Friday, January 9, 2009

Oracle Streams Step by Step setup

The first step is to create our streams administrator. I will follow the guidelines from the oracle docs exactly for this:
Connect as sysdba:
sqlplus / as sysdba

Create the streams tablespace (change the name and/or location to suit):
create tablespace streams_tbs datafile 'c:\temp\stream_tbs.dbf' size 25M
reuse autoextend on maxsize unlimited;

Create our streams administrator:

create user strmadmin identified by strmadmin
default tablespace streams_tbs
quota unlimited on streams_tbs;

I haven't quite figured out why, but we need to grant our administrator DBA privs. I think this is a bad thing. There is probably a work around where I could do some direct grants instead but I haven't had time to track those down.
grant dba to strmadmin;

We also want to grant streams admin privs to the user.

BEGIN
DBMS_STREAMS_AUTH.GRANT_ADMIN_PRIVILEGE(
grantee => 'strmadmin',
grant_privileges => true);
END;
/

The next steps we'll run as the HR user.
conn hr/hr

Grant all access to the employee table to the streams admin:
grant all on hr.employees to strmadmin;

We also need to create the employee_audit table. Note that I am adding three columns in this table that do not exist in the employee table.

CREATE TABLE employee_audit(
employee_id NUMBER(6),
first_name VARCHAR2(20),
last_name VARCHAR2(25),
email VARCHAR2(25),
phone_number VARCHAR2(20),
hire_date DATE,
job_id VARCHAR2(10),
salary NUMBER(8,2),
commission_pct NUMBER(2,2),
manager_id NUMBER(6),
department_id NUMBER(4),
upd_date DATE,
user_name VARCHAR2(30),
action VARCHAR2(30));

Grant all access to the audit table to the streams admin user:
grant all on hr.employee_audit to strmadmin;

We connect as the streams admin user:
conn strmadmin/strmadmin

We can create a logging table. You would NOT want to do this in a high-volume production system. I am doing this to illustrate user defined monitoring and show how you can get inside the capture process.
CREATE TABLE streams_monitor (
date_and_time TIMESTAMP(6) DEFAULT systimestamp,
txt_msg CLOB );

Here we create the queue. Unlike AQ, where you have to create a separate table, this step creates the queue and the underlying ANYDATA table.
BEGIN
DBMS_STREAMS_ADM.SET_UP_QUEUE(
queue_table => 'strmadmin.streams_queue_table',
queue_name => 'strmadmin.streams_queue');
END;
/

This just defines that we want to capture DML and not DDL.
BEGIN
DBMS_STREAMS_ADM.ADD_TABLE_RULES(
table_name => 'hr.employees',
streams_type => 'capture',
streams_name => 'capture_emp',
queue_name => 'strmadmin.streams_queue',
include_dml => true,
include_ddl => false,
inclusion_rule => true);
END;
/

Tell the capture process that we want to know who made the change:
BEGIN
DBMS_CAPTURE_ADM.INCLUDE_EXTRA_ATTRIBUTE(
capture_name => 'capture_emp',
attribute_name => 'username',
include => true);
END;
/

We also need to tell Oracle where to start our capture. Change the source_database_name to match your database.
DECLARE
iscn NUMBER;
BEGIN
iscn := DBMS_FLASHBACK.GET_SYSTEM_CHANGE_NUMBER();
DBMS_APPLY_ADM.SET_TABLE_INSTANTIATION_SCN(
source_object_name => 'hr.employees',
source_database_name => 'ORCL',
instantiation_scn => iscn);
END;
/

And the fun part! This is where we define our capture procedure. I'm taking this right from the docs but I'm adding a couple steps.

CREATE OR REPLACE PROCEDURE emp_dml_handler(in_any IN ANYDATA) IS
lcr SYS.LCR$_ROW_RECORD;
rc PLS_INTEGER;
command VARCHAR2(30);
old_values SYS.LCR$_ROW_LIST;
BEGIN
-- Access the LCR
rc := in_any.GETOBJECT(lcr);
-- Get the object command type
command := lcr.GET_COMMAND_TYPE();

-- I am inserting the XML equivalent of the LCR into the monitoring table.
insert into streams_monitor (txt_msg)
values (command ||
DBMS_STREAMS.CONVERT_LCR_TO_XML(in_any) );

-- Set the command_type in the row LCR to INSERT
lcr.SET_COMMAND_TYPE('INSERT');
-- Set the object_name in the row LCR to EMP_DEL
lcr.SET_OBJECT_NAME('EMPLOYEE_AUDIT');

