Re: Re: significant performance hit whenever autovacuum runs after upgrading from 9.0 -> 9.1 - Mailing list pgsql-general

From Tom Lane
Subject Re: Re: significant performance hit whenever autovacuum runs after upgrading from 9.0 -> 9.1
Date
Msg-id 6087.1337791074@sss.pgh.pa.us
Whole thread Raw
In response to Re: Re: significant performance hit whenever autovacuum runs after upgrading from 9.0 -> 9.1  (Lonni J Friedman <netllama@gmail.com>)
Responses Re: Re: significant performance hit whenever autovacuum runs after upgrading from 9.0 -> 9.1
List pgsql-general
Lonni J Friedman <netllama@gmail.com> writes:
> After banging my head on the wall for  a long time, I happened to
> notice that khugepaged was consuming 100% CPU every time autovacuum
> was running.  I did:
> echo "madvise" > /sys/kernel/mm/transparent_hugepage/defrag
> and immediately the entire problem went away.

Fascinating.

> So this looks like a nasty Fedora16 kernel bug to me, or maybe
> postgresql & Fedora16's default kernel settings are just not
> compatible?

I agree, kernel bug.  What kernel version are you using exactly?

> Is anyone else using Fedora16 & PostgreSQL-9.1 ?

I use an F16 box daily, but can't claim to have done major performance
testing with it.  Can you put together a summary of your nondefault
Postgres settings?  I wonder whether it only kicks in for a certain
size of shared memory for instance.

            regards, tom lane

pgsql-general by date:

Previous
From: Tom Lane
Date:
Subject: Re: Migrating from 8.2 to 9.1 : invalid port number
Next
From: Tom Lane
Date:
Subject: Re: FATAL: lock file "postmaster.pid" already exists