Oracle user process uses all memory and swap on server

From: Tom Dale <tom.dale_at_fivium.co.uk>
Date: Mon, 12 Nov 2012 15:04:43 +0000
Message-ID: <CAGDf7wSZTY=BwGFHg_8V5Y57n=+nEtVL3rcyCxBSMMJJLUuwXw_at_mail.gmail.com>

Hi Oracle-l,
When I run a simple select on one 160mb clob, the oracle foreground process uses all memory, swap then hangs the server.

This happens on red hat 4.5, 5.7 and OEL 6.3 Tested Oracle 10.2.0.4 and 11.2.0.3.4 Tested

Oracle Support :

---

What is happening is that single block reads are being done which presumably also takes up more and more buffer space, as well as making for very poor performance. This will give the appearance of a hang.

---

I do not think that is the case!!

My question is :

How can the Oracle OS user process consume all the memory when the the SGA is small and v$pgastat shows a small maxpga?

Test script is on OTN

https://forums.oracle.com/forums/thread.jspa?threadID=2463164&tstart=0

Bit more info...

When I run an OS strace and a 10046 trace, after the last traced event in the oracle session trace, the process strace continues with

brk(0x60013000) = 0x5ffef000
mmap(NULL, 1048576, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0)
= 0x2b9e775b5000

brk(0x60013000) = 0x5ffef000
mmap(NULL, 1048576, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0)
= 0x2b9e776b5000

...
Repeating
...

Any answers would enlightening!

Thanks,
Tom

--
http://www.freelists.org/webpage/oracle-l

Received on Mon Nov 12 2012 - 16:04:43 CET