Short answer: when LAN or cloud connectivity fails, furnace data acquisition should continue locally where the project requires continuity. The system should mark the outage, buffer records with stable timestamps/identity, show backlog health and forward them after recovery without duplicates or silent gaps.
A remote dashboard is a consumer of data, not the furnace's memory. If a WAN connection is in the critical path between sensor and durable storage, an internet outage can become a permanent hole in the batch record.
Separate four states
| State | What the operator should know |
|---|---|
| Measurement healthy, remote link healthy | Local and remote views are current |
| Measurement healthy, remote link down | Furnace record continues locally; remote view is stale |
| Measurement fault, network healthy | Remote connection does not make the source data valid |
| Local recorder/storage fault | High-priority data-integrity condition requiring action |
Conflating these states creates bad alarms. “Cloud unavailable” should not claim that the furnace stopped logging if the local store is healthy, and a green cloud icon should not hide an invalid channel.
Design store-and-forward explicitly
- Assign each record a source identity and sequence or collision-resistant key.
- Write locally before or independently of remote acknowledgement.
- Track the last confirmed forwarded record and the oldest pending record.
- Retry with bounded backoff; do not reorder or overwrite newer records.
- Make the backlog, storage headroom and last successful contact visible.
- After recovery, reconcile counts and expose any irrecoverable gap.
Time remains important during an outage
Use a defined time source and store source timestamps. If the clock becomes uncertain or changes, record the event; do not silently rewrite a sequence to make the graph look tidy. Recovery logic should not use arrival time as though it were measurement time.
Test failure, not only connection
- Disconnect the network for longer than a brief cable pull.
- Restart the remote service while records are pending.
- Fill the local buffer toward its warning threshold.
- Create a duplicate acknowledgement or out-of-order retry.
- Verify the report before, during and after the outage.
Data hosting, support access, retention and ownership are project-agreement matters. Pyrolog can support local, LAN, managed cloud or hybrid architectures; cloud is a capability, not a mandatory subscription model.