11.12. Managing Deleted Entries with Replication

When an entry is deleted, it is not immediately removed from the database. Rather, it is converted into a tombstone entry, a kind of backup entry that is used by servers in replication to resolve conflicts. The tombstone entry preserves state information about the original entry.
If there is ever a replication conflict, then the supplier uses the replica ID (the server where the change was initiated) and the timestamp of the change in the change sequence number to resolve the conflict. The oldest change wins.
As with deleted entries, deleted attributes are also kept in tombstone entries.
Tombstones are not preserved indefinitely. A purge job is run periodically, at a specified interval (set in the nsDS5ReplicaTombstonePurgeInterval attribute); the purge removes old tombstone entries. Tombstone entries are saved for a given amount of time (set in the nsDS5ReplicaPurgeDelay attribute); one a tombstone entry is older than the delay period, it is reaped at the next purge job.
Both the purge delay and the purge interval are set on the replica entry for a supplier server in the cn=replica,cn=replicated suffix,cn=mapping tree,cn=config configuration entry.
There are two considerations when defining the purge settings for replication:
  • The purge operation is time-consuming, especially if the server handles a lot of delete operations. Do not set the purge interval too low or it could consume too many server resources and affect performance.
  • Suppliers use change information, including tombstone entries, to prime replication after initialization. There should be enough of a backlog of changes to effectively re-initialize consumers and to resolve replication conflicts. Do not set the purge delay (the age of tombstone entries) too low or you could lose information required to resolve replication conflicts.
    Set the purge delay so that it is slightly longer than the longest replication schedule in the replication topology. For example, if the longest replication interval is 24 hours, keep tombstone entries around for 25 hours. This ensures that there is enough change history to initialize consumers and prevent the data stored in different suppliers from diverging.
To change the purge settings: 
[root@server ~]# ldapmodify -D "cn=directory manager" -W -x  -h supplier1.example.com

dn: cn=replica,cn=dc\=example\,dc\=com,cn=mapping tree,cn=config
changetype: modify
replace: nsDS5ReplicaTombstonePurgeInterval
nsDS5ReplicaTombstonePurgeInterval: 43200 # in seconds, 12 hours

changetype: modify
replace: nsDS5ReplicaPurgeDelay
nsDS5ReplicaPurgeDelay: 90000 # in seconds, 25 hours

Note

To clean up the tombstone entries and the state information immediately, set set a very small value to the nsDS5ReplicaTombstonePurgeInterval and nsDS5ReplicaPurgeDelay attributes. Both attributes have values set in seconds, so the purge operations can be initiated almost immediately.

Warning

Always use the purge intervals to clean out tombstone entries from the changelog. Never delete tombstone entries manually.