I've been knee-deep in evaluating managed database offerings for a new multi-region application, and Google's recent announcement about Spanner's regional configurations caught my eye. For years, the conversation around Spanner started with "globally distributed, strong consistency, but the cost..." The new regional single-region and multi-region configurations seem like a direct attempt to address that last part, positioning it more squarely against the likes of AWS Aurora or Azure Cosmos DB with a single region write.
My initial take is that this is a significant move for GCP, making Spanner a more practical option for workloads that don't *need* global writes but do need that proven horizontal scale and the strong consistency model. The pricing drop for the regional instance is substantial. However, I'm trying to piece together the real-world trade-offs beyond the headline cost-per-operation.
Here’s what I’m currently weighing in my vendor evaluation:
* **Resiliency vs. Cost:** The regional multi-region config (e.g., `nam3`) offers read-write replicas across three zones in one continent. This is fantastic for zone-level resilience within a region, but how does the latency and cost compare to a traditional multi-AZ deployment in another cloud? The documentation suggests lower cost than a true global config, but is it competitive with, say, a Multi-AZ Aurora Provisioned instance for the same throughput?
* **The "No Planned Downtime" Promise:** This is a key Spanner differentiator, even in regional mode. For our compliance needs, operations like OS updates without downtime are a huge vendor risk reduction. Has anyone run a regional Spanner instance through a maintenance event yet? Any observable impact?
* **Workload Fit:** This seems ideal for:
* Applications expecting to outgrow a single PostgreSQL or MySQL instance but wary of the sharding complexity.
* Systems where strong consistency is non-negotiable, and you previously ruled out global Spanner on cost.
* Migrations off of legacy Oracle RAC where the consistency model is similar.
My main question for the community: **Are you considering these new configs for workloads you'd previously have placed elsewhere?** I'm particularly interested in any early latency benchmarks or cost-per-GiB-storage comparisons you might have seen against Azure Cosmos DB's single-write-region mode or Amazon Aurora. The pricing transparency on the GCP side is clearer now, but the true total cost of ownership for a mid-tier workload still feels like it needs some real-world data points.
— frank
buyer beware, but buy smart