Digging into ADD SCHEMATRANDATA … what is #GoldenGate doing?

In my post on the differences between ADD TRANDATA and ADD SCHEMATRANDATA, I highlighted the differences between the two ways of adding supplemental logging to tables. One of the things I pointed out was that ADD SCHEMATRANDATA doesn’t add any supplemental log groups (SLG). Without SLGs then how does ADD SCHEMATRANDATA work? That was the question I was left with. So I started digging around to find the answer and I think you may be interested in this as well.

Tracing

In order to figure out what is actually going on within the database when ADD SCHEMATRANDATA is run, I had to trace the GoldenGate session. In order to do this, I first had to login to the database from GGSCI using “dblogin useridalias ggate”. After logging in, I had to identify the session and then setup tracing. This was done with the following scripts:

—SQL to identify the GGate session
select 'exec dbms_monitor.session_trace_enable(session_id=>'||sid||', serial_num=>'||serial#||', binds=>true, waits=true)' 
from v$session where username = 'GGATE';

—Start tracing of the session
exec dbms_monitor.session_trace_enable(session_id=>156, serial_num=>15799, binds=>true, waits=>true);
—Disable tracing after done
exec dbms_monitor.session_trace_disable(session_id=>156, serial_num=>15799);

Now with tracing enabled for the session, I’m able to trace the GoldenGate session from within GGSCI. At this time, I’m able to run DELETE SCHEMATRANDATA and ADD SCHEMATRANDATA against the schema I want to add the logging to.

All the last step of tracing, I need to disable the tracing (script above) of the GoldenGate user and identify the trace file. By default the trace file should be in $ORACLE_BASE/diag/rdbms/<db>/<db>/trace. In this directory, you need to identify the trace files for the session that was traced.

oracle >ls -ltr src12c_ora_23267.*
-rw-r-----. 1 oracle oracle   2788 Aug 15 16:09 src12c_ora_23267.trm
-rw-r-----. 1 oracle oracle 300531 Aug 15 16:09 src12c_ora_23267.trc

After you know the trace file, you can perform a TKPROF on the trace file.

oracle > tkprof src12c_ora_23267.trc src12c_ora_23267.tkprofs

This is generate a tkprofs file that will show specifics for the sessions, such as the SQL that is ran. This is what I’m more interested in, what SQL is ran with ADD SCHEMATRANDATA is called.

Quick Review of TKProf file

I’m not going to go into all the details that are in a tkprof generated file, but I took a look through the file trying to figure out what is going on when running the ADD SCHEMATRANDATA. As I was searching the file, I found a reference to LOGMNR$ALWAYS_SUPLOG_COLUMNS.

If you have kept up with the versions of Oracle GoldenGate, you will know that this is a reference to LogMiner and that Oracle is integrating the Oracle GoldenGate processes with it. This is also a hint to where to look, a.k.a Oracle Streams. As I continued to look through the tkprof file, I found a few references to a streams package – DBMS_CAPTURE_ADM; along with sql statements making calls to views like STREAMS$_PREPARE_DDL.

BEGIN sys.dbms_capture_adm.PREPARE_SCHEMA_INSTANTIATION('soe','ALLKEYS_OFF');
  END;
BEGIN sys.dbms_capture_adm.ABORT_SCHEMA_INSTANTIATION('soe'); END;

At this point, it is safe to say, that Oracle is merging Oracle GoldenGate into Oracle Streams.

Review of Packages

Note: These are Oracle packages that are encrypted in the database. You can use UnwrapIt to view if needed. Output here is only for teaching purposes.

Now that I’m armed with what package the ADD SCHEMATRANDATA is calling, I can drill into what exactly is happening.

In taking a look at the DBMS_CAPTURE_ADM package, I wanted to look at the PREPARE_SCHEMA_INSTANTIATION procedure. I can quickly see that the procedure takes the schema_name and that defaults will be used for supplemental logging and container.

