11 Database Synchronization

Normally, GDBM functions don’t flush changed data to the disk immediately after a change. This allows for faster writing of databases at the risk of having a corrupted database if the application terminates in an abnormal fashion. The following function allows the programmer to make sure the disk version of the database has been completely updated with all changes to the current time.

gdbm interface: int gdbm_sync (GDBM_FILE dbf)

Synchronizes the changes in dbf with its disk file. The parameter is a pointer returned by gdbm_open.

This function would usually be called after a complete set of changes have been made to the database and before some long waiting time. This set of changes should preserve application-level invariants. In other words, call gdbm_sync only when the database is in a consistent state with regard to the application logic, a state from which you are willing and able to recover. You can think about all database operations between two consecutive gdbm_sync calls as constituting a single transaction. See Synchronizing the Database, for a detailed discussion about how to properly select the synchronization points.

The gdbm_close function automatically calls the equivalent of gdbm_sync so no call is needed if the database is to be closed immediately after the set of changes have been made.

Gdbm_sync returns 0 on success. On error, it sets gdbm_errno and system errno variables to the codes describing the error and returns -1.

Opening the database with GDBM_SYNC flag ensures that gdbm_sync function will be called after each change, thereby flushing the changes to disk immediately. You are advised against using this flag, however, because it incurs a severe performance penalty, while giving only a moderate guarantee that the structural consistency of the database will be preserved in case of failure, and that only unless the failure occurs while being in the fsync call. For the ways to ensure proper logical consistency of the database, see Crash Tolerance.