Strange network fault

[Andrey Oktyabrskiy]

On 29.12.2013 22:09, Matthew Dillon wrote:
> I've never seen the rtsock warning before.  My guess is that the network
> problems are related to the cluster mbuf running out of memory.  So the
> question is... what is eating up all the cluster mbufs?
May be Transmission? At least I see Transmissions coredumps from time to 
time.

> There are a couple of things that can be used to monitor this.  'netstat
> -m' gives you current statistics.  Make sure they aren't being exhausted
> just from the initial ring loads (check after a fresh boot), perhaps
> there are simply not enough left to handle the TCP sockets.
>
> Second thing is e.g. 'netstat -p tcp -n' and observe the transmit and
> receive buffer levels to see if the clusters are being eaten up by
> active TCP streams.
Thank you. I will try to see it.

> Also, how much memory does this laptop have?
2GB

> If it looks like the machine is configuring too few cluster mbufs you
> can increase the number in /boot/loader.conf with something like this:
>
> kern.ipc.nmbclusters=16000
I can, but I want to see - does it make OS unreachable from network or 
killing Transmission solves the problem?

> On Sun, Dec 29, 2013 at 7:51 AM, Andrey Oktyabrskiy <ano at bestmx.net
> <mailto:ano at bestmx.net>> wrote:
>     I have home server on Dell D631 laptop with DragonFly 3.6. There is
>     NFS server with 2x1TB HDD. Today is the second time the network has
>     stopped working with such symptoms (from dmesg):
>     rtsock: received more addr bits than sockaddrs.
>     Warning, objcache (cluster mbuf): Exhausted!
>     dsched_thread_io: destroy race tdio=0xffffffe048fa04e0
>
>     Today it happened when copying large amount of data via NFS. But
>     first time there was no NFS. That time I copied data by SFTP. There
>     is no way to make the network work again - reboot only.
>
>     What other information from me would be useful for debugging?