PROCEDURE PREPARE_SCHEMA_INSTANTIATION( 
 SCHEMA_NAME               IN VARCHAR2, 
 SUPPLEMENTAL_LOGGING      IN VARCHAR2 DEFAULT 'KEYS', 
 CONTAINER                 IN VARCHAR2 DEFAULT 'CURRENT') 
 IS 
 BEGIN 
 DBMS_CAPTURE_ADM_IVK.PREPARE_SCHEMA_INST_IVK( 
 SCHEMA_NAME => SCHEMA_NAME,  
 SUPPLEMENTAL_LOGGING => SUPPLEMENTAL_LOGGING, 
 CONTAINER => CONTAINER); 
 END;

After the schema is passed to the procedure, this procedure passes all three of the parameters to another package and procedure for execution. This package is DBMS_CAPTURE_ADM_IVK.PREPARE_SCHEMA_INST_IVK.

PROCEDURE PREPARE_SCHEMA_INST_IVK(
SCHEMA_NAME               IN VARCHAR2,
SUPPLEMENTAL_LOGGING      IN VARCHAR2 DEFAULT 'KEYS',
CONTAINER                 IN VARCHAR2 DEFAULT 'CURRENT')
IS
SUPP_LOG_LEVEL      BINARY_INTEGER;    
SYNCHRONIZATION     VARCHAR2(4) := 'LOCK';
CANON_CONTAINER     DBMS_ID;
BEGIN
  DBMS_LOGREP_UTIL.WRITE_TRACE('prepare_schema_inst_ivk()+ container: ' ||
  CONTAINER, DBMS_LOGREP_UTIL.TRACE_FLAG_PUBLISH);
  COMMIT;

  SUPP_LOG_LEVEL := DBMS_CAPTURE_ADM_IVK.SUPP_LOG_LEVEL_CHECK(SUPPLEMENTAL_LOGGING, 
  IF DBMS_XSTREAM_GG_ADM.SYNCHRONIZATION IS NOT NULL 
  THEN
         SYNCHRONIZATION := UPPER(DBMS_XSTREAM_GG_ADM.SYNCHRONIZATION);
  END IF;

  IF (SYNCHRONIZATION != 'NONE' AND SYNCHRONIZATION != 'LOCK' AND
           SYNCHRONIZATION != 'WAIT') 
  THEN
           DBMS_LOGREP_UTIL.RAISE_SYSTEM_ERROR(
           DBMS_STREAMS_ADM.INVALID_PARAMETER_NUM,
           DBMS_LOGREP_UTIL.GET_PROC_USE_CONTEXT, 'SYNCHRONIZATION'); 
  END IF;

  DBMS_UTILITY.CANONICALIZE(CONTAINER, CANON_CONTAINER, 30);   
  IF CANON_CONTAINER = 'CURRENT' 
  THEN
     DBMS_LOGREP_UTIL.WRITE_TRACE('prepare schema objects for current' ||
            ' container: ', DBMS_LOGREP_UTIL.TRACE_FLAG_PUBLISH);
     DBMS_CAPTURE_ADM_INTERNAL.PREPARE_SCHEMA_INSTANTIATION(
        SCHEMA_NAME, SYS_CONTEXT('USERENV','_USER'), SUPP_LOG_LEVEL,
        SYNCHRONIZATION, CANON_CONTAINER);
  ELSE
     DBMS_CAPTURE_ADM_INTERNAL.PREPARE_SCHEMA_INST_CDB(
            SCHEMA_NAME, CANON_CONTAINER,
            SYS_CONTEXT('USERENV','_USER'), SUPP_LOG_LEVEL,
           SYNCHRONIZATION);
  END IF;

  COMMIT;
  END;

In this procedure, you will notice that the input is all three of the parameters that are passed from the PREPARE_SCHEMA_INSTANTIATION procedure. Then this procedure precedes to check the supplemental logging and attempts to sync the tables with the DBMS_XSTREAM_GG_ADM package. After the tables have been synced, then the procedure calls another package and procedure to prepare (DBMS_CAPTURE_ADM_INTERNAL.PREPARE_SCHEMA_INSTANTIATION) the tables with the required supplemental logging level.

