Fixed issues

DBMI-28261

Database ingestion and replication combined load jobs that have persistent storage enabled might encounter unexpected errors related to leaving an orphaned InputPersist helper thread running after a task fails.

If you add multiple new apply jobs to a CDC staging group and then deploy and run them while the group’s CDC Staging Task is running, the CDC Staging Task fails with an error and all of the other apply jobs in the group that were previously running stop.

A database ingestion and replication combined load job that has schema drift enabled might fail after a restart if the first schema received in the CDC change stream contains a DDL change such as an added table. In this case, the system might not have the previous schema in cache to process older pending data in the stream.

Database ingestion and replication combined load jobs might fail during the initial unload phase. When the backlog process detects that the unload failed, it clears the name of the table that was being processed. Then, if another unload event occurs, the backlog process checks if it’s for the same table. At this point, the backlog process issues the following error because that table name was cleared

Adding the customPrimaryKeys source custom property to an existing incremental load or combined initial and incremental load task might result in a primary key change not allowed error. Following a resync, new DML changes are not replicated to the target, and a restart of the task results in the primary key change not allowed error for the resynced table.

When communication between the source CDC capture process and Data Ingestion and Replication fails, the monitoring interface still shows the job status as Up and Running although no data is being processed.

The intermediateStorageDirectory and storageBackupDirectory properties of the Database Ingestion service that runs under the Secure Agent must specify unique directory locations. However, no checking for their uniqueness occurs. With this fix, an exception will occur if the properties point to the same location.

If you resume database ingestion and replication apply jobs that are part of a CDC staging group after they’ve stopped, the resume attempt initially fails. If you manually retry the resume operation multiple times, the apply jobs eventually resume processing. The problem occurs because of unnecessary scanning for the latest schema information in the schema history table.