REPLICATING DELETION INFORMATION USING SETS OF DELETED RECORD IDS

Deleted database records are identified not by the use of tombstones (pairs of record IDs and change numbers) but by storing for each database replica a set of deleted record IDs without associated change numbers (DRID data). The DRID data can include not only IDs of deleted records but also IDs that were never assigned to the database by the server (e.g., IDs assigned to records of other databases). The DRID data can be compactly stored as a list of intervals, with each interval describing a range of IDs across which records have either been deleted or never assigned to the database. Replication is accomplished by periodically sending the DRID data, or incremental versions thereof, to other replicas, which delete records identified by the received DRID data, and update their existing DRID data by combining it with received DRID data.