-- Set the new values to the old values for update and delete
IF command IN ('DELETE', 'UPDATE') THEN
-- Get the old values in the row LCR
old_values := lcr.GET_VALUES('old');
-- Set the old values in the row LCR to the new values in the row LCR
lcr.SET_VALUES('new', old_values);
-- Set the old values in the row LCR to NULL
lcr.SET_VALUES('old', NULL);
END IF;

-- Add a SYSDATE for upd_date
lcr.ADD_COLUMN('new', 'UPD_DATE', ANYDATA.ConvertDate(SYSDATE));
-- Add a user column
lcr.ADD_COLUMN('new', 'user_name',
lcr.GET_EXTRA_ATTRIBUTE('USERNAME') );
-- Add an action column
lcr.ADD_COLUMN('new', 'ACTION', ANYDATA.ConvertVarChar2(command));

-- Make the changes
lcr.EXECUTE(true);
commit;
END;
/

Create the DML handlers:
BEGIN
DBMS_APPLY_ADM.SET_DML_HANDLER(
object_name => 'hr.employees',
object_type => 'TABLE',
operation_name => 'INSERT',
error_handler => false,
user_procedure => 'strmadmin.emp_dml_handler',
apply_database_link => NULL,
apply_name => NULL);
END;
/
BEGIN
DBMS_APPLY_ADM.SET_DML_HANDLER(
object_name => 'hr.employees',
object_type => 'TABLE',
operation_name => 'UPDATE',
error_handler => false,
user_procedure => 'strmadmin.emp_dml_handler',
apply_database_link => NULL,
apply_name => NULL);
END;
/
BEGIN
DBMS_APPLY_ADM.SET_DML_HANDLER(
object_name => 'hr.employees',
object_type => 'TABLE',
operation_name => 'DELETE',
error_handler => false,
user_procedure => 'strmadmin.emp_dml_handler',
apply_database_link => NULL,
apply_name => NULL);
END;
/

Create the apply rule. This tells streams, yet again, that we in fact do want to capture changes. The second calls tells streams where to put the info. Change the source_database_name to match your database.

DECLARE
emp_rule_name_dml VARCHAR2(30);
emp_rule_name_ddl VARCHAR2(30);
BEGIN
DBMS_STREAMS_ADM.ADD_TABLE_RULES(
table_name => 'hr.employees',
streams_type => 'apply',
streams_name => 'apply_emp',
queue_name => 'strmadmin.streams_queue',
include_dml => true,
include_ddl => false,
source_database => 'ORCL',
dml_rule_name => emp_rule_name_dml,
ddl_rule_name => emp_rule_name_ddl);
DBMS_APPLY_ADM.SET_ENQUEUE_DESTINATION(
rule_name => emp_rule_name_dml,
destination_queue_name => 'strmadmin.streams_queue');
END;
/


We don't want to stop applying changes when there is an error, so:

BEGIN
DBMS_APPLY_ADM.SET_PARAMETER(
apply_name => 'apply_emp',
parameter => 'disable_on_error',
value => 'n');
END;
/

Turn on the apply process:
BEGIN
DBMS_APPLY_ADM.START_APPLY(
apply_name => 'apply_emp');
END;
/

Turn on the capture process:
BEGIN
DBMS_CAPTURE_ADM.START_CAPTURE(
capture_name => 'capture_emp');
END;
/

Connect as HR and make some changes to Employees.
sqlplus hr/hr

INSERT INTO hr.employees VALUES(207, 'JOHN', 'SMITH',
'JSMITH@MYCOMPANY.COM',
NULL, '07-JUN-94', 'AC_ACCOUNT', 777, NULL, NULL, 110);
COMMIT;
UPDATE hr.employees SET salary=5999 WHERE employee_id=206;
COMMIT;
DELETE FROM hr.employees WHERE employee_id=207;
COMMIT;

It takes a few seconds for the data to make it to the logs and then back into the system to be applied. Run this query until you see data (remembering that it is not instantaneous):

SELECT employee_id, first_name, last_name, upd_Date, action
FROM hr.employee_audit
ORDER BY employee_id;

Then you can log back into the streams admin account:
sqlplus strmadmin/strmadmin

View the XML LCR that we inserted during the capture process:
set long 9999
set pagesize 0
select * from streams_monitor;

That's it! It's really not that much work to capture and apply changes using Oracle 10g. Of course, it's a little bit more work to cross database instances, but it's not that much.

