Re: Occasional insert slow down with CBC waits.

From: Jonathan Lewis <jlewisoracle_at_gmail.com>
Date: Thu, 14 Apr 2022 17:48:59 +0100
Message-ID: <CAGtsp8=qmW_UhM3gK6T3j1Z5inCxOj+gLC4kt3UP_u52xR-s6Q_at_mail.gmail.com>



One of the buffers has state PI, which means you must be running RAC, and at some recent point in the past another instance has been calling for this block. (If you're not running RAC then seeing a PI is probable the problem you need to address first).

It's possible that the CBC latch wait is a side effect of having more than one instance inserting into the same LOB segment at the same time. Waiting for a latch, though, often means that your CPU is overloaded - so perhaps the LMS processes are getting too busy and putting pressure on the system when the rate of insert gets too high.

Regards
Jonathan Lewis

On Thu, 14 Apr 2022 at 16:41, Goti <aryan.goti_at_gmail.com> wrote:

> Hi,
>
> ENV: Oracle 12.1.0.2 on RHEL 7.
>
> There is one single pconventional insert which slows down occasionally.
> The wait event during the slow down is "latch: cache buffer chains". Using
> Tanel Poder's latchprofx.sql , it was identified that the wait event was
> mainly for the LOB segment with 1 level bmb as the block class. Details are
> there in the gists below. Is there any way to identify what is causing this
> and how the issue can be fixed?
>
> https://gist.github.com/aryangoti/06bcd2ead5efbea910afa4e7e5af5e4a
>
> https://gist.github.com/aryangoti/7d462225877c648f8a88b669b4b68b08
>
>
> Thanks,
>
> Goti
>

--
http://www.freelists.org/webpage/oracle-l
Received on Thu Apr 14 2022 - 18:48:59 CEST

Original text of this message