[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: innd & mmap



On Thu, 28 Nov 1996, Miquel van Smoorenburg wrote:

> You (Craig Sanders) wrote:
> > my news box is regularly throttling with the log message "File exists
> > writing symlinking article file -- throttling".
> 
> How's your disk? Do you see many ext2 errors?

yeah, a few

there's some weird corrupted files in /var/spool/news/control, too:

c---r-----   1 8736     30054    116,  32 Apr  4  1987 504023
cr-Sr-sr-x   1 11565    8224      32,  32 Jan 30  1987 504024
br-x-w----   1 26740    28270    110, 105 Mar 21  2029 504027
br-xr----t   1 11891    28015    108, 101 May 22  2031 504028
br-sr-srwx   1 28193    28265    112,  45 May 22  2031 504029
c---r-----   1 28499    30068     32, 101 Mar 30  2029 504048
br-xrwS---   1 26994    8294     101, 109 Mar  9  2029 504049
br-xrwSr--   1 14962    28520    104, 116 Nov 26  2023 504050

These can't be deleted, moved, chmod-ed, chattr-ed, or anything.
Any attempt to do something to the files results in "Operation not
permitted" (btw, this is what they look like AFTER running e2fsck on the
partition - unmounted, of course - they used to have unreasonably large
sizes before that...4gb and more)

I'm going to reformat the drive when i have the time to bring the system
down for a few hours.

> > >                 Trying to use MMAP on Linux
> > 
> > Is this still true for linux?
> 
> No. we are running INN with mmap() on on all our news servers (inn 1.4unoff4,
> linux-2.0.25 or so) and we have 5 feeds in totalling about 400 Mb/day, 30
> or 40 feeds out (small ones like linux.* and a couple of "full" ones).

OK, sounds good to me.

Are any of these feeds uucp?

My main newsfeed is uucp over a trailblazer modem - bandwidth is
expensive here in australia and there's better things to use it for than
usenet news.

> > Has inn for debian been compiled with MMAP turned off?  If not, should
> > this be reported as a bug against inn? 
> 
> AFAIK, nobody has complained about this yet - however I have seen the
> same problem in the following cases:

from what you say, it isn't a problem anyway.

> 1) rmgroup immideately followed by a makegroup control message (resets
>    the counters to 0 but doesn't delete the articles)


> 2) Harddisk problems, especially 4Gb disks go bad very fast IME.

i suspect that this is the likely cause. either the hard disk or the
ide controller. it's a new quantum bigfoot, 2.5GB, on an ASUS SP3G PCI
motherboard. The bigfoot drive only got put in a few months ago, to
replace a seagate 1GB scsi drive (which was big, fast, and extremely
noisy - the bigfoot is slower but bigger and wonderfully quiet)

I used to get the same problem with the old seagate scsi drive, just not
as often...which could be due to the fact that i'm now keeping news for
twice as long.

I've kludged a temporary fix. I run 'ctlinnd renumber' before and after
news.daily in my fixnews.nightly script. My fixnews.weekly script
rebuilds the history file and then runs fixnews.nightly.

it's not a real fix, but it stops the error messages, and prevents
/var/spool/news/.incoming from filling up with hundreds of megabytes of
compressed rnews batches which inn cant process due to being throttled
(news is very low priority - sometimes i dont notice theres a problem
for a day or two).


> BTW, I'm not planning a 1.4unoff4-2 at this time (though I have one
> available) but instead I want inn1.5 in "bo". Under Linux, it doesn't
> make much difference if you have mmap() on or off because the buffer
> cache is so good. If people prefere to have it turned off, that's no
> problem with me.

1.5 went out of beta just recently, i saw the release version on the ISC
INN web page.


Craig


--
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
debian-user-REQUEST@lists.debian.org . Trouble? e-mail to Bruce@Pixar.com


Reply to: