2003-09-12 16:48:56

by Alberto Bertogli

[permalink] [raw]
Subject: file_lock_cache won't shink down


Hi there!

I've put 2.6.0-test5 yesterday on a mail server and today I noticed that
there was a huge amount of memory 'missing' from vmstat.

So I took a look at slabinfo and found out that file_lock_cache has a huge
amount of objects that won't shink down even under memory pressure.

The line, as it is now, looks like:
file_lock_cache 5828600 5828600 96 40 1 : tunables 120 60 \
8 : slabdata 145715 145715 0

The whole slabinfo is attached.

This is a dual Pentium III with 1Gb RAM and three SCSI disks in two
Adaptec aic7896/97.

I don't know if I can hold the server up for much longer so please if you
want me to get any other information let me know.


Thanks a lot,
Alberto



2003-09-13 00:51:25

by Matthew Wilcox

[permalink] [raw]
Subject: Re: file_lock_cache won't shink down


Hi, Alberto. Please try
http://ftp.linux.org.uk/pub/linux/willy/patches/flock-memleak.diff

--
"It's not Hollywood. War is real, war is primarily not about defeat or
victory, it is about death. I've seen thousands and thousands of dead bodies.
Do you think I want to have an academic debate on this subject?" -- Robert Fisk

2003-09-13 06:59:21

by Alberto Bertogli

[permalink] [raw]
Subject: Re: file_lock_cache won't shink down

On Sat, Sep 13, 2003 at 01:51:23AM +0100, Matthew Wilcox wrote:
> Hi, Alberto. Please try
> http://ftp.linux.org.uk/pub/linux/willy/patches/flock-memleak.diff

I've just put it in, it seems to be working so far but I'll tell you
tomorrow for sure after it's been under some pressure.


Thanks a lot,
Alberto


2003-09-13 13:09:24

by Alberto Bertogli

[permalink] [raw]
Subject: Re: file_lock_cache won't shink down

On Sat, Sep 13, 2003 at 03:59:14AM -0300, Alberto Bertogli wrote:
> On Sat, Sep 13, 2003 at 01:51:23AM +0100, Matthew Wilcox wrote:
> > Hi, Alberto. Please try
> > http://ftp.linux.org.uk/pub/linux/willy/patches/flock-memleak.diff
>
> I've just put it in, it seems to be working so far but I'll tell you
> tomorrow for sure after it's been under some pressure.

It's been tested enough to confirm that it's working great!

Thanks a lot for the patch, I really hope this goes into mainline as it
fixes a very serious and easily reproduceable bug.


Thanks again,
Alberto