Zhejiang Mobile’s PostgreSQL Migration: Real‑World Challenges and Lessons Learned

This article compiles the interactive Q&A from the “Efficient Operations” WeChat group’s Operations Forum, where Zhejiang Mobile shared its PostgreSQL migration experience.

Participants

Gai Guoqiang @ Yunhe Enmo

Gong Yong @ Foreign‑owned Auto Plant – Shanghai

Chen Yang @ Youbao – Beijing

Li Zhuping @ Taoshi Cloud Union – Beijing

Editors

Li Ning @ TaOle‑Deep (content collection & organization)

Dong Wei, Xiao Tianguo (review & publishing)

Guest Introduction

Wang Xiaozheng, Assistant General Manager of the IT Department at China Mobile Zhejiang, led the session. He oversees application operations and platform architecture, graduated from Zhejiang University in 1997, and has experience as a developer, system administrator, and DBA. Additional guest: Tang Renjie, architect at Zhejiang Mobile.

Question 1: How far has Zhejiang Mobile’s PostgreSQL migration progressed and what pitfalls have been encountered?

Answer: The migration is divided into three phases—analysis of usable databases, small‑scale production rollout, and full de‑O (de‑IOE) implementation. The project is currently in the second phase, using PostgreSQL‑XC for distributed scalability. The biggest issue so far is occasional data inconsistencies between nodes, identified as a bug that is being jointly addressed by China Mobile’s Suzhou R&D center, Huawei, Tencent, and the PostgreSQL community.

Question 2: Our Oracle instance is small with many materialized views and stored procedures—must we refactor everything to move to PostgreSQL?

Answer: Materialized views need to be rebuilt; stored procedures can be migrated with minor modifications. The recommendation is to refactor completely, letting the database handle what it does best, and leverage ORM frameworks for a smoother transition.

Question 3: Is the de‑O effort driven only by the IT department, or does the business have a strong incentive?

Answer: The initiative was not originally a top‑down decision; it emerged from external pressure and the need to demonstrate technical competence. The team seeks to voice its perspective and has been working on de‑O solutions even before the “IOE” terminology became popular.

Question 4: Should we consider non‑relational databases?

Answer: Relational databases remain the most versatile choice for OLTP workloads, though specific domains may benefit from NewSQL, NoSQL, or other specialized systems. A balanced approach using different databases for different scenarios is recommended.

OldSQL, NewSQL, NoSQL each have their own use cases; de‑O is more about selecting the right tool for each scenario rather than abandoning relational databases entirely.

Question 5: Has de‑IOE delivered economic benefits for Zhejiang Mobile?

Answer: The effort is nearing completion, with a noticeable reduction in total cost of ownership (TCO) for servers, though exact figures are not disclosed.

Question 6: If we still need Oracle, what application‑level constraints should we be aware of?

Answer: The team imposes minimal constraints—SQL must conform to the SQL‑92 standard, avoiding Oracle‑specific functions that are not portable.

They have established a DBA team and a data‑management team to guide and review all SQL code.

Question 7: Can PostgreSQL build an ecosystem comparable to Oracle’s?

Answer: It is possible; Oracle’s ecosystem succeeded due to product maturity, open knowledge sharing, strong partner collaboration, and extensive training. PostgreSQL can follow a similar path by fostering community partnerships and practical adoption.

1. Product maturity and stability 2. Open knowledge base 3. Active third‑party partner ecosystem 4. Effective training and user community

Both MySQL and PostgreSQL need to learn from these factors to develop a thriving ecosystem.

Question 8: What are the main shortcomings of PostgreSQL compared to Oracle?

Answer: Challenges include less comprehensive monitoring tools, slower bug‑fix turnaround for open‑source issues, and less mature disaster‑recovery solutions, particularly the lack of heterogeneous replication or synchronization tools for PostgreSQL.

—‑ Interaction End —‑

How to Grow Together

This new era invites every voice to be heard and respected.

The “Efficient Operations” WeChat groups, founded in April 2015, now host over 800 senior operations professionals from diverse industries, serving as a benchmark for vertical social networking in the ops field.

Follow the official “Efficient Operations” public account for weekly original articles, event highlights, and community insights.

For those interested in joining the discussion groups, contact Liu Yuqiang (WeChat ID: kidult0314).

Important: Please obtain authorization before reproducing this article; full attribution and the QR code must be included.