Nano 1.3 Batched Sync Flush: Memory and Latency Trade-offs

Flush Behavior Transition

Nano 1.3 introduces batched sync flush, moving away from the asynchronous flush model used in version 1.2. This change alters how entries are committed to storage, specifically impacting the relationship between batch size and write latency.

Configuration Constraints

The default batch size is 1000 entries. While increasing this limit to 5000 entries can reduce per-entry overhead and overall flush frequency, there is a risk of increased memory pressure on resource-constrained devices, potentially leading to out-of-memory (OOM) conditions.

When transitioning to Nano 1.3, it is unclear how to balance the reduction in latency against the increased memory footprint without a standardized baseline for flush time per entry.

• What is the expected memory overhead increase when scaling the batch size from 1000 to 5000 entries?
• Does the batched sync flush in 1.3 maintain consistent latency across varying entry sizes?