Struct ReplayOutcome 

pub struct ReplayOutcome {
    pub committed_events: Vec<MutationEvent>,
    pub max_lsn: Lsn,
    pub torn_tail: Option<TornTailInfo>,
    pub checkpoint_lsn_observed: Option<Lsn>,
}

Outcome of a full replay walk.

Fields

committed_events: Vec<MutationEvent>

Mutation events from committed transactions, in append order. Apply these to a fresh store (or one freshly loaded from a snapshot at checkpoint_lsn) to reproduce the pre-crash state.

max_lsn: Lsn

Highest LSN observed in any segment, regardless of whether the owning transaction committed. Used to seed next_lsn for new appends so we never reuse an already-allocated id.

torn_tail: Option<TornTailInfo>

Torn-tail diagnostic. Some iff a record failed to decode before the natural end-of-log. The caller must truncate the affected segment to last_good_offset before resuming appends, otherwise replay-after-replay will keep tripping the same CRC.

checkpoint_lsn_observed: Option<Lsn>

Newest checkpoint LSN observed in a [WalRecord::Checkpoint] marker.

Informational only. The recovery contract today is “the snapshot’s wal_lsn is the replay fence” — if the operator passes an older snapshot than the newest checkpoint marker on disk we still replay every record above the snapshot’s fence, which is conservative-correct (the only cost is duplicated work). A tighter “marker overrides snapshot” contract is deferred to v2 because it would require us to know that the marker’s snapshot file actually exists where the operator can reach it, which is a separate observability concern.

Surfaced so callers (e.g. lora-database::Database::recover) can log a warning when the snapshot is older than the newest observed marker.

Trait Implementations

impl Debug for ReplayOutcome

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.