When looking at this procedure, it takes in five different parameters; two of which are defaults. After the procedure sets up tracing, it checks to see if the calling procedure is specifying GoldenGate. Once everything is confirmed and synchronization is complete, then the procedure grabs the CURRENT_SCN, waits for any inflight transactions and prepares the schema tables before exiting the package.

PROCEDURE PREPARE_SCHEMA_INSTANTIATION(
SCHEMA_NAME               IN VARCHAR2,
CANON_INVOKER             IN VARCHAR2,
SUPP_LOG_LEVEL            IN BINARY_INTEGER,
SYNCHRONIZATION           IN VARCHAR2 DEFAULT 'LOCK',
CONTAINER                 IN VARCHAR2 DEFAULT 'CURRENT')
IS
CANON_SCHEMA  DBMS_ID;
IDX           NUMBER :=0;
NAME_ARRAY    DBMS_UTILITY.NAME_ARRAY;
OWNER_ARRAY   DBMS_UTILITY.NAME_ARRAY;
CURRENT_SCN   NUMBER;
WFIT          BOOLEAN := FALSE;
RUN_TKLRWT1   BOOLEAN := FALSE;
SUPPLOG       BINARY_INTEGER := SUPP_LOG_LEVEL;
TEST_EVENT_LVL   BINARY_INTEGER := 0;
BEGIN
 DBMS_LOGREP_UTIL.WRITE_TRACE(
 'dbms_capture_adm_internal.prepare_schema_instantiation()+', DBMS_LOGREP_UTIL.TRACE_FLAG_PUBLISH);
 DBMS_LOGREP_UTIL.WRITE_TRACE('schema_name='||SCHEMA_NAME,   DBMS_LOGREP_UTIL.TRACE_FLAG_PUBLISH);
  DBMS_LOGREP_UTIL.WRITE_TRACE('supp_log_level='||SUPP_LOG_LEVEL, DBMS_LOGREP_UTIL.TRACE_FLAG_PUBLISH); 
 DBMS_UTILITY.CANONICALIZE(SCHEMA_NAME, CANON_SCHEMA, 30);  

 IF ((SUPP_LOG_LEVEL = DBMS_STREAMS_DECL.SUPPLOG_ALLKEYS_ON) OR (SUPP_LOG_LEVEL = DBMS_STREAMS_DECL.SUPPLOG_ALLKEYS_OFF)) 
 THEN
    STORE_PREPARE_INFO(CANON_SCHEMA, SUPP_LOG_LEVEL);
    RETURN;
 END IF;
    DBMS_LOGREP_UTIL.WRITE_TRACE('dbms_capture_adm_internal.prepare_schema_instantiation()+ before read_ev', DBMS_LOGREP_UTIL.TRACE_FLAG_PUBLISH);
    DBMS_SYSTEM.READ_EV(DBMS_LOGREP_UTIL.EXT_TRACE_EV, TEST_EVENT_LVL);

    IF (DBMS_XSTREAM_GG_ADM.IS_GOLDENGATE AND TEST_EVENT_LVL = 0) 
    THEN
     SUPPLOG := 0;
    END IF;
   DBMS_LOGREP_UTIL.WRITE_TRACE('dbms_capture_adm_internal.prepare_schema_instantiation()+ before prepare_ddl',DBMS_LOGREP_UTIL.TRACE_FLAG_PUBLISH);
    PREPARE_DDL(CANON_SCHEMA, SUPPLOG);
    DBMS_LOGREP_UTIL.WRITE_TRACE('dbms_capture_adm_internal.prepare_schema_instantiation()+ before prepare_schema_tables',DBMS_LOGREP_UTIL.TRACE_FLAG_PUBLISH);
	RUN_TKLRWT1 := RUN_WFIT_TEST;

