Thursday, November 11, 2010

Re-Synchronize Streams objects online


In-cases we need to re-synchronize the data between the replicating sites, like for example in cases of errors like ORA-01403 / ORA-1422 reported by the apply process because data is out of sync, and re-synchronizing manually is not feasible.

Here are the steps we need to do to re-synchronize the data between the source and target site in the streams environment, without the need to stop the activity on the replicated objects on the source site.



Steps to follow

1. Stop the capture and apply processes:

-- Stop the capture process:

exec dbms_capture_adm.stop_capture('');

-- Make sure it is stopped:

select capture_name, status from dba_capture;



-- Stop the apply process:

exec dbms_apply_adm.stop_apply('');

-- Make sure it is stopped:

select apply_name, status from dba_apply;



2. Purge the Capture and apply queue tables:


- As we will re-synch; The messages in both source queue that has not been propagated yet, and messages in the apply queue that has not been applied yet, can both be purged. You can do this step of purging the capture and apply queue if and only if you are doing complete resync of all the streams replication objects. If you are doing synchronization for some of the objects, then do not attempt to purge any of the queues or the other objects will get out of sync

- There will be no problem if you skipped that step, the apply process will still bypass all those messages, we just do that incase we got lot of messages that may take long time to be propagated and skipped by the apply process.



-- Stop the propagation:

exec dbms_propagation_adm.stop_propagation('');

select propagation_name, status from dba_propagation;



-- To purge perform the following for both the capture and apply queue tables:

DECLARE
options dbms_aqadm.aq$_purge_options_t;
BEGIN
options.delivery_mode:=DBMS_AQADM.BUFFERED;
DBMS_AQADM.PURGE_QUEUE_TABLE('',NULL,options);
END;
/



-- check that the messages has been successfully purged for capture and apply queue:

SELECT QUEUE_SCHEMA, QUEUE_NAME, (NUM_MSGS - SPILL_MSGS) MEM_MSG, SPILL_MSGS, NUM_MSGS
FROM V$BUFFERED_QUEUES
where queue_name='';



-- Start the propagation again:

exec dbms_propagation_adm.start_propagation('');
select propagation_name, status from dba_propagation;



3. Export the source schema:

exp system/oracle owner=HR file=HR.dump log=HR.log object_consistent=Y



*Object_consistent must be set to Y, so the imported data would be consistent with the source.

4) Delete the apply errors:



As we will re-synchronize, we can delete the errors that are in the error_queue of the apply process.

To do so, we can execute either DBMS_APPLY_ADM.DELETE_ERROR or DBMS_APPLY_ADM.DELETE_ALL_ERRORS at the apply site.



You need to make sure that you are not deleting any needed transaction, specially if you are not synchronizing all the replicated for that apply process.

If you are synchronizing all the data you can execute directly DBMS_APPLY_ADM.DELETE_ALL_ERRORS.



-- Check the errors in the error queue:

select apply_name, LOCAL_TRANSACTION_ID, ERROR_MESSAGE from dba_apply_error;



-- To delete specific transaction:

exec DBMS_APPLY_ADM.DELETE_ERROR('');



-- To delete all errors:

exec DBMS_APPLY_ADM.DELETE_ALL_ERRORS('');







If you need to check the contents of the transactions before deleting check the procedures in documentation:

Reference :
Oracle® Streams Concepts and Administration 10g Release 2 (10.2) Part Number B14229-04



5. Import the source schema to target:
imp system/oracle file=HR.dump full=y ignore=y STREAMS_INSTANTIATION=Y


When doing STREAMS_INSTANTIATION=Y and having the export done with object_consistent=Y, the instantiation SCN for the apply will be modified to the SCN at the time the export was taken, and this will insure that data at the target is consistent with data at the source.



In the example above of exp/imp we are assuming that we are replicating schema MARS, and we are trying to synchronize it, but we may also exp/imp specific tables, we do not have to synchronize the whole schema when we are doing schema level streams.



Remember that when you do the import, you will need to truncate the tables being re-synchronized to avoid the failure of import due to ORA-00001 "unique constraint violated", incase we have unique keys on the table. Or, to avoid duplicate rows, incase we have do not have unique keys.




Note:

- After doing the import, we are now sure that the data is now consistent, and the apply process will only apply changes starting from the SCN after the time of the export.

- If the exp/imp will be done using datapump, then we will not need to set any parameters, as the imported objects will be automatically imported with the SCN of the source at point in time where the export was taken, as long as there is an existing apply process on the apply site.



6. Start capture and apply:

-- Start the apply process:

exec dbms_apply_adm.start_apply('');

-- Make sure apply is started

select apply_name, status from dba_apply;



-- Start the capture process:

exec dbms_capture_adm.start_capture('');

-- Make sure capture is started

select capture_name, status from dba_capture;