CDC performance

[Roshan]

Hello Erman,

I am trying to improve the CDC replication from Oracle to Kudu using Striim. CDC is very slow. I have checked the archivelog performance on the Oracle source. Below are the results.

SQL> EXECUTE DBMS_LOGMNR.ADD_LOGFILE(LOGFILENAME => '+RECO/DWARE1/ARCHIVELOG/2020_10_06/thread_2_seq_49647.6929.1053080591', OPTIONS => DBMS_LOGMNR.NEW);

PL/SQL procedure successfully completed.

SQL> EXECUTE DBMS_LOGMNR.ADD_LOGFILE(LOGFILENAME => '+RECO/DWARE1/ARCHIVELOG/2020_10_06/thread_1_seq_45836.6826.1053080243', OPTIONS => DBMS_LOGMNR.ADDFILE);

PL/SQL procedure successfully completed.

SQL> EXECUTE DBMS_LOGMNR.ADD_LOGFILE(LOGFILENAME => '+RECO/DWARE1/ARCHIVELOG/2020_10_06/thread_1_seq_45837.5797.1053081777', OPTIONS => DBMS_LOGMNR.ADDFILE);

PL/SQL procedure successfully completed.

SQL> EXECUTE DBMS_LOGMNR.ADD_LOGFILE(LOGFILENAME => '+RECO/DWARE1/ARCHIVELOG/2020_10_06/thread_2_seq_49648.5048.1053081569', OPTIONS => DBMS_LOGMNR.ADDFILE);

PL/SQL procedure successfully completed.

SQL> EXECUTE DBMS_LOGMNR.ADD_LOGFILE(LOGFILENAME => '+RECO/DWARE1/ARCHIVELOG/2020_10_06/thread_2_seq_49649.5024.1053081777', OPTIONS => DBMS_LOGMNR.ADDFILE);

PL/SQL procedure successfully completed.

SQL> EXECUTE DBMS_LOGMNR.START_LOGMNR(OPTIONS => DBMS_LOGMNR.SKIP_CORRUPTION+DBMS_LOGMNR.NO_SQL_DELIMITER+DBMS_LOGMNR.NO_ROWID_IN_STMT+ DBMS_LOGMNR.DICT_FROM_ONLINE_CATALOG);

PL/SQL procedure successfully completed.

SQL> SET TIMING ON
SQL> Set arraysize 1000
SQL> SELECT * FROM V$LOGMNR_CONTENTS WHERE TABLE_NAME = 'TABLENOTEXIT';


no rows selected

Elapsed: 00:30:32.08

OUTPUT: 7.62580871582 * (60/31)= 14.76 GB/H

Can you please advise how I can improve the performance of CDC from Oracle to Kudu?

Regards,

Roshan

[ErmanArslansOracleBlog]

You need compare that log read rate with the archivelog generation rate in your source to understand if the CDC-read can keep up with..
For instance, is 25 GB per hour sufficient for Log Miner read? The answer is, it  depends..
Ofcourse, it is not sufficient in a mission critical production environment, which generates high redo (for instance 30GB per hour)..

read -> https://ermanarslan.blogspot.com/2020/03/rdbms-cdc-log-miner-read-performance.html