One of the things that amazes me is how little code is required to accomplish this. The less code I have to write, the less code I have to maintain.

AQ 10g to 9i

NOTE: Streams is not available with Oracle 10g XE. Download and install EE. If you have 1 GB or more of RAM on your PC, you can download EE and use the DBCA to run two database instances. You do not physically need two machines to get this to work.

You need two Oracle instances for this. I used a 10g instance as my source (required) and a 9i database as my target (you could also use a 10g instance here). I called my first instance ORCL (how creative!) and I called my second instance SECOND. It works for me!

As I said, ORCL will be my source instance and SECOND will be my target instance. You should already have the CDC code from the last article running in ORCL. If you are not using the same names for your databases and you are not sure of the exact name of your databases (including domain), use global_name, i.e. select * from global_name;

ORCL must be in archivelog mode to run CDC. SECOND does not need archivelog mode. Having two databases running on a single PC in archivelog mode can really beat up a poor IDE drive.

You already created your streams admin user in ORCL so now do the same thing in SECOND. The code below is mostly the same code that you ran on ORCL. I made a few minor changes in case you are running both instances on a single PC:
sqlplus / as sysdba

create tablespace streams_second_tbs datafile 'c:\temp\stream_2_tbs.dbf' size 25M
reuse autoextend on maxsize unlimited;

create user strmadmin identified by strmadmin
default tablespace streams_second_tbs
quota unlimited on streams_second_tbs;

grant dba to strmadmin;

Connect as strmadmin. You need to create an AQ table, AQ queue and then start the queue. That's what the code below does.
BEGIN
DBMS_AQADM.CREATE_QUEUE_TABLE(
queue_table => 'lrc_emp_t',
queue_payload_type => 'sys.anydata',
multiple_consumers => TRUE,
compatible => '8.1');


DBMS_AQADM.CREATE_QUEUE(
queue_name => 'lrc_emp_q',
queue_table => 'lrc_emp_t');


DBMS_AQADM.START_QUEUE (
queue_name => 'lrc_emp_q');
END;
/

You also need to create a database link. You have to have one from ORCL to SECOND but for debugging, I like a link in both. So, while you're in SECOND, create a link:
CREATE DATABASE LINK orcl.world
CONNECT TO strmadmin
IDENTIFIED BY strmadmin
USING 'orcl.world';

Log into ORCL as strmadmin and run the exact same command there. Most of the setup for this is exactly the same between the two instances.

Create your link on this side also.

CREATE DATABASE LINK second.world
CONNECT TO strmadmin
IDENTIFIED BY strmadmin
USING 'second.world';

Ok, now we have running queues in ORCL and SECOND. While you are logged into ORCL, you will create a propagation schedule. You DO NOT need to run this in SECOND.
BEGIN
DBMS_STREAMS_ADM.ADD_TABLE_PROPAGATION_RULES(
table_name => 'hr.employees',
streams_name => 'orcl_2_second',
source_queue_name => 'strmadmin.lrc_emp_q',
destination_queue_name => 'strmadmin.lrc_emp_q@second.world',
include_dml => true,
include_ddl => FALSE,
source_database => 'orcl.world');
END;
/

This tells the database to take the data in the local lrc_emp_q and send it to the named destination queue.

We're almost done with the propagation now. We are going to modify the EMP_DML_HANDLER (from above) so that we get an enqueue block just above the execute statement:

CREATE OR REPLACE PROCEDURE emp_dml_handler(in_any IN ANYDATA) IS
lcr SYS.LCR$_ROW_RECORD;
rc PLS_INTEGER;
command VARCHAR2(30);
old_values SYS.LCR$_ROW_LIST;
BEGIN
-- Access the LCR
rc := in_any.GETOBJECT(lcr);
-- Get the object command type
command := lcr.GET_COMMAND_TYPE();
-- I am inserting the XML equivalent of the LCR into the monitoring table.
insert into streams_monitor (txt_msg)
values (command ||
DBMS_STREAMS.CONVERT_LCR_TO_XML(in_any) );
-- Set the command_type in the row LCR to INSERT
lcr.SET_COMMAND_TYPE('INSERT');
-- Set the object_name in the row LCR to EMP_DEL
lcr.SET_OBJECT_NAME('EMPLOYEE_AUDIT');
-- Set the new values to the old values for update and delete
IF command IN ('DELETE', 'UPDATE') THEN
-- Get the old values in the row LCR
old_values := lcr.GET_VALUES('old');
-- Set the old values in the row LCR to the new values in the row LCR
lcr.SET_VALUES('new', old_values);
-- Set the old values in the row LCR to NULL
lcr.SET_VALUES('old', NULL);
END IF;
-- Add a SYSDATE value for the timestamp column
lcr.ADD_COLUMN('new', 'UPD_DATE', ANYDATA.ConvertDate(SYSDATE));
-- Add a user value for the timestamp column
lcr.ADD_COLUMN('new', 'user_name',
lcr.GET_EXTRA_ATTRIBUTE('USERNAME') );
-- Add an action column
lcr.ADD_COLUMN('new', 'ACTION', ANYDATA.ConvertVarChar2(command));


