[Raw Msg Headers][Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: router, mailing lists, memory and suns...
> It appears that the router can corrupt disks on SunOS 4.1.2 and
> probably other SunOS releases. This appears related to memory bloat in
> the router when expanding mailing lists. At one time I believed this
> only happened with Rayans zmailer, and I switched to Matti's. While
> Matti's has some wonderful improvements, it seems that slowly, over
> time the router grows in memory usage until it scribbles all over the
> disk.
You can grow it rather rapidly to that large mode too, but
then it stabilizes. I have heard of real leak on Linux systems
with DNS resolver appearing to be the culprit.
There are two different reasons for the bloat:
- using zmsh script attribute processing does about
N^2 memory usage in respect to the number of recipients
- ZMSH scripting language has a bit inefficient memory
management, while being able to do mark()/release()
type of thing on objects, it copies them a bit too much..
> This is obviously a SunOS bug, but we don't have a maintanence
> contract and want to stick with something more BSD-ish than their
> latest releases, so I was hoping someone here could suggest something,
> or perhaps Matti will be able to track it down.
>
> Here is a ps -auxww after expanding two large (> 1500) lists:
>
> root 26148 2.7 5.0 5508 1144 ? R Feb 14 4:38 /usr/local/lib/zmailer/bin/router -dkn 3
> root 26147 2.3 5.5 5516 1252 ? R Feb 14 4:36 /usr/local/lib/zmailer/bin/router -dkn 3
> root 26139 0.0 4.2 888 960 ? S Feb 14 2:43 /usr/local/lib/zmailer/bin/router -dkn 3
For a comparison, on SunOS 4.1.4: (vger.rutgers.edu)
bash# ps auxww|grep router
root 16892 2.0 4.4 6864 3104 ? R Feb 17123:16 /vger/u3/zmailer/bin/router -dkn 4
root 16894 2.0 4.1 6872 2880 ? R Feb 17123:18 /vger/u3/zmailer/bin/router -dkn 4
root 16889 2.0 6.2 6976 4424 ? R Feb 17122:57 /vger/u3/zmailer/bin/router -dkn 4
root 16895 1.6 4.8 6752 3416 ? R Feb 17123:40 /vger/u3/zmailer/bin/router -dkn 4
There are several lists of about 1500-1700 recipients apiece..
> I now have to kill and restart the router or else spend a morning
> restoring disks and listening to user complaints :-(
Restoring disks ? Does your partitioning have some overlaps ?
I recall several bug-fixes to SunOS 4.1.3 with regards to UFS
panics, in fact one of my systems was practically unusable until
those fixes were applied.
> Thanks for any advice - I wish I had time to debug this myself, but I
> am presently overloaded at work.
>
> The router version is 2.99.20.
At ..25 I found one nasty bug at mailbox TA, which did caused
mailbox zero-truncation, in case the write failed.
> Thanks again,
> sdb
> --
> sdb@ssr.com
/Matti Aarnio <mea@utu.fi> <mea@nic.funet.fi>