On Fri, Oct 10, 2008 at 11:43:39AM +0200, Laszlo Nagy wrote:
>>>> If find / -sx is running and is consuming all CPU, what is the 
>>>> value of vfs.ufs.dirhash_mem: 
>>>> # sysctl -a | grep dirhash
>>> shopzeus# sysctl -a | grep dirhash
>>> vfs.ufs.dirhash_docheck: 0
>>> vfs.ufs.dirhash_mem: 2095818
>>> vfs.ufs.dirhash_maxmem: 2097152
>>> vfs.ufs.dirhash_minsize: 2560
>>>> Make sure vfs.ufs.dirhash_mem: is not close to vfs.ufs.dirhash_maxmem:
>>> All right. It is close to it. Which one should I increase? I put this 
>>>  into /etc/sysctl.conf:
>>> vfs.ufs.dirhash_maxmem=8228608
>>> Would it be scufficient?
>> We don't know, and can't tell you.  You'll have to monitor
>> vfs.ufs.dirhash_mem occasionally to see if you start to reach
>> vfs.ufs.dirhash_maxmem.
>> I have a tendency to use vfs.ufs.dirhash_maxmem=16777216, which is
>> 16384*1024 (16MBytes).
>> I'm not fully confident this is what's causing your problem, but it's
>> definitely a recommendation by Johan.
> Thank you very much! Probably you are right. Our users use shared IMAP  
> folders and sometimes they keep ten thousands of messages in one folder.  
> I have increased dirhash_maxmem to 64MB and see what happens.
> Unfortunately, I cannot play with the hardware because it is in a server  
> park, and it must be up 99.99% on workdays.
> I hope dirhash will solve the problem. I'm setting this to [SOLVED] and  
> come back if it happens again. (Maybe on monday?)
> By the way, there is nothing in /etc/periodic that would execute "find /  
> -sx". Can somebody explain what is this for, and why it was started by  
> root? Is it being used instead for enumerating files in a directory,  
> when dir hash is full?

Firstly, I see a periodic(8) job that DOES use find -sx, which means
your attempt to track it down was faulty, and your syntax should have
been "find -sx /" not "find / -sx".  See here:

/etc/periodic/security/100.chksetuid:   find -sx $MP /dev/null -type f \

$MP == mountpoint, e.g. /, /var, or any other mounted filesystem.

So, what you saw was the periodic check looking for setuid-root

Secondly, the kernel does not spawn userland processes like find(1).

Thirdly, dirmem and dirmem_max are *pure* kernel things.  What they do
is control the amount of memory used for directory structure caching;
rather than continually hit the disk every time and spend all that time
handling directory contents, the kernel can cache previously-fetched
contents in memory.

| Jeremy Chadwick                                jdc at parodius.com |
| Parodius Networking                       http://www.parodius.com/ |
| UNIX Systems Administrator                  Mountain View, CA, USA |
| Making life hard for others since 1977.              PGP: 4BD6C0CB |

freebsd-questions@freebsd.org mailing list
To unsubscribe, send any mail to "[EMAIL PROTECTED]"

Reply via email to