|< < 44 > >|

Checkpointing

Algorithm

  • Stop accepting new transactions.

  • Wait for all current transactions to log COMMIT or ABORT.

  • flush().

  • Write checkpoint() into the log.

  • Resume normal operation.

Recovery

  • Same recovery algorithm, but backward scan stops when checkpoint() is seen.

  • All updates before the checkpoint either committed or aborted.

|< < 44 > >|