tradindexed - Interrupted system call

[Andreas M. Kirchwitz]

Mirek Luc wrote:

 >>>>> Could this just be a symptom of "normal" NFS behavior under heavy
 >>>>> load: Lack of NFS processes relative to the queue of requests?
 >>>> No. And after some more debugging and researching, I don't even
 >>>> think it's a NFS problem at all.
 >>> You're wrong. It's the problem(?) with I/O. The Solaris operating system
 >>> told you that very clearly.
 >>
 >> Where is it documented that you get EINTR only if the file is
 >> on an NFS volume?
 > I've never wrote that.

Sorry for that implication. It wasn't meant to offend you.
Your article simply was the last one in the thread at that
time. ;-)

This discussion is just a little bit confusing. Let me try
to repeat it...

Bettina found those "Interrupted system call" errors.
Under the assumption that INN should run on Solaris 2, those errors
are caused by a bug in the source code not properly checking for the
EINTR return code (which you explained very good).

Then somebody said the "forbidden" word "NFS" - and now it looked like
"NFS" is the problem. If I get your articles right, you probably agree,
that NFS is not the problem, but just makes the problem (no check for
EINTR) occur.

However, you already wrote something similar today to Russ Allbery.
While Russ trusts Stevens, that open() should not be interrupted,
you (and me) say, that EINTR is documented in the man page and must
be checked.

I guess, the question now is: Who is authoritative? Stevens or the
man pages (shipped with Solaris, and available on http://docs.sun.com).

	Greetings, Andreas