TAF On 11g DataGuard with FSFO

Question

Hi All,
Today, i have created TAF for my 11.1.0.7 Oracle database running on RHEL 5.4 64 bit.
I have configured my databases with FSFO enabled. (MaxAvailibility with LGWR SYNC AFFIRM transport)
I created TAF with following entries:
1) Created and enabled a dedicated service in primary :
I have local_listener parameter set in my primary and standby databases.
begin
dbms_service.create_service('taf_test','taf_test');
end;
begin
DBMS_SERVICE.START_SERVICE('taf_test');
end;
begin
dbms_service.modify_service
('taf_test',
FAILOVER_METHOD => 'BASIC',
FAILOVER_TYPE => 'SELECT',
FAILOVER_RETRIES => 200,
FAILOVER_DELAY => 1);
end;
/2) Made sure that listener is listening to above created service:
lsnrctl services l_payee1fe_dg_001
LSNRCTL for Linux: Version 11.1.0.7.0 - Production on 21-JAN-2010 09:52:27
Copyright (c) 1991, 2008, Oracle.  All rights reserved.
Services Summary...
Service "taf_test.us" has 1 instance(s).
  Instance "payee1fe", status READY, has 1 handler(s) for this service...
    Handler(s):
      "DEDICATED" established:13 refused:0 state:ready
         LOCAL SERVER3) TNS Entry :
taf_test.us=
(DESCRIPTION=
        (SDU= 32767)
        (ENABLE=BROKEN)
        (ADDRESS_LIST=
                (FAILOVER=ON)
                (LOAD_BALANCE=YES)
                (ADDRESS=(PROTOCOL=TCP)(HOST=payee1fe-orasvr.db.us.com)(PORT=58003))
                (ADDRESS=(PROTOCOL=TCP)(HOST=payee1fe-orasvr.db.us.com)(PORT=58003))
        (CONNECT_DATA=
                (SERVICE_NAME = taf_test.us)
  )4)Trigger for setting service on appropriate primary database :
create trigger taf_test after startup on database
declare
v_role varchar(30);
begin
select database_role into v_role from v$database;
if v_role = 'PRIMARY' then
DBMS_SERVICE.START_SERVICE('taf_test');
else
DBMS_SERVICE.STOP_SERVICE('taf_test');
end if;
end;
/5) Ran a big SELECT on primary and after 5 sec, kill primary's pmon to do fail-over.
6) After sometime, i saw that SELECT was hanged for some time and once, DGMGRL has open new primary (org standby), startup trigger got fired and SELECT started fetching rows from the point where it hanged.
In that way, my TAF was working properly as expected.
My question is :
How come new Primary got the session state of SELECT which originated on old primary and then failed-over to new primary ? I confimed that,SELECT was NOT re-executed on new primary and started fetching rows from the row where it hanged at the time of fail-over.
Since both the databases have different cache and controlfiles, i want to understand how TAF works on dataguard.
In RAC,there is always a GRD through which session state can be co-ordinated between different instances. But this is not the case in DG.
I did not find anything in PRIMARY's alert log. Though STANDBY alert log was containing below statement:
ALTER SYSTEM SET service_names='taf_test' SCOPE=MEMORY SID='payee1fe';Can anyone shed light on internal working of TAF in DG.
Regards,
Bhavik Desai

Answer 1

Hi,
Just now i tested,SELECT FAIL-OVER for SELECT with PARALLEL (4) hint. I got below msg and SELECT did not executed on new primary:
ERROR:
ORA-25401: can not continue fetches
4350 rows selected.However, i observed that the session is failed-over to new primary. After getting above msg in SQLPLUS window, when i saw number of slaves given to my new sessions, i got :
SQL> select *From v$pq_tqstat;
DFO_NUMBER      TQ_ID SERVER_TYP   NUM_ROWS      BYTES  OPEN_TIME AVG_LATENCY      WAITS   TIMEOUTS PROCES   INSTANCE
         1          0 Producer        19800      65296          0           0          3          0 P002            1
         1          0 Producer        19800      65296          0           0          4          1 P001            1
         1          0 Producer        19800      65294          0           0          3          0 P000            1
         1          0 Producer        19800      65296          0           0          3          0 P003            1
         1          0 Consumer         4351      65296          0           0          8          0 QC              1Does it mean that, SELECT fail-over is only supported for serialized SELECTs ? Or there is other alternative to achive PARALLEL SELECT fail-over ?
Regards,
Bhavik Desai
Edited by: BhavikDe on Jan 24, 2010 11:30 PM