Trait TransactionLog 

pub trait TransactionLog:
    Send
    + Sync
    + Debug {
    // Required methods
    fn append<'life0, 'async_trait>(
        &'life0 self,
        entry: LogEntry,
    ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait;
    fn read_after<'life0, 'async_trait>(
        &'life0 self,
        after: HlcTimestamp,
        limit: usize,
    ) -> Pin<Box<dyn Future<Output = Result<Vec<LogEntry>>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait;
    fn sweep<'life0, 'async_trait>(
        &'life0 self,
        policy: RetentionPolicy,
    ) -> Pin<Box<dyn Future<Output = Result<usize>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait;

    // Provided method
    fn append_batch<'life0, 'async_trait>(
        &'life0 self,
        entries: Vec<LogEntry>,
    ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>
       where Self: 'async_trait,
             'life0: 'async_trait { ... }
}

Per-deployment append-only log of every committed write.

One instance per deployment process; lives for the deployment’s lifetime. Entries are HLC-ordered and survive process restart (durability is the foundational contract — replication’s catch- up window relies on it).

Contract

• append is the durability sink. Returns only after the entry has been fsynced (or the implementation’s chosen durability point); subsequent read_after calls on the same instance must see the new entry.
• read_after returns entries strictly greater than the given HLC, in ascending HLC order, up to limit. Used by replication for streaming tail-reads and catch-up on peer reconnection.
• sweep removes entries that exceed the retention bounds. Returns the count of removed entries. Sweep is the only operation that DELETES from the log — append is the only operation that ADDS.

Implementations

The canonical implementation is yeti_store::transaction_log::KvTransactionLog, backed by a dedicated KvBackend opened at {root_directory}/logs/transactions/{deployment_hash}/. Tests use the in-memory MemoryBackend under the same wrapper.

Required Methods

fn append<'life0, 'async_trait>( &'life0 self, entry: LogEntry, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Append a new entry. The entry’s hlc is assigned by the caller (typically from a HybridLogicalClock::now call at commit time). The log persists the entry verbatim.

Errors

• crate::error::YetiError::Internal if encoding fails.
• Whatever the underlying storage backend returns from a write.

fn read_after<'life0, 'async_trait>( &'life0 self, after: HlcTimestamp, limit: usize, ) -> Pin<Box<dyn Future<Output = Result<Vec<LogEntry>>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Read up to limit entries with hlc > after, in ascending HLC order. after = HlcTimestamp::ZERO reads from the beginning of the log.

Errors

• crate::error::YetiError::Internal if decoding any entry fails (indicates on-disk corruption or version skew).
• Whatever the underlying storage backend returns from a scan.

fn sweep<'life0, 'async_trait>( &'life0 self, policy: RetentionPolicy, ) -> Pin<Box<dyn Future<Output = Result<usize>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Remove entries that exceed the retention policy. Returns the number of entries removed.

Errors

Whatever the underlying storage backend returns from scan + batch-delete operations.

Provided Methods

fn append_batch<'life0, 'async_trait>( &'life0 self, entries: Vec<LogEntry>, ) -> Pin<Box<dyn Future<Output = Result<()>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Append many entries in one call. Implementations should fuse the appends into a single batched backend write where possible — the default implementation falls back to a per-entry loop and is correct but pays one awaited backend call per entry.

Used by LoggingBackend::write_batch so a 100-record table batch produces one storage write per table + one batched write on the log, instead of 100 serial awaited log puts.

Errors

• crate::error::YetiError::Internal if encoding any entry fails.
• Whatever the underlying storage backend returns from a write.

Dyn Compatibility

This trait is dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety".

Implementors