Parallel issue

Question

Hi ,
i have been trying to figure this for some time now, so if you are interested , here we go.
Os is hpux 11, (i can't tell exact release because that machine and db is currently unavailable...)
and there are 4 dual core cpus
DB is 10.2.0.4.0
parallel max servers = 16
parallel min servers = 0
parallel message size is at default value, 2100 and something
parallel auto tuning is false
shared pool = 400M
while sga target is 0
sga max size is cca 11G
db cache size is 2G
and there is 16k cache size of 8G
SQL structure
(called inside function)
INSERT INTO big (col1,col2....)
        SELECT col1,col2...
        FROM
(select
        sum(col1),col2...
from (
        select col1,col2...
        from
             a
            ,b
            ,lc
            ,d
            ,e
            ,f
        where
               union
               select
            col1,col2...
        from
            a
     group by
            col1,col2...
)a -> global temporary table
big -> over 600GB in size partitioned table
e -> table created with parallel default option
f -> table created with parallel default option
there are also two more indexes created with parallel default option on those two table... sorry, i can't be more precise currently due to the db unavailability..
There are several succesful execution of that function which takes about 20-45 seconds and inserts always about 10,000 records, but then i see that session that does this job that is active bit apears to be stucked , and toad and V$ views are showing me tha above SQL is current statemnt.
That session is in this state until is killed. We left for few days and still it was showing the same waits and current SQL
wait events are :
PX Deq: Table Q Normal
and
PX Deq: Execution Msg
for all px processes of that sessoin
those two all have the same values
and for one, maybe this was the session that originaly called px procesess , there is
PX Deq Credit: send blkd
this one usual have the highest wait value
Also when i execute only those selects except the firts insert part, there are no problems, it finishes in about 20-45 seconds. I tryied this realy many times and i couldn't reproduce hang outside of that function.... maybe i would if i could run it with insert part, i don't know...
what do you think ?
any similiar experiences ?
also when that db will be up/available again i'll post any valuable info that i at this time could not....
Edited by: Toni Lazarin on Mar 19, 2009 1:26 PM
Edited by: Toni Lazarin on Mar 19, 2009 2:12 PM

Answer 1

Great.
It seems i manage to force the issue when executing SQL only.
I executed that select in toad and it was finished in about 30 seconds.
But , look at this , after select was done :
SQL> select * from v$px_process;
SERV STATUS           PID SPID                SID    SERIAL#
P000 IN USE            34 7606                229         85
P001 IN USE            42 7614                243        332
P002 IN USE            44 7621                241        245
P003 IN USE            43 7626                228          3repeated this select 10 minutes now and im getting the same result.
Note that my session in toad from which i executed select appeared inactive right after result were shown.
SQL> select sql_id,program,status from v$session where sid=242 and serial#=156;
SQL_ID        PROGRAM                                          STATUS
c9q8wuhdhgmdg toad.exe                                         INACTIVE..and then i did this :
SQL>  select sql_id,program,status from v$session where sid=241 and serial#=245;
SQL_ID        PROGRAM                                          STATUS
c9q8wuhdhgmdg oracle@fms-a (P002)                              INACTIVE
SQL> select * from v$px_process;
SERV STATUS           PID SPID                SID    SERIAL#
P000 IN USE            34 7606                229         85
P001 IN USE            42 7614                243        332
P002 IN USE            44 7621                241        245
P003 IN USE            43 7626                228          3
SQL> select sql_id,program,status from v$session where sid=243 and serial#=332;
SQL_ID        PROGRAM                                          STATUS
c9q8wuhdhgmdg oracle@fms-a (P001)                              ACTIVE
SQL> select * from v$px_process;
SERV STATUS           PID SPID                SID    SERIAL#
P000 IN USE            34 7606                229         85
P001 IN USE            42 7614                243        332
P002 IN USE            44 7621                241        245
P003 IN USE            43 7626                228          3
SQL> r
  1  select name,value from v$sysstat
  2  where upper(name) like '%PARALLEL OPERATIONS%'
  3* or upper(name) like '%PARALLELIZED%' or upper(name) like '%PX%'
NAME                                                                  VALUE
queries parallelized                                                      2
DML statements parallelized                                               0
DDL statements parallelized                                               0
DFO trees parallelized                                                    2
Parallel operations not downgraded                                        2
Parallel operations downgraded to serial                                  1
Parallel operations downgraded 75 to 99 pct                               0
Parallel operations downgraded 50 to 75 pct                               0
Parallel operations downgraded 25 to 50 pct                               0
Parallel operations downgraded 1 to 25 pct                                0
PX local messages sent                                               625988
NAME                                                                  VALUE
PX local messages recv'd                                             625983
PX remote messages sent                                                   0
PX remote messages recv'd                                                 0
14 rows selected.am i getting this right ?
Parallel select is finished but there are still some active slaves that were used for that parallel query ?
Also v$session and v$px_process are showing contradictory data ?
Edited by: Toni Lazarin on Mar 19, 2009 3:13 PM
Edited by: Toni Lazarin on Mar 19, 2009 3:14 PM