If any column that is part of a unique constraint is updated, then SQL Server implements the update as a "deferred update", which means as a pair of DELETE/INSERT operations. This "deferred update" causes replication to send a pair of DELETE/INSERT statements to the subscribers. There are also other situations that might cause a deferred update. Therefore, any business logic that you implement in your UPDATE triggers or custom stored procedures at the Subscriber should also be included in the DELETE/INSERT triggers or custom stored procedures.
The default behavior in transactional replication is to use INSERT
custom stored procedures to apply changes at the subscribers.INSERT
statements made at the Publisher are applied to subscribers through an INSERT
stored procedure call. Similarly, a DELETE
statement is applied through a DELETE
stored procedure call.
However, when an UPDATE
statement is executed as a "deferred update", the logreader agent places a pair of DELETE
stored procedure calls in the distribution database to be applied to the Subscribers rather than an update stored procedure call. For example, suppose you have a publishing table, named TABLE1, with these three columns:
- col1 int
- col2 int
- col3 varchar(30).
The only unique constraint on TABLE1 is defined on col1 through a primary key constraint. Assume that you have one record (1,1,'Dallas').
When you execute this code:
UPDATE TABLE1 set col1 = 3 where col2 = 'Dallas'
statement is implemented by SQL Server as a pair of DELETE
statements since you are updating col1, which has a unique index defined. Thus, the logreader places a pair of DELETE
calls in the distribution database. This can impact any business logic that is present in the triggers or custom stored procedures at the Subscriber. You should incorporate the additional business logic in DELETE
triggers or stored procedures to handle this situation.
If you prefer to use single logic and you want all your UPDATE
commands replicated as DELETE
pairs, you can enable a trace flag as described in this Microsoft Knowledge Base article:
INF: Trace Flag to Replicate UPDATE as DELETE/INSERT Pair
Additionally, if you use a horizontal filter in your publication and if the updated row does not meet a filter condition, only a DELETE
procedure call is sent to the subscribers. If the updated row previously did not meet the filter condition but meets the condition after the update, only the INSERT
procedure call is sent through the replication process.
In the preceding example, assume that you also have a horizontal filter defined on TABLE1: where col2 = 'Dallas'. If you execute this code:
UPDATE table1 set col2 = 'New York' where col1 = 3
the logreader agent only places a DELETE
stored procedure call to be applied to the subscribers since the updated row does not meet the horizontal filter criteria.
Now, if you execute this code:
UPDATE table1 set col2 = 'Dallas' where col1 = 3
the logreader generates only the INSERT
stored procedure call, since the row did not previously meet the filter condition.
Although an UPDATE
operation was performed at the Publisher, only the appropriate commands are applied at the Subscriber.