Description

A transaction is given a unique ID number which

A transaction which has been committed is given a unique integer transaction_id, which in turn can be used to mark the tables to indicate the last committed transaction. This value is used in snapshots and logs, and effectively indicates the total number of transactions which have been committed against a particular database. This number can't loop around safely, so a 64-bit int is used.

A transaction which is in progress is identified by an integer which I've decided to call a lock_id. This is because it is most typically used to mark rows in shared data which the transaction has locked. The "lock_id" is an id assigned to a transaction when it is started, and a 'transaction_id' is an id which is assigned when the transaction commits. The reason for the two identifiers is because transactions can start and commit in different orders, and it would complicate the tracking of committed transactions if one one ID was used.