Synchronize inventory data

Use a single-flight or lease around each inventory refresh, cache successful snapshots, and apply bounded retry behaviour.

Prerequisites

Complete Build an inventory integration first. Decide which component owns scheduling, the maximum acceptable snapshot age, and whether your product may serve a labelled stale snapshot.

Freshness model

Define separate values for retrieval time, acceptable freshness, and maximum stale-serving duration. Publish a new snapshot only when the paginated read completes.

Coordinate refresh work

1. Derive a stable refresh key from the inventory identity.
2. Acquire a bounded lease or use a single-flight mechanism.
3. Retrieve all pages within one workflow deadline.
4. Publish the completed candidate atomically.
5. Release the lease and schedule the next attempt with jitter.

The coordination mechanism must expire after a crashed worker. A waiting request should use the current snapshot or return a clear in-progress state rather than starting unbounded duplicate work.

Fail without losing known data

Keep the last successful snapshot when a refresh is private, rate limited, unavailable, or incomplete. Record the refresh outcome separately. Retry only transient outcomes and stop when the workflow deadline or attempt cap is reached.

Track snapshot age, refresh duration, page count, outcome, and request IDs. Do not attach API keys or full inventory payloads to metrics, traces, or alerts. Review Retries & timeouts and Caching.