Oracle export hang

[Alexandre Arents]

Hi all,

ORACLE 8.1.7.4.0
HPUX PARISC 11i

We're experencing problem during an export of a big table, the exp dump core with the following message:

gefec: fatal error 0
********** Internal heap ERROR 17113 addr=0x0
******************************************************
HEAP DUMP heap name="" desc=0x0
extent sz=0x0 alt=0 het=0 rec=0 flg=0 opc=0
parent=0 owner=0 nex=0 xsz=0x0
Hla: 0

I found this link on ITRC
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=566652

It sounds like the server is out of memory, but
at the time the export running, there is a lot of memory aviaible on the server:

# swapinfo -tm
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 4096 0 4096 0% 0 - 1 /dev/vg00/lvol2
dev 8192 0 8192 0% 0 - 0 /dev/vg00/swap
dev 8192 0 8192 0% 0 - 0 /dev/vg00/swap2
reserve - 9463 -9463
memory 25552 4516 21036 18%
total 46032 13979 32053 30% - 0 -

and the ulimit of dba user account is:
ulimit -a
time(seconds) unlimited
file(blocks) unlimited
data(kbytes) 1310720
stack(kbytes) 381952
memory(kbytes) unlimited
coredump(blocks) 4194303

the "exp" prog coredump at the size of 650MByte(memory size of proc), which is under the above limitation..

I don't know what memory parameter can limit that..

Other thing:
-When it failed, the export ouput file is 6,9 GBytes instead of 9Go
-the export with "buffer=1024" gives similar result
-the export of the table in "direct=y" mode is OK
-the export of the whole shema which contain the table is OK.
-database is 64 bits but "exp" command is 32bits exec.

-and finally, we succesfully run this export on our developpement system very similar except with lower memory physical memory and memory parameter value...

Any Idea ?
Alex

[Steven E. Protter]

Shalom Alex,

If the Oracle version is the same on both systems, consider relinking oracle on the effected system.

SEP

[Eric Antunes]

Hi Alexandre,

If you are running the export from a client (32 bits) and the database is not the same
wordsize (64 bits), you got bug 2298968.

Is it the case?

Best Regards,

Eric Antunes

[Alexandre Arents]

Steven,
The release is the same on dev and prod server.
I will relink soon.

Eric,
We run 32bit client "exp" on the database 64bit (default install with oracle 8i database??) in our prod and dev server.

The bug 2298968 is a memory leak in the client side. If we run this bug, it is on our dev -and- prod platform.
When I take a look at the memory growth of the exp process, the memory usage is the same on dev and prod server:
but on the prod serveur it failed à 650MB
and on the dev server, the dump finish at 690MB

when an export of the prod is launch from the dev server (I mean over the network) It works.
and when it is launch to dev from prod it failed.

So, I want to tell the pb is in the client side from on the prod server..

If the problem is in client side I don't know if a relink of the database will help ?

thanks

[Eric Antunes]

Hi Alex,

Try it on the production server with DIRECT=Y and BUFFER=3000 (for example). If it fails, reduce the BUFFER parameter further.

Best Regards,

Eric Antunes