Struct TransactionManager 

pub struct TransactionManager { /* private fields */ }

Manages all transactions in the database

Implementations

impl TransactionManager

pub fn new() -> Self

Create a new transaction manager

For new databases, use this constructor. For restoring from a persisted database, use TransactionManager::with_start_txn_id() instead.

pub fn with_start_txn_id(start_txn_id: u64) -> Self

Create a new transaction manager with a specific starting transaction ID.

Used when restoring from a persisted database to maintain MVCC visibility. The start_txn_id should be the value persisted from the previous session, ensuring transaction IDs remain monotonically increasing across restarts.

For new databases, use TransactionManager::new() instead.

pub fn next_txn_id(&self) -> u64

Get the current next transaction ID (for persistence)

This value should be persisted to ensure transaction IDs remain monotonically increasing across database restarts.

pub fn begin(&self, mode: TransactionMode) -> Transaction<'_>

Begin a new transaction with default configuration

pub fn begin_with_config( &self, mode: TransactionMode, config: TransactionConfig, ) -> Transaction<'_>

Begin a new transaction with specific configuration

pub fn begin_read_only(&self) -> Transaction<'_>

Begin a read-only transaction

pub fn begin_read_write(&self) -> Transaction<'_>

Begin a read-write transaction

pub fn commit_txn(&self, txn_id: TransactionId) -> Result<()>

Commit a transaction (public API without WAL)

pub fn abort_txn(&self, txn_id: TransactionId) -> Result<()>

Abort a transaction (public API without WAL)

pub fn add_snapshot_ancestor( &self, txn_id: TransactionId, ancestor: TransactionId, )

Record a write operation for conflict detection (first-committer-wins).

Called by the DML executor for each row modified by INSERT, UPDATE, or DELETE. The (table_name, pk_bytes) pair is added to the transaction’s write set, which is checked against committed write sets at commit time. Add an ancestor to a transaction’s snapshot (for savepoint support).

When a savepoint creates a sub-transaction, the sub-transaction needs to recognize the parent transaction as an ancestor so that write-write conflict detection doesn’t flag writes from the same logical transaction.

pub fn record_write( &self, txn_id: TransactionId, table: String, pk_bytes: Vec<u8>, ) -> Result<()>

Record a write and check for conflicts (first-committer-wins).

Returns Err(WriteConflict) if another transaction has already written to the same (table, pk) — either committed after our snapshot or still active.

pub fn acquire_write_lock(&self) -> impl Drop + '_

Acquire write lock for exclusive writing

pub fn oldest_active_txn(&self) -> Option<TransactionId>

Get the oldest active transaction ID

pub fn version_store(&self) -> &RwLock<VersionStore>

Get the version store

pub fn active_count(&self) -> usize

Get number of active transactions

pub fn has_aborted_transactions(&self) -> bool

Quick check whether any aborted transactions exist. When false, all is_aborted calls are guaranteed to return false.

pub fn is_aborted(&self, txn_id: TransactionId) -> bool

Check if a transaction was aborted

pub fn is_row_visible( &self, snapshot: &Snapshot, created_by: TransactionId, deleted_by: Option<TransactionId>, ) -> bool

Check if a row is visible to a snapshot, accounting for aborted transactions.

A row is visible if:

1. The creating transaction was NOT aborted
2. The snapshot can see the creating transaction (per standard MVCC rules)
3. Either the row is not deleted, the deleting transaction was aborted, or the snapshot cannot see the deletion

pub fn aborted_count(&self) -> usize

Get the number of aborted transactions being tracked

pub fn cleanup_aborted_txns(&self, older_than: TransactionId)

Clean up old aborted transaction IDs that are no longer needed.

Aborted transaction IDs can be removed once all their written data has been garbage collected. This is typically safe to do when the aborted transaction ID is older than any active transaction’s snapshot, since all such data would have been cleaned up by GC.

pub fn register_dirty_page(&self, txn_id: TransactionId, page_id: PageId)

Register a page as dirtied by a transaction

This is called when a transaction modifies a page. The page ID is tracked so that on commit, only this transaction’s pages are flushed, and on abort, they can be discarded.

pub fn get_dirty_pages(&self, txn_id: TransactionId) -> HashSet<PageId>

Get all pages dirtied by a transaction

pub fn clear_dirty_pages(&self, txn_id: TransactionId)

Clear dirty page tracking for a transaction (called on commit/abort)

pub fn dirty_page_count(&self, txn_id: TransactionId) -> usize

Get the number of dirty pages for a transaction

pub fn run_gc(&self) -> GcStats

Run garbage collection on old versions.

Removes old versions that no active transaction can see. Returns statistics about what was collected.

pub fn version_count(&self) -> usize

Get the number of old versions in the version store

pub fn long_running_transactions( &self, threshold: Duration, ) -> Vec<TransactionInfo>

Get information about long-running transactions

Returns a list of transactions that have been running longer than the specified threshold, sorted by age (oldest first).

pub fn gc_status(&self) -> GcStatus

Get current GC status

Returns information about whether garbage collection is blocked by active transactions.

pub fn metrics(&self) -> TransactionMetricsSnapshot

Get transaction metrics snapshot

Returns a snapshot of transaction metrics including active count, commits, rollbacks, and average transaction duration.

Trait Implementations

impl Default for TransactionManager

fn default() -> Self

Returns the “default value” for a type.