DECLARE
enqueue_options DBMS_AQ.enqueue_options_t;
message_properties DBMS_AQ.message_properties_t;
message_handle RAW(16);
recipients DBMS_AQ.aq$_recipient_list_t;
BEGIN
recipients(1) := sys.aq$_agent(
'anydata_subscriber',
'strmadmin.lrc_emp_q@second.world',
NULL);
message_properties.recipient_list := recipients;


DBMS_AQ.ENQUEUE(
queue_name => 'strmadmin.lrc_emp_q',
enqueue_options => enqueue_options,
message_properties => message_properties,
payload => anydata.convertObject(lcr),
msgid => message_handle);
EXCEPTION
WHEN OTHERS THEN
insert into streams_monitor (txt_msg)
values ('Anydata: ' || DBMS_UTILITY.FORMAT_ERROR_STACK );
END;


-- Make the changes
lcr.EXECUTE(true);
commit;
END;
/

The declaration section above created some variables required for an enqueue. We created a subscriber (that's the name of the consumer). We will use that name to dequeue the record in the SECOND instance.

We then enqueued our LCR as an ANYDATA datatype.

I put the exception handler there in case there are any problems with our enqueue.

That's all it takes. Insert some records into the HR.employees table and commit them. Then log into strmadmin@second and select * from the lrc_emp_t table. You should have as many records there as you inserted.

There are not a lot of moving parts so there aren't many things that will go wrong. Propagation is where I have the most troubles. You can query DBA_PROPAGATION to see if you have any propagation errors.

That's it for moving the data from 10g to 9i.
Manual Creation of an LCR
This part of the paper is not required to move data from 10g to 9i. It is very handy if you need to manually create test data though. I recently had the need to create some test data in LCR format. If you have manual control of your LCR, you can test specific data issues a lot easier.

Anyway, I (and a coworker) searched the web and looked through various documentation but was not able to find a concise description of how to go about creating an LCR manually. So I decided to write one.

NOTE: Everything below is specifically for a ROW type LCR as opposed to a DDL type LCR. The concepts would be the same but the specific code would change.

First a little definition. An LCR format is the format of data that Oracle uses in the redo logs and is used for Oracle Streams (and probably data guard although I am guessing about that). The LCR has information about what the object is as well as the old and new values. The old and new values are exactly the same as :old and :new in a trigger.

You can see the definition of the LCR format by viewing the LCR XML Schema. http://download-east.oracle.com/docs/cd/B14117_01/server.101/b10727/ap_xmlsc.htm

An LCR is an object type. Actually, it is two table collections of an object type embedded within another object type. You can get the details of that from the documentation in the supplied PL/SQL Packages and Types documentation. http://www.oracle.com/pls/db102/to_toc?pathname=appdev.102%2Fb14258%2Ftoc.htm&remark=portal+%28Application+development%29

The short story is that an LCR stored object level information (database, object_name, owner, and command type) at the top level. Beneath that is column and data information in a name/value pair collection.

I tend to think of things like this in a relational format. If I put it in database terms, it might look something like:


Figure 1 - LCR Structure
Excuse the poor diagram. An artiste I am not.

You can do a describe to see the methods as well as view the documentation. The important thing to note is the constructor. Normally, a constructor has the same name as the object type. In this case, they chose to name it CONSTRUCT.
SQL> desc sys.lcr$_row_record
METHOD
------
STATIC FUNCTION CONSTRUCT RETURNS LCR$_ROW_RECORD
Argument Name Type In/Out Default?
------------------------------ ----------------------- ------ --------
SOURCE_DATABASE_NAME VARCHAR2 IN
COMMAND_TYPE VARCHAR2 IN
OBJECT_OWNER VARCHAR2 IN
OBJECT_NAME VARCHAR2 IN
TAG RAW IN DEFAULT
TRANSACTION_ID VARCHAR2 IN DEFAULT
SCN NUMBER IN DEFAULT
OLD_VALUES LCR$_ROW_LIST IN DEFAULT
NEW_VALUES LCR$_ROW_LIST IN DEFAULT

Based on that info, populating the test LCR is relatively straight-forward.

