Disconnected and Partitioned Operation
Keep the application useful when its node loses contact with other nodes. Start by separating rebuildable local work from actions that require a shared decision.
| Work | What to plan for |
|---|---|
| Reading resident state | Local state may remain available; it cannot include updates the node has not received |
| Non-transactional writes | Writes can continue and reconcile after the link returns |
MULTI/EXEC on unreachable shared state | Transactions can fail; define retry or rejection behavior |
| Fetching absent state | A disconnected node may be unable to reach a holder or storage tier |
For a disconnected field application, keep the required working set available before losing connectivity. Treat external actions as retryable work with stable identifiers.
After recovery, check application outcomes and transaction errors as well as peer health. Do not assume that a restored link makes every interrupted request safe to retry.
The partition guide covers detection races and current recovery limits. Holographic divergence is not a public self-service deployment option.