    IF (UPPER(SYNCHRONIZATION) = 'NONE') 
    THEN
		PREPARE_SCHEMA_TABLES(CANON_SCHEMA, FALSE, CONTAINER);
    ELSIF (UPPER(SYNCHRONIZATION) = 'LOCK') 
    THEN
		PREPARE_SCHEMA_TABLES(CANON_SCHEMA, TRUE, CONTAINER);
    ELSIF (UPPER(SYNCHRONIZATION) = 'WAIT') 
    THEN
		WFIT := TRUE;
 		CURRENT_SCN  := DBMS_FLASHBACK.GET_SYSTEM_CHANGE_NUMBER;
        DBMS_CAPTURE_ADM_INTERNAL.WAIT_FOR_INFLIGHT_TXN(CURRENT_SCN, 'Prepare_schema_instantiation');
        PREPARE_SCHEMA_TABLES(CANON_SCHEMA, FALSE, CONTAINER);
    END IF;
        DONE_WFIT_TEST(WFIT, RUN_TKLRWT1);
    END;

Up to this point, the last three packages have been preparing the tables. Seems like a lot of preparing to get tables set for supplemental logging. Well the last call in the previous package called the procedure to prepare the tables. When I look at this procedure, I see that there is a call to a PREPARE_TABLE_INST procedure (getting a bit lazy on copying the procedure – don’t wan to put to much since this is Oracle’s code).

FOR I IN 1..IDX LOOP
BEGIN
   DBMS_LOGREP_UTIL.WRITE_TRACE('prepare_schema_tables():' || CANON_OWNER_ARRAY(I) || '.' || CANON_NAME_ARRAY(I), DBMS_LOGREP_UTIL.TRACE_FLAG_PUBLISH);
   PREPARE_TABLE_INST(CANON_OWNER_ARRAY(I), CANON_NAME_ARRAY(I),DBMS_STREAMS_DECL.SUPPLOG_NONE, LOCKING, CONTAINER);
EXCEPTION WHEN OTHERS THEN
   DBMS_LOGREP_UTIL.DUMP_TRACE('error:dbms_capture_adm_internal.prepare_table_inst('||'owner='||CANON_OWNER_ARRAY(I)||' name='||CANON_NAME_ARRAY(I));
END;

Now, when I go an look at the PREPARE_TABLE_INST procedure, I see that the procedure is calling a C package called “knlcpreptabinst” to set the supplemental logging on the tables associated with the schema.

PROCEDURE PREPARE_TABLE_INST(
CANON_OWNER                 IN VARCHAR2,
CANON_TABLE_NAME            IN VARCHAR2,
SUPPLEMENTAL_LOGGING_LEVEL  IN BINARY_INTEGER,
LOCKING                     IN BOOLEAN,
CONTAINER                   IN VARCHAR2) 
IS
EXTERNAL
 NAME "knlcpreptabinst"
 LIBRARY DBMS_LOGREP_LIB
 WITH CONTEXT
 PARAMETERS(CONTEXT,    
      CANON_OWNER       OCISTRING, CANON_OWNER       INDICATOR SB2,
      CANON_TABLE_NAME  OCISTRING, CANON_TABLE_NAME  INDICATOR SB2,
      SUPPLEMENTAL_LOGGING_LEVEL UB4,
      LOCKING           UB2,       LOCKING           INDICATOR SB2,
      CONTAINER         OCISTRING, CONTAINER         INDICATOR SB2)
LANGUAGE C;

Without drilling down into the C package, I cannot see exactly how the supplemental logging is added or where it is exactly stored in the database. I can only assume (we know what that means … lol), that it is in the data dictionary some where.

