Edge buffering & offline recovery

Design for a connection that eventually comes back.

Proposed store-and-forward service

An edge connector can persist accepted local events before attempting upstream delivery. Use a bounded durable queue, stable event identity and explicit retention. Design disk exhaustion as an expected failure mode.

Recovery rules

• Preserve measurement and enqueue times separately.
• Retry with backoff and jitter; cap concurrent sends.
• Require upstream acknowledgement before clearing a queued item.
• Deduplicate upstream because acknowledgement can be lost.
• Define which data may expire and which requires manual intervention.

Device constraints

Not every gateway stores application messages or exposes the same interfaces. Confirm the gateway operating system, available storage, network-server topology and supported integration path. An edge service must not be sold as a universal plug-in.

Acceptance test

Disconnect the upstream link in a controlled environment, create representative events, restart the edge process and restore connectivity. Verify bounded storage, preserved identities and one business effect per event.