Just for your info, the type LCR$_ROW_LIST is a collection of LCR$_ROW_UNIT. Those types are also documented in the reference guide I mentioned above. You will not want to access those directly though. You can use the built-in LCR$_ROW_RECORD methods to populate those fields.

One thing to remember is that the data values that you are dealing with are sys.AnyData data types. That type has it's own rules and deserves a blog entry all to itself.

Here is a function that will create an empty LCR for you automatically for any table. Once you have the LCR you can modify the values to suit.

CREATE OR REPLACE FUNCTION create_lcr(
p_table_owner IN all_tables.owner%TYPE,
p_table_name IN all_tables.table_name%TYPE,
p_command IN VARCHAR2 )
RETURN sys.lcr$_row_record
AS
v_lcr sys.lcr$_row_record;
v_database global_name.global_name%TYPE;
BEGIN

-- verify the command type
IF p_command NOT IN ('INSERT', 'UPDATE', 'DELETE')
THEN
RETURN v_lcr;
END IF;

-- Get the database name
-- This could be parameterized
SELECT global_name
INTO v_database
FROM global_name;

-- Construct the LCR
v_lcr := sys.lcr$_row_record.construct(
source_database_name => v_database,
command_type => p_command,
object_owner => p_table_owner,
object_name => p_table_name );

-- You can override the values in the constructor by calling these methods
v_lcr.set_command_type(p_command);
v_lcr.set_object_name(p_table_name);
v_lcr.set_object_owner(p_table_owner);
v_lcr.set_source_database_name(v_database);

-- Loop through the columns and add new and old values
FOR c1 IN (
SELECT column_name, data_type
FROM all_tab_columns
WHERE owner = p_table_owner
AND table_name = p_table_name
ORDER BY column_id )
LOOP

-- Create an anydata based on column data type
-- You would expand this for all data types
-- I'm going to keep this example fairly simple
CASE c1.data_type
WHEN 'VARCHAR2'
THEN
v_lcr.add_column('new', c1.column_name, sys.AnyData.convertVarChar2(TO_CHAR(NULL)));
v_lcr.add_column('old', c1.column_name, sys.AnyData.convertVarChar2(TO_CHAR(NULL)));
WHEN 'DATE'
THEN
v_lcr.add_column('new', c1.column_name, sys.AnyData.convertDate(TO_DATE(NULL)));
v_lcr.add_column('old', c1.column_name, sys.AnyData.convertDate(TO_DATE(NULL)));
WHEN 'NUMBER'
THEN
v_lcr.add_column('new', c1.column_name, sys.AnyData.convertNumber(TO_NUMBER(NULL)));
v_lcr.add_column('old', c1.column_name, sys.AnyData.convertNumber(TO_NUMBER(NULL)));
END CASE;
END LOOP;

RETURN v_lcr;
END;

To call this function and manipulate it, you might write something like the following:
DECLARE
v_lcr sys.lcr$_row_record;
BEGIN
v_lcr := create_lcr( 'HR', 'EMPLOYEES', 'INSERT' );

-- Set some values
v_lcr.set_value('new', 'first_name', sys.anyData.convertVarchar2('Lewis'));
v_lcr.set_value('old', 'first_name', sys.anyData.convertVarchar2('George'));

-- Display Some Values
DBMS_OUTPUT.PUT_LINE(
'Database: ' || v_lcr.get_source_database_name() ||
', Object Owner: ' || v_lcr.get_object_owner() ||
', Object Name: ' || v_lcr.get_object_name() ||
', Command: ' || v_lcr.get_command_type() );

DBMS_OUTPUT.PUT_LINE(
'New First Name: ' || sys.anyData.accessVarchar2(v_lcr.get_value('new', 'first_name')) ||
', Old First Name: ' || sys.anyData.accessVarchar2(v_lcr.get_value('old', 'first_name')) );
END;

The output from this is:
Database: XE, Object Owner: HR, Object Name: EMPLOYEES, Command: INSERT
New First Name: Lewis, Old First Name: George