CONCEPT Cited by 1 source

Flush interval / RPO tradeoff

Definition

When an LSM-tree's SSTables and manifest are stored in object storage, the memtable flush interval determines how stale the remote representation can be relative to the in-memory state. This interval directly defines:

1. Recovery Point Objective (RPO): The maximum amount of metadata that could be lost in a catastrophic cluster failure (only data in the memtable since the last flush).
2. Read-replica staleness: External consumers reading from object storage can be at most one flush interval behind the source.
3. Write amplification: More frequent flushes increase object-storage writes and compaction pressure.

Redpanda's choice

The Cloud Topics metastore defaults to a 10-minute flush interval, meaning: - RPO ≤ 10 minutes of metadata - Read-replica lag ≤ 10 minutes - Described as "a comfortable balance between durability and write amplification" for their users

(Source: sources/2026-06-09-redpanda-cloud-topics-the-metastore)

Seen in

• systems/redpanda-cloud-topics-metastore — 10-minute default flush, defines RPO and read-replica lag bound