An application serving users on three continents has a problem no single-region database can solve cleanly: write latency goes up for whoever is farthest from the primary, and a regional outage can take the whole app down. Multi-region, active-active-capable databases exist to fix this by replicating data — and often accepting writes — across regions. CockroachDB, Google Cloud Spanner, and YugabyteDB are the three most mature options, and they make genuinely different trade-offs to get there.
At a Glance
| Database | Replication model | Consistency | Starting cost basis |
|---|---|---|---|
| CockroachDB | Synchronous Raft consensus across regions by default | Serializable (strongest SQL isolation level) | Request Units: $0.20/1M RUs + $0.50/GiB storage (Basic tier) |
| Google Cloud Spanner | Synchronous, Paxos-based replication; true multi-region configs need Enterprise Plus edition | External consistency (strict serializability) | Node-hours: Enterprise Plus multi-region from $3.705/node-hour (on-demand) |
| YugabyteDB | Synchronous Raft (default) or asynchronous xCluster for cross-region active-active | Serializable with synchronous mode; “some data loss” possible on xCluster failover | Plan-based: $0.17/vCPU-hour + $0.10/GB-month storage (YugabyteDB Aeon) |
Short Verdict
All three can survive a region going dark without losing committed data, but they get there differently. Spanner’s multi-region configurations are the most “set it and forget it” if you’re already on Google Cloud and willing to pay Enterprise Plus pricing. CockroachDB gives you the most control over per-database region placement on a Postgres-compatible engine. YugabyteDB is the only one of the three that explicitly offers a true asynchronous active-active mode (xCluster) for teams who will accept some replication lag in exchange for lower cross-region write latency.
How Each One Actually Replicates Data
CockroachDB distributes data in ranges and replicates each range via Raft consensus. In a multi-region cluster, you choose a primary region, and by default data gets three replicas in that region and one in each additional region; two regions can survive a zone failure, and surviving a full regional outage requires at least three regions. Every write that touches a range requires a quorum of its replicas to acknowledge it, which is why cross-region writes get slower as regions spread farther apart — that’s the price of synchronous consistency.
Google Cloud Spanner only offers true multi-region and dual-region configurations on its Enterprise Plus edition; Standard and Enterprise editions are regional only. A multi-region instance runs 5 or more replicas per node and uses a Paxos-based consensus protocol to give external consistency — Google’s term for consistency that holds even across the uncertainty window of distributed clocks. Inter-region replication is billed separately, per GiB, based on which regions a write has to reach.
YugabyteDB gives you a choice the other two don’t: synchronous replication (the default, via Raft, with the same latency trade-off as CockroachDB and Spanner) or xCluster, an asynchronous replication mode that supports unidirectional and bidirectional (“active-active”) replication between regions or clusters. xCluster keeps writes local to a region for low latency, but accepts some data loss on an unplanned failover since replication isn’t synchronous — a materially different risk profile than the other two.
Pricing Breakdown
| Cost component | CockroachDB (Basic/Cloud) | Spanner (Enterprise Plus) | YugabyteDB Aeon |
|---|---|---|---|
| Compute | $0.20 per 1M Request Units | $1.71–$3.705/node-hour depending on config (on-demand) | $0.17/vCPU-hour |
| Storage | $0.50/GiB, prorated | ~$0.52/GiB-month (multi-region SSD) | $0.10/GB-month |
| Cross-region data transfer | Consumes RUs; cross-region reads flagged as an anti-pattern | $0.04–$0.17/GiB depending on source/destination continent | $0.01/GB same region, $0.02–$0.08/GB cross-region |
| Free tier | $15/month free usage (pay-as-you-go only) | 90-day free trial, 10 GiB, Standard edition only (no multi-region) | Not publicly listed on the pricing page reviewed |
Note: CockroachDB’s Request Unit model applies to its Basic serverless tier; its Standard and Advanced provisioned tiers use different, dedicated-vCPU pricing that Cockroach Labs doesn’t fully itemize on its public Basic-tier docs page, so budget accordingly if you need a provisioned multi-region cluster rather than serverless Basic.
Pros and Cons
- CockroachDB pros: Postgres wire-compatible, transparent range-based sharding, clear survival-goal documentation (zone vs. region failure)
- CockroachDB cons: Cross-region write latency is a hard physics constraint of its synchronous design; Basic tier RU limits can throttle a cluster
- Spanner pros: Deepest operational track record (runs Google’s own planet-scale services), strong consistency guarantees, integrates natively with Google Cloud IAM and networking
- Spanner cons: True multi-region requires the most expensive edition; locked into Google Cloud; moving an existing regional instance to multi-region requires an edition upgrade first
- YugabyteDB pros: Only one of the three with an explicit async active-active mode; Postgres (YSQL) and Cassandra (YCQL)-compatible APIs; can self-host or use managed Aeon
- YugabyteDB cons: xCluster’s asynchronous mode trades durability guarantees for latency — “some data loss” on failover is stated plainly in Yugabyte’s own docs; smaller managed-cloud footprint than the other two
Who Each One Fits
Pick Spanner if you’re already committed to Google Cloud and want the most battle-tested multi-region consistency model, and the Enterprise Plus pricing is acceptable. Pick CockroachDB if you want Postgres compatibility and fine-grained control over where data lives, and can tolerate synchronous cross-region latency. Pick YugabyteDB if you specifically need a region to keep serving fast local writes during a cross-region network partition and can tolerate the small window of potential data loss that async replication implies.
Bottom Line
There’s no free lunch in distributed databases: synchronous replication (Spanner, CockroachDB’s default, Yugabyte’s default) buys you zero data loss at the cost of cross-region write latency; asynchronous replication (Yugabyte’s xCluster) buys you lower latency at the cost of a small failover risk window. The right choice depends on which of those two costs your application can least afford.
FAQ
Is any of these truly “active-active” in the sense of accepting writes to the same row from two regions simultaneously with no coordination?
Not without trade-offs. CockroachDB and Spanner’s synchronous models mean any write is coordinated via consensus regardless of which region it originates from. YugabyteDB’s bidirectional xCluster is the closest to classic active-active, but it uses asynchronous replication and last-writer-wins-style conflict handling rather than true multi-master consensus on every write.
Can I move a single-region Spanner instance to multi-region later?
Yes, but you must first upgrade it to the Enterprise Plus edition before changing the instance configuration to dual-region or multi-region.
Which one is cheapest?
It depends entirely on workload shape — Request-Unit and vCPU-hour pricing models aren’t directly comparable without running your own query patterns through each vendor’s calculator. Don’t assume list prices translate linearly across models.
