| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
 |
 | |||
| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
|
| |||
Home -> Community -> Mailing Lists -> Oracle-L -> systemstate dump
We have an Oracle/PeopleSoft implementation (8.1.7 on Solaris) which on occasion stops any new logons from happening. Performance for existing sessions is so slow as to be non-existent. This is a two instance OPS system with the second instance existing solely for failover. The data I have gathered comes from three sources:statspack snapshots at 15 minute intervals, some 10046 traces (level 12) which have since been deleted, systemstate dump.
>From statspack, I compared the 15 (closer to 20) minute snapshot where the issue arose to the 15 minute snapshot just prior to that. What I found was:
-CPU time almost unchanged
-a huge change in the Total Wait Time (cs) for the following events (the number of waits were all within an order of 2 of each other except for latch free which jumped from 200 to 150,000).
NAME TIME-initial (cs) TIME-final (cs)
latch free 200 2,800,000
enqueue 900 280,000
global cache lock busy 8,200 30,000
library cache pin 500 18,000
row cache lock 600 15,000
library cache lock 600 8,000
DFS lock handle 30 5,000
PX Deq: Execution Msg -- 5,000
-latch sleeps skyrocketed for library cache, shared pool, and row cache objects with many having at least 4 sleeps. The 'dlm resource hash list' latch had about the same number of sleeps, but the number of gets increased by ~x10.
For the 10046 traces (taken a few weeks ago) I had a chance to look through a few of the multitude before they were deleted. -Lots of latch free:library cache, shared pool, and row cache object waits. The library cache wait always pointed to the same address (the same 1 of 5 children). -Increase in library cache lock and library cache pin. These are pointing to the same handle address. Also an increase in the library cache pin instance lock enqueue.
My working model is that the latch free (library cache) waits are a symptom. Something is holding a library cache pin/lock and another process wants it. Get a latch. Someone else wants it. Latch contention. Latch contention builds. Spin, sleep, another process, more latch contention, ... Take a systemstate dump to see what is happening.
>From systemstate dump. (I thought I once saw a document on metalink to help read these things, but I can't find it. Is that FTMN??) I searched on the addresses of all 5 library cache latches. All Processes which were holding a library cache latch were waiting on a shared pool latch (address 8000ebec). Does a process State Object hold a latch independent from a transaction? Don't know where to go from here. The beginning of the systemstate dump for the process which holds the shared pool latch follows:
PROCESS 23:
int error: 0, call error: 0, sess error: 0, txn error 0 (post info) last post received: 3648316968 0 1
last post received-location: kslfre
last process to post me: d9522cd0 4 0
last post sent: 2147544044 0 1
last post sent-location: kslfre
last process posted by me: d9525220 41 0
(latch info) wait_event=0 bits=90
holding 8000ebec shared pool level=7
Location from where latch is held: kghfrunp: alloc: clatch nowait:
Context saved from call: 0
state=busy
holding 8000c114 row cache objects level=4
Location from where latch is held: kghfrunp: clatch: nowait:
Context saved from call: 0
state=busy
SO: d95991a0, type: 3, owner: d9523624, pt: 0, flag: INIT/-/-/0x00
(session) trans: 0, creator: d9523624, flag: (41) USR/- BSY/-/-/-/-/-
DID: 0000-0000-00000000, short-term DID: 0000-0000-00000000
txn branch: 0
oct: 47, prv: 0, user: 20/SYSADM
O/S info: user: dp943, term: PHSPSPAPP8, ospid: 2028:3424, machine: PARTNERS\PHSPSPAPP8
program: PSAPPSRV.exe
address=d974e628, number=6a, tries=1
----------------------------------------
SO: d97c6674, type: 36, owner: d95991a0, flag: INIT/-/-/0x00
LIBRARY OBJECT PIN: pin=d97c6674 handle=da02bb1c mode=X lock=d97d5c84
user=d95991a0 session=d95991a0 count=0 mask=007d savepoint=68 flags=[00]
----------------------------------------
SO: d97d5c84, type: 35, owner: d95991a0, flag: INIT/-/-/0x00
LIBRARY OBJECT LOCK: lock=d97d5c84 handle=da02bb1c mode=N
call pin=0 session pin=d97c6674
user=d95991a0 session=d95991a0 count=1 flags=PNS/[08] savepoint=68
LIBRARY OBJECT HANDLE: handle=da02bb1c
namespace=CRSR flags=RON/KGHP/PN0/[10010000]
kkkk-dddd-llll=0000-0000-0000 lock=N pin=X latch=2
lwt=da02bb34[da02bb34,da02bb34] ltm=da02bb3c[da02bb3c,da02bb3c]
pwt=da02bb4c[da02bb4c,da02bb4c] ptm=da02bba4[da02bba4,da02bba4]
ref=da02bb24[cac6f6d8,cac6f6d8] lnd=da02bbb0[da02bbb0,da02bbb0]
LIBRARY OBJECT: object=dadfc344
type=CRSR flags=EXS/BCM[0001] pflags=NST [101] status=VALD load=0
DEPENDENCIES: count=4 size=16
ACCESSES: count=1 size=16
TRANSLATIONS: count=1 size=16
DATA BLOCKS:
data# heap pointer status pins change
----- -------- -------- ------ ---- ------
0 c9cee16c dadfc1b4 I/P/A 0 NONE
2 dadfc3d8 c9c4bb3c I/P/A 1 NONE
3 dadfc438 0 -/P/- 1 NONE
4 dadfc498 cf49aef0 I/P/A 1 NONE
5 dadfc104 0 -/P/- 1 NONE
6 dadfc164 cb43e3e4 I/P/A 1 NONE
----------------------------------------
SO: d97cc9cc, type: 35, owner: d95991a0, flag: INIT/-/-/0x00
LIBRARY OBJECT LOCK: lock=d97cc9cc handle=cac6f768 mode=N
Any ideas are welcome. If you made it this far, thanks for the perseverance and time you already invested.
Henry
-- Archives are at http://www.freelists.org/archives/oracle-l/ FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html -----------------------------------------------------------------Received on Tue Apr 27 2004 - 21:05:20 CDT
 |
 |