Re: syslogd

["Barrett G. Lyon" <blyon netpr com>]

I do infact have an inode problem, and a fds problem.  I have kernel panics
where I get messages like 'no free inodes', or the system just seems to
lock up, but its really out of fds, so anything needing disk IO just stops
working.  My best uptime for the box is 9 days, but it has a very serious
load, but the adv. uptime is around 5 days.

I think I just have a bad motherboard, because anything in the PCI bus
seems to just be fucked up.

I did fix ftp however, I replaced wu-ftp with proftpd 1.1.6pl2 and that
works awesome. 

Another thing that seems to eat the system alive, is when it starts to
teeter-totter on the edge of running out of FDS, syslog starts to go nuts.
It can't write out to some files, so it starts to loop messages at 1000nds
per minute.  I don't even feel like looking up the exact message, but I
have seen lots of people post pastes of the same thing I got.

The only soution I have/had (and its a very very poor one at that) is to up
file-max and inode-max to a huge ungodly number.  

I am thinking about trying a dev kernel 2.1.122 maybe that will fix some of
my problems.  Ohh, that reminds me, what kernel do you run?

-Barrett

>Quoting Barrett G. Lyon (blyon@netpr.com):
>> Nobody knows what eats up FDS and inodes, but it happens to me also.  I
>
>Nothing eats up file descriptors.  file-nr is always around 470-500, even
when
>syslogd is spewing.
>
>> also noticed huge problems with wu-ftpd.  I think there are problems in the
>
>I'm no longer doing any ftping on that system - I moved the ftp setup over to
>the Intel Linux box which attaches to my cable modem instead of the alpha
>which attaches to the regular phone modem.  (Gee, I wonder why?)
>
>> Linux kernel for the 21164sx support PCI maybe?  Maybe there is some
>> relation to the odd clock problems that were 'fixed' with milo.  I wonder
>> if people that have the clock that runs 2X faster than normal, and have the
>> milo fix are the same ones that have the inode/fds/pci problems.
>
>I have no problems with my clock.  I set it every day against my provider's
>time server (which in turn is set to the NIST clock), and it varys by less
>than a second each time.  I do have the latest milo, however.
>
>One thing which I doubt is significant but I thought I'd mention is that this
>time when I booted, I booted originally into single user mode because I
had to
>fix something in the news setup before the news server started.  I had
>forgotten about that, but I noticed it when I rebooted to fix the syslogd
>problem.  I guess shutdown -r causes milo to boot with whatever kernel
>arguments it had been booted with last, and this time it went into single
user
>again.
>
>Previous to this 11 day uptime, I had a 39+ day uptime with no syslogd
>problems (that one was rebooted due to the power fluctuating during an
>electrical storm).
>

--
Barrett G. Lyon                               PGP Key: http://www.doxx.net/pgp
Data & Network Security Consultant                Fax: 310-737-0196
Network Presence, LLC                           Email: blyon@netpr.com
--