Hacker News Reader

I don’t agree with that characterization about availability, this is exactly what Spanner is designed for since replication can be cross-region for very good reasons (to improve latency, save on bandwidth, improve e2e availability) and you actually do want to maintain availability in the case of a network paritition between regions (which is a thing that actually happens)

View on HN

Spanner does not give you availability in the case of a network partition between regions.

I’m pretty sure it does give you read-availability, just not write availability?

I could be wrong but I think it’s: To serve a read from a replica in region X, where the write replica is in region Y s.t. there is a partition between X and Y, I’m pretty sure in the normal case X does not need to wait for Y to tell it to let the read go. Instead Y locks X on-write to implement consistency. So in the case of a partition X still does not wait for Y to execute a read, but writes become unavailable.

Reply on news.ycombinator.com