Replication overview
The WAL streaming model behind Silos replicas, its current maturity, and what to expect.
Silos keeps replicas current by streaming the database's write-ahead log (WAL) from the primary to replicas. Replicas apply those records to stay close to the primary's state, and each query chooses how fresh its data must be.
WAL streaming and edge replica hydration are rolling out — they are not yet functional end-to-end. The streaming code and consistency types are implemented and wired, but cross-region distribution is not production-proven, and the subscriber side does not yet fully apply records to replicas. Plan production reads against the primary region as the dependable baseline until this matures.
How replication works
- The primary owns writes and appends every change to the WAL.
- WAL records are streamed out toward replicas (the design uses a durable streaming layer for this transport).
- Each replica applies the WAL it receives to advance its own state.
- A query's consistency level decides whether a read may use a possibly-lagging replica, must wait for catch-up, or must hit the primary.
Sub-millisecond global synchronization is physically impossible — replicas in distant regions are always at least a network round-trip behind. Silos's answer is WAL streaming plus per-query consistency: you decide, per read, how much staleness you can tolerate. The design target for typical replication lag is ~50–200 ms (a target, not a measured guarantee).
What's in this section
Consistency levels
Eventual, read-your-writes (default), bounded-staleness, and strong — tunable per query.
Read-your-writes
How LSN tokens guarantee you see your own writes.
Failover
Primary failure handling — designed-for behavior and current status.
Monitoring
Replication lag and health via Prometheus-style metrics.
Current status at a glance
| Capability | Status |
|---|---|
| WAL streaming transport | Implemented and wired; not yet functional end-to-end |
| Replica apply (subscriber) | In progress — records received, full apply maturing |
| Consistency levels (types + routing) | Implemented |
| Read-your-writes via LSN tokens | Implemented |
| Automatic failover | Designed for, not guaranteed |
See When to use Silos for an honest fit assessment.