Timestamping after commit

Many applications need transaction-consistent pictures of past states of the database. These applications use the commit time of the transaction to timestamp the data. When a transaction is distributed, the cohorts must vote on a commit time and the coordinator must choose a commit time based on the votes of the cohorts. This implies that timestamps are applied after commit. Until the timestamps are on all the records, one must keep a table of all the committed transaction identifiers and their commit times. The main problem solved is that of determining when all timestamps corresponding to a given transaction have been placed in the records so that a committed transaction entry can be erased from the table. This information must be stable. Logging and recovery details are included.<
>