CONCEPT Cited by 1 source

Manifest replication via Raft

Definition

In an LSM-tree database, the manifest tracks which SSTables exist and their key ranges. Traditionally it's a local file rewritten on each flush. Manifest replication via Raft writes the manifest to durable storage (e.g., object storage) on each flush AND replicates it to all Raft-group followers, so every replica immediately knows the current SSTable layout.

Why it matters

1. Fast failover: A new leader doesn't need to read the manifest from remote storage — it already has a local copy from the Raft replication.
2. Log truncation signal: The same Raft entry that carries the manifest update marks a truncation point — all WAL entries already captured in the manifest can be trimmed from the Raft log.
3. External readability: Because the manifest also goes to object storage, external systems (disaster recovery, read replicas) can bootstrap without connecting to the cluster.

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

Seen in

• systems/redpanda-cloud-topics-metastore — manifest written to object storage + Raft-replicated to followers on every flush