Monitoring replication

To rely on replicas safely, you need to know how far behind they are. Silos exposes Prometheus-style metrics so you can watch replication lag and replica health and alert when a replica drifts too far from the primary.

What to watch

The single most important signal is replication lag — how far a replica trails the primary, measured by the gap between the primary's latest WAL position and the position the replica has applied.

When lag is low, eventual and read-your-writes reads from that replica are fresh. When lag spikes, read-your-writes and strong consistency protect correctness by waiting for catch-up or routing to the primary — but high lag is still worth alerting on, because it signals a streaming or capacity problem.

How to consume the metrics

Metrics are exposed in a Prometheus-compatible format, so you can scrape them with Prometheus and visualize or alert with your existing stack (Grafana, Alertmanager, or any Prometheus-compatible tooling). A typical setup:

• Scrape the metrics endpoint with Prometheus.
• Graph replication lag per replica/region.
• Alert when lag exceeds your tolerance (tie the threshold to the staleness your app accepts — see Consistency levels).

Using lag in your decisions

• If a replica's lag is within your app's tolerance, prefer local reads there.
• If lag is high, prefer strong reads (primary) for correctness-sensitive paths until the replica recovers.
• Persistent high lag in a region usually points to the streaming path or replica capacity — investigate rather than masking it with stronger consistency everywhere.