Re: server crashing

[Al - Image Hosting Services <azick%zickswebventures.com@localhost>]

Hi,

Sorry about the delay in getting back to you.

On Tue, 23 Aug 2011, Matthias Scheler wrote:

> Date: Tue, 23 Aug 2011 20:15:35 +0100
> From: Matthias Scheler <tron%zhadum.org.uk@localhost>
> To: Al Zick <al%familysafeinternet.com@localhost>
> Cc: netbsd-users <netbsd-users%netbsd.org@localhost>
> Subject: Re: server crashing
>
> On Mon, Aug 22, 2011 at 02:50:12PM -0400, Al Zick wrote:
> > I just leased a new server with NetBSD 5.1 on it. I have over 600 GB
> > of websites on it. It seems to run fine, serving pages happily until
> > I try to back it up with rsync over an nfs to another server on the
> > back end network.
>
> Is your machine the NFS server or client in this setup? If it is the
> NFS server you might suffer from this problem:
>
>         http://gnats.netbsd.org/45093
>         http://releng.netbsd.org/cgi-bin/req-5.cgi?show=1644
>
> The above problem has been fixed in the "netbsd-5" branch.

This server is running NetBSD 5.1. I did an install from source maybe 6 
weeks ago. Last night there were two consecutive crashes. It does 
seem to be related to nfsd and sakva.

From NetBSD Problem Report #45093:
> Fix:
        workaround: either disable page loaning in nfs server, or
        change uvm_unloanpage() to use yield() instead of kpause()
        (the later has been confirmed to work around the issue).

It looks like this work around requires editing the source. Could someone 
outline how to, or at least tell me what source file?

From NetBSD Problem Report #45093:
sokva_reclaim_callback() and sokvareserve() may have the same issue,
        if called with the socket locked.

This greatly concerns me. I have had major problems with processes being 
in a state of sakva in NetBSD 4. I added: /sbin/sysctl -w 
kern.somaxkva=75497472 I also added this to the server that keeps 
crashing. It seems to help, but I think there must be a better solution.

What else can I try in an effort to make this server stable?

Thanks,
Al