Re: what's mean of "enqueue hash chains" latch?

From: <jo_holvoet_at_amis.com>
Date: Mon, 10 Jan 2005 10:03:31 +0100
Message-id: <OF02966E68.10C46353-ONC1256F85.00316128@eu.amis.com>

I'm jumping in a bit late, but last year I filed a tar due to an enqueue hash chains latch problem (related to a deadlock). But you didn't mention a deadlock so this probably has nothing at all to do with your issue.

>From the description (8.1.7.4.1 on Win2k BTW):

---
---

User runs a program to load data into the DB. At a certain point, this causes
an ORA-00060 (deadlock). So far no real problem, but a few minutes later (while
deadlock cleanup is still going on), every other session in the DB hangs until
deadlock cleanup is completed. 10046 traces of the hanging processes show that
they are all waiting on 'latch free'; trace files contain almost nothing but
lines of the type :
...

WAIT #6: nam='latch free' ela= 207 p1=635234948 p2=13 p3=136
WAIT #6: nam='latch free' ela= 206 p1=635234948 p2=13 p3=137
WAIT #6: nam='latch free' ela= 206 p1=635234948 p2=13 p3=138
WAIT #6: nam='latch free' ela= 207 p1=635234948 p2=13 p3=139
WAIT #6: nam='latch free' ela= 206 p1=635234948 p2=13 p3=140

...
The latch with number p2=13 is the 'enqueue hash chains' latch.
---
---

And Oracle's answer :

---
---

The problem you see is when the ora-60 error will generate an systemstate dump automatically.
Then we need to take enqueue locks of the structure and this is causing your problems.

This behaviour can be changed by setting event 10027 in your init.ora file,

Put this is your init.ora file.
---
---

mvg/regards

"Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> Sent by: oracle-l-bounce_at_freelists.org
01/09/2005 23:40
Please respond to jonathan

        To:     <oracle-l_at_freelists.org>
        cc: 
        Subject:        Re: what's mean of "enqueue hash chains" latch?

Note also how:

    enqueue gets and releases = 23M
    db block gets = 23 M
    execute count = 23M

You can't help feeling that that's not a complete coincidence. I think your trigger theory sounds quite good at the moment,
only

I'd like to find that 23M executions in v$sql. Why isn't it appearing in statspack ?
There's no redo generated to match the volume of locking.

I'd like to see the snapshot of x$ksqst
One thought - for parallel DML, maybe
there are lots of DX locks being taken
and dropped.

Regards

Jonathan Lewis

http://www.jlcomp.demon.co.uk/faq/ind_faq.html The Co-operative Oracle Users' FAQ

http://www.jlcomp.demon.co.uk/seminar.html Public Appearances - schedule updated Dec 23rd 2004

Original Message ----- From: "Riyaj Shamsudeen" <rshamsud_at_jcpenney.com> To: <oracle.unix_at_gmail.com>; <jonathan_at_jlcomp.demon.co.uk> Cc: <oracle-l_at_freelists.org> Sent: Sunday, January 09, 2005 9:09 PM Subject: RE: what's mean of "enqueue hash chains" latch?

Eygle
Looking at the excel spreadsheet, one of the child latch has 2.4 Billion gets and other child latches have less then 76K latch gets. So, sessions are contending to lock one table. Looking at the statspack report, you have numerous enqueue gets and releases too. Is there any foreign key constraint on this table ? If yes, do you have any index on that foreign key column ? Further, what mode locks are being acquired ?
do you have any triggers on this table, by any chance ?

Thanks
Riyaj "Re-yas" Shamsudeen
Certified Oracle DBA

--

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

--

http://www.freelists.org/webpage/oracle-l Received on Mon Jan 10 2005 - 02:59:05 CST