Edge Database overview
How Silos materializes read replicas close to your users while writes forward to the primary.
Silos can serve reads from on-demand replicas that materialize in regions near your users, while all writes go to a single primary. The goal is local-latency reads globally without you running and operating a multi-region cluster by hand.
The model in one picture
- A database has one primary that owns writes and is the source of truth.
- When traffic arrives from a region without a local copy, Silos can materialize a replica there: it hydrates from object storage and then follows the write-ahead log (WAL) to stay current.
- Reads are served from the nearest healthy replica (subject to the consistency level you choose). Writes are forwarded to the primary.
- Routing to the nearest region is handled by GeoDNS.
Edge replicas and global WAL streaming are rolling out. The materialization, hydration, and consistency-routing mechanisms are implemented, but multi-region production availability — and end-to-end WAL streaming between regions — is still maturing. Don't depend on cross-region replicas for a production SLA yet. See the Replication overview for current status.
What's in this section
Replica materialization
How a replica is created on demand, hydrated from storage, and kept current.
Region selection & routing
How GeoDNS routes a client to a nearby region.
Read & write paths
Reads served locally; writes forwarded to the primary.
Latency expectations
Design targets for cold start and edge reads — what to expect and what not to assume.
Why edge reads
A traditional single-region database makes distant users pay the round-trip to that region on every query. Putting a read replica close to the user removes that round-trip for reads, which is the dominant operation for most read-heavy apps. The trade-off is staleness: a replica may lag the primary slightly. Silos makes that trade-off explicit and tunable per query — see Consistency levels.
Regions
Silos is rolling out across a growing set of regions. A small canonical set (roughly ten regions across North America and Europe) is the live baseline, and the network is expanding toward a 100+ region target.
"100+ regions" is a rollout target, not the current live footprint. The console's region list reflects the platform's planned topology; for the regions available to a specific database, use the CLI or API rather than assuming global coverage.
Where to go next
- New to the architecture? Start with How it works.
- Need to understand freshness guarantees? Read Consistency levels and Read-your-writes.