Data concurrency on the active standby database

Changes on the HADR primary database may not necessarily be reflected on the HADR active standby database. Uncommitted changes on the primary may not replicate to the standby until the primary flushes, or sends, its logs to disk.

Logs are only guaranteed to be flushed to disk-and, therefore sent to the standby-after they have been committed. Log flushes can also be triggered by undeterministic conditions such as a log buffer full situation. As a result, it is possible for uncommitted changes on the primary to remain in the primary's log buffer for a long time. Because the logger avoids flushing partial pages, this situation may particularly affect small uncommitted changes on the primary.

If your workload running on the standby requires the data to be virtually identical to the data on the primary, you should consider committing your transactions more frequently.