” as a core theme, making “Agentic Data System” a recurring conference keyword.

Dimension	Traditional Data System	Agentic Data System
Users	Humans (DBAs, developers, business users)	AI Agents (autonomous retrieval, reasoning, tool invocation)
Access Patterns	SQL queries (one‑off)	Multi‑round interaction, cross‑session memory, tool invocation chains
Data Types	Primarily structured	Vector, graph, multi‑modal, structured hybrid
Resource Management	Fixed resource pools	Dynamic elasticity, task state and memory persistence
Optimisation Goals	Single‑query latency, throughput	End‑to‑end task completion rate, agent decision quality

In simple terms, an Agentic Data System is not about “supporting AI” in the database, but about making the database the “memory and decision infrastructure” for AI agents – the complete closed loop of agent information retrieval, reasoning, tool invocation, and memory write‑back is all completed with the database as the foundation.

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

Segment 2: HiddenMerit In‑Depth · Issue 6

Focus on Database Frontiers, In‑Depth Analysis September 3, 2026 ｜ 1 In‑Depth Feature Article

When Workloads Are Rewritten by AI: The Database Architecture Signals from VLDB 2026

From August 31 to September 4, the 52nd International Conference on Very Large Databases (VLDB 2026) is being held in Boston. The signals from the conference floor are more direct than any single paper: database research is shifting from “performance optimisation” to “AI‑Ready architectural restructuring.”

1. From “Storing Fast” to “Being Understood by AI”

For a long time, the goal of databases was clear enough: store data reliably, query it fast, and withstand pressure during business peaks.

But AI is rewriting that proposition. New data objects – vectors, graphs, multi‑modal data – are entering production pipelines, and heterogeneous hardware like GPUs and RDMA are becoming system design variables. More critically, AI agents are emerging as a new type of persistent “user” – they retrieve, reason, invoke tools, and continuously write back task progress, memory, and context. Compared to a single SQL query, this is a longer, more dynamic, and harder‑to‑predict task chain.

VLDB 2026 has listed “Agentic Data Systems” as one of the conference’s core themes. This is not an academic “forward‑looking exploration” – it is a response to real production needs. When agents become the “new users” of databases, database access patterns, resource management, and optimisation goals all need to be redesigned.

2. The “GPU Moment” for Data Formats

Researcher Li Huan from Zhejiang University, speaking at Tencent Cloud DBTalk’s “Three Consecutive Top Conference Hits” event, raised a key observation: data formats are being restructured by GPUs.

Take the columnar storage format L3 as an example: in the GPU era, high compression ratio does not necessarily mean fast computation. If the traditional path of CPU‑side encoding, CPU‑GPU data movement, and decoding during execution is maintained, the benefits of compression are easily offset by data conversion and movement costs.

L3’s solution is to redesign GPU‑native data formats, transforming learned compression from offline CPU‑side processing into executable data formats that can directly participate in GPU analytical computation.

Similar transformations are also occurring in vector and graph data systems. Vector databases need to simultaneously consider retrieval, partitioning, load skew, and cross‑node communication. Compressed graph direct updates achieve online updates and query performance while maintaining compression effectiveness through “lightweight local modifications in the foreground, with background continuous compression rule organisation.”

[quads id="805"]

The implication for DBAs is: the boundaries of tuning are expanding from “database parameters” to “data format and hardware coordination.” In the future, DBAs will need to consider not just indexes and SQL structure, but also how data is represented on GPUs.

3. Resources Can Be Decoupled, But Elasticity Doesn’t Come Automatically

Cloud‑native databases continue to evolve along the lines of storage‑compute decoupling, resource pooling, and serverless data services. Persistent data and compute nodes are no longer strictly bound, and background tasks like memory, caching, and compression are gaining more independent resource boundaries.

But Li Huan highlighted an easily overlooked fact: just because resources can be decoupled doesn’t mean elasticity problems have disappeared.

Remote access brings network overhead; new nodes need to complete state recovery and topology convergence after joining; cache scaling affects hit rates and consistency. True elasticity cannot wait for load spikes to start remediating – it requires a closed loop of observation, prediction, decision, execution, and feedback.

Li Huan summarised the elasticity goals as “fast, stable, and cost‑efficient”:

Fast: Can resources be provisioned in time, and can running queries utilise newly added compute capacity?
Stable: Admission control, overload protection, and SLO guarantees under multi‑tenant, multi‑concurrency scenarios.
Cost‑efficient: Can resources be allocated and reused at fine granularity according to query needs?

The three cannot be optimised infinitely simultaneously – the system needs to find the right control points among response speed, stability margin, and resource utilisation. Agents make this problem more complex: compute resources can be reclaimed, but agent task progress, long‑term memory, execution branches, and permission context cannot be lost with them.

4. Agentic Data Systems: The Next Form of Databases

The thematic directions of multiple papers at VLDB 2026 reveal a common theme: databases are moving from “passive storage” to “actively participating in AI decision‑making chains.”

Professor Zhang Feng from Renmin University focused on the collaborative evolution of database kernels. In the past, system optimisation often targeted a specific algorithm or operator; now, research is beginning to follow the complete data path: how data is represented, how it is laid out, how it moves, where it executes, and how it remains stable under changing workloads.

As data types continue to increase and access paths continue to lengthen, the goals of database optimisation also shift – no longer just pursuing local optimality for individual modules, but driving cross‑layer coordination of data representation, execution engines, and heterogeneous hardware around real workloads.

5. Implications for DBAs

The signals from VLDB 2026 converge into three implications for DBAs:

1. Agents are the new users – databases need to redesign access patterns

Agent access patterns are completely different from human ones – high‑frequency small batches, cross‑session persistence, multi‑tenant logical isolation. Future databases need to redesign permission models, connection management, and audit trail mechanisms for agents.

2. GPUs are becoming a core variable in database architecture

Data formats, execution engines, and storage layouts all need to consider GPU characteristics. DBA tuning dimensions will expand from “CPU core count, memory allocation, disk I/O” to “GPU video memory management, parallel stream scheduling, and data residency strategies.”

3. Elasticity is not “automatic” – it requires closed‑loop governance

Resource pooling does not mean elasticity is automatically achieved. DBAs need to establish a closed‑loop mechanism of observation → prediction → decision → execution → feedback, rather than passively scaling after load spikes.

The “AI‑Ready” transformation of databases has begun. For DBAs, this is not a threat – it is an opportunity to redefine the value of the profession.