Lost write detection and repair

A computer-implemented method comprising:

• reading, from a redo log to which changes made to blocks in a primary database are recorded, a log entry that indicates a change to a first block in the primary database;
• wherein the log entry includes data representing (a) a current version number of the first block, (b) a previous version number of the first block, and (c) an identifier of the first block;
• determining whether the previous version number indicated in the log entry is less than, equal to, or greater than a particular version number indicated in a second block that is (a) in a standby database for the primary database and (b) indicates an identifier that matches the identifier in the log entry;
• if the previous version number indicated in the log entry is greater than the particular version number, then overwriting the second block in the standby database with data from the first block in the primary database;
• if the previous version number indicated in the log entry is less than the particular version number, then overwriting the first block in the primary database with data from the second block in the standby database;
• and
• if the previous version number indicated in the log entry is equal to the particular version number, then maintaining the first block and the second block in the present states of the first block and the second block;
• wherein the standby database serves as a replica of the primary database;
• wherein, in an event that the primary database becomes inaccessible, the standby database is substituted for the primary database to allow access of data replicated in the standby database to continue even though the primary database is inaccessible, and
• where in the steps of reading, determining, overwriting, and maintaining are performed by a computing device.

A method comprising the computer-implemented steps of:

• reading, from a read log that is updated when data units are read, a first version indicator that indicates a current version of a first instance of a data unit, wherein the first version indicator was written into the read log in response to the first instance of the data unit being read;
• reading a second version indicator that indicates the version of a second instance of the data unit;
• determining that the first version indicator does not indicate the same version of the data unit as the second version indicator;
• and
• in response to determining that the first version indicator does not indicate the same version as the second version indicator, then (a) indicating that a write operation made by an I/O device in an I/O subsystem has not been completed, even though the I/O subsystem reports the write operation has been completed, and (b) performing one or more specified actions relative to at least one of the first instance of the data unit and the second instance of the data unit,
• wherein the steps of reading, determining, and performing are performed by a computing device.