Fusion upsert idempotency during internal retry cycles
0 reputation · 18 Sept 2020, 00:20 UTC
Upsert Operation and Idempotency
Fusion provides an upsert operation designed to prevent duplicate records by utilizing a user-defined unique key. This mechanism ensures that retried requests update existing entries rather than creating redundant rows in the target store.
Retry Behavior Uncertainty
While the upsert API handles idempotency based on the provided key, the behavior of Fusion's internal retry mechanism during transient network failures is not fully documented. There is uncertainty regarding whether a failed request triggers a repeat of the original upsert call or if the system falls back to a sequential insert-then-update logic.
This distinction is critical for maintaining exactly-once semantics, as a fallback sequence could potentially introduce race conditions or duplicate entries if the initial write partially succeeded.
- Does the internal retry mechanism maintain the idempotency of the original upsert request?
- Is there a configurable idempotency key separate from the record's unique key to manage these retries?