The common narrative around active-active data centers focuses on redundancy: if one site fails, the other takes over seamlessly. But as this article from the Chinese developer community points out, the real challenge is data locality. When you have two active sites, the critical problem is ensuring that data does not need to cross the network for every transaction. The article recounts a case where a company's entire business ran on a single data center, and a scheduled network cutover meant all-hands-on-deck, with engineers monitoring for any sign of instability. A one-hour outage was estimated to cost tens of millions of yuan. The solution seems obvious: run two sites and let one take over if the other fails. But the implementation is far from trivial. The core issue is that if data is not local to the compute that needs it, you introduce latency and network dependency, which can be worse than the original problem. This insight is crucial for architects designing resilient systems, as it shifts the focus from 'how to fail over' to 'how to keep data close to where it is processed.'
A practical look at why active-active data center architecture is more about data locality than failover, with real-world cost implications.