Re: ??: postgres cpu 100% need help - Mailing list pgsql-general

From Merlin Moncure
Subject Re: ??: postgres cpu 100% need help
Date
Msg-id CAHyXU0xcCk1DD_oC5P-6b1uFQ2si12Kp3wHduG=LDZt_UK4UUQ@mail.gmail.com
Whole thread Raw
In response to Re: ??: postgres cpu 100% need help  (Kevin Grittner <kgrittn@gmail.com>)
List pgsql-general
On Mon, Nov 23, 2015 at 9:30 AM, Kevin Grittner <kgrittn@gmail.com> wrote:
> On Mon, Nov 23, 2015 at 12:20 AM, 657985552@qq.com <657985552@qq.com> wrote:
>
>> [root@pg1 ~]# cat /sys/kernel/mm/transparent_hugepage/enabled
>> [always] madvise never
>> [root@pg1 ~]# cat /sys/kernel/mm/transparent_hugepage/defrag
>> [always] madvise never
>
> There's your problem.  You need to set those to "never".  You can
> get immediate relief by echoing 'never' to those pseudo-files, but
> you need to configure your system to set them at OS boot time, too.

yup.  IMNSHO automatic enabling of THP defrag is one of the worst
performance related mistakes I've ever seen in the history of the
linux kernel (although maybe the blame lies with the distros...I'm not
sure).  It caused a number of terrible outages here before we figured
out the underlying cause.  (of course, everyone blamed postgres until
then)

merlin


pgsql-general by date:

Previous
From: Merlin Moncure
Date:
Subject: Re: using a postgres table as a multi-writer multi-updater queue
Next
From: John R Pierce
Date:
Subject: Re: using a postgres table as a multi-writer multi-updater queue