Expand description
Caller-owned, failure-aware buffering. Construction never starts work.
Records remain in memory until persistence acknowledges them. Sink errors pause the affected
lane; only explicitly retryable failures can be resumed. Dropping the owner cancels/aborts its
worker, while stop additionally joins it before dependencies may close.
Structs§
- Buffer
Stop Report - Drain result. Once returned, the worker is joined and database dependencies may close.
- Buffered
Batch Config - Bounded in-memory buffering policy. Zero capacities normalize to one.
- Buffered
Writer Snapshot - Observable per-instance state. Counts include in-flight, unacknowledged records.
- Managed
Batch Writer - Explicitly started and stopped buffering with one owned worker and bounded overflow.
- Write
Failure - A failure retained alongside the unconfirmed records.
Enums§
- Buffer
Flush Error - Flush failure; inspect the snapshot and unconfirmed records before recovery.
- Buffered
Writer Event - Low-cardinality events delivered with
try_send; persistence never calls observer code. - Buffered
Writer State - Writer lifecycle. A stopped writer cannot be restarted.
- Overflow
Policy - Explicit admission policy when the bounded batch queue is full.
- Record
Admission - Successful admission does not imply batch durability.
- Record
Error - Admission failure. Rejected records are returned; accepted failed direct writes stay in memory.
- Write
Failure Cause - Original persistence error, or a worker-side interruption with unknown outcome.
- Write
Failure Class - Whether an unacknowledged write can safely be replayed.