Queries that are executed against the mirror database can place locks, depending on the isolation level specified. If locks interfere with operations being applied from the primary server, then the connections holding the locks have their transactions rolled back and any open cursors for those connections are closed. Applications running at isolation level 0 do not add row locks, but still acquire schema locks. If the schema locks interfere with operations being applied from the primary server, the transaction on the mirror database is rolled back.

Applications that require a consistent view of the database (and so cannot use isolation level 0) should consider using snapshot isolation. To do so, the allow_snapshot_isolation option must be set to On. This option takes effect on both the primary server and the mirror server, so the costs associated with snapshot isolation need to be considered.

Connections to the mirror database are affected by transactions against the primary server, since those operations are then processed and applied by the mirror server. There can be a small delay between the time an update on the primary server is committed and the time that the update is available on the mirror server. Normally this delay is short, but you should keep this in mind when you are accessing the database running on the mirror server.

 See also