Looking for tables or views that may shed some light on this as well, I’ve found LOGMNR$SCHEMA_ALLKEY_SUPLOG that will show you the schema, if all keys are in supplemental log mode and if no validated pks are allowed. The following query is what I used to extract information about the SOE schema:

select * from LOGMNR$SCHEMA_ALLKEY_SUPLOG
where allkey_suplog = 'YES';

 

Output is as follows:

SCHEMA_NAME     ALL ALL
--------------- --- ---
SOE             YES NO

In drilling down further, after have a friend of mine pointed out a function to me (follow him on twitter -> @resetlogs). You can get down to the table level on supplemental logging when using ADD SCHEMATRANDATA. There is a log miner function that has to be called when using SQL to pull the correct information. This function is similar named to the table I referenced above … LOGMNR$ALWAYS_SUPLOG_COLUMNS.

This function takes two parameters. The first is the schema that holds the objects and the second is the table name. So in the following example, I can see that the ORDERS table of the SOE schema has supplemental logging added.

select * from table(logmnr$always_suplog_columns('SOE','ORDERS'));

OWNER           TABLE_NAME                     COLUMN_NAME                        INTCOL     SEGCOL    USERCOL
--------------- ------------------------------ ------------------------------ ---------- ---------- ----------
SOE             ORDERS                         ORDER_ID                                1          1          1
SOE             ORDERS                         ORDER_TOTAL                             6          6          6
SOE             ORDERS                         COST_OF_DELIVERY                       11         11         11
SOE             ORDERS                         DELIVERY_ADDRESS_ID                    13         13         13
SOE             ORDERS                         ORDER_DATE                              2          2          2
SOE             ORDERS                         CUSTOMER_CLASS                         14         14         14
SOE             ORDERS                         CUSTOMER_ID                             4          4          4
SOE             ORDERS                         ORDER_STATUS                            5          5          5
SOE             ORDERS                         PROMOTION_ID                            8          8          8
SOE             ORDERS                         ORDER_MODE                              3          3          3
SOE             ORDERS                         SALES_REP_ID                            7          7          7
SOE             ORDERS                         WAREHOUSE_ID                            9          9          9
SOE             ORDERS                         DELIVERY_TYPE                          10         10         10
SOE             ORDERS                         WAIT_TILL_ALL_AVAILABLE                12         12         12
SOE             ORDERS                         CARD_ID                                15         15         15
SOE             ORDERS                         INVOICE_ADDRESS_ID                     16         16         16

 

I know this has been a long post, but hopefully, I’ve been able to somewhat show how the ADD SCHEMATRANDATA command within GGSCI works and where you can see if supplemental logging is turned on for the selected schema. As I was trying to dig to the root of this issue, I found it interesting that so many packages are involved with setting the supplemental logging on a schema/tables; while identifying if it is enabled is not as easy as ADD TRANDATA. Where when you use ADD TRANDATA the tables are easily identified and can quickly see that the supplemental log groups have been added.

As Oracle GoldenGate for Oracle moves to a more integrated approach to replication, I think more items will be tied to the log miner and streams architecture.

Enjoy!!

@dbasolved
http://about.me/dbasolved

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Mrs. Jaye's Class Blog

Coordinate Algebra & Advanced Algebra

Pardy DBA

ORA-00001: unique constraint (ORA.BLOG_TAGLINE_PK) violated

Martin Widlake's Yet Another Oracle Blog

Oracle performance, Oracle statistics and VLDBs

HeliFromFinland

Heli's thoughts on Database Designing, Oracle SQL Developer Data Modeler, User Groups etc.

Julian Dontcheff's Database Blog

The good DBA is one who learns from his mistakes, the best DBA is one who learns from other DBA's mistakes

Martins Blog

Trying to explain complex things in simple terms

The Data Warrior

Changing the world, one data model at a time. How can I help you?

Maaz Anjum's Blog

A life yet to be lived...

PeteWhoDidNotTweet.com

Stuff that interests me, if not you!

%d bloggers like this: