LMDB checkpoints of the memory state

The memory state is saved to an LMDB file four times a second. No client waits for it: it decides how much log a restart replays. Every value carries a checksum.

Updated Oct 24, 2026View as Markdown

The state in memory is saved to an LMDB file, store/data.mdb, four times a second. This is a checkpoint, and no client waits for it: it only decides how much of the log a restart has to replay.

Every value carries a checksum, and the node checks all of them at boot. Every 6 hours a background scan reads the whole store again, to find damage before a restart does.

Traces have their own LMDB file, traces/data.mdb.

Source of truth: server/src/rsm/qlog/mod.rs
server/src/rsm/qlog/set.rs
server/src/rsm/qlog/dirx.rs
server/src/rsm/qlog/index.rs
server/src/rsm/qlog/record.rs
server/src/rsm/qlog/codec.rs
server/src/rsm/store/mod.rs
server/src/rsm/store/heed_store.rs
server/src/rsm/store/integrity.rs
server/src/rsm/apply.rs
server/src/rsm/scrub.rs
server/src/rsm/planner/txn.rs
server/src/rsm/maintenance.rs
server/src/rsm/retention_scan.rs
server/src/rsm/replicator/raft/mod.rs
server/src/rsm/replicator/raft/log_store.rs
server/src/rsm/replicator/raft/snapshot.rs

Edit this page