[OpenAFS-devel] callback handling problem

[Hartmut Reuter]

In older versions of callback.c was no "if ( !" around the call to
ClearHostCallbacks_r. Therefore the loop was left. I suppose we need a
line

hp1 = hp;

after the "return,". That would lih_r() start searching with the next
host.

Here the original part:

    do {
        lih_host = 0;
        h_Enumerate_r(lih_r, (char *)hp1);
        hp = lih_host;
        if (hp) {
            cbstuff.GSS4++;
            if ( ! ClearHostCallbacks_r(hp, 0 /* not locked or held */)
)
                return;
        } else {
            hp1 = hostp;
            cbstuff.GSS1++;
            ViceLog(5,("GSS: Try harder for longest inactive host cnt=
%d\n", i));
            /*
             * Next time try getting callbacks from any host even if
             * it's deleted (that's actually great since we can freely
             * remove its callbacks) or it's held since the only other
             * option is starvation for the file server (i.e. until the
             * callback timeout arrives).
             */
            i++;
        }
    } while (i < 2);


Hartmut

Thomas Mueller wrote:
> 
> Hi folks,
> 
> we still have the problem which I reported last October.
> I now can say what happens but I have no idea why it happens.
> 
> Occasionally the fileserver ends up with all threads running an infinite
> loop in GetSomeSpace_r() in viced/callback.c (do { ... } while (i < 2);)
> 
> If
> 
> h_Enumerate_r(lih_r, (char*)hp1);
> 
> finds at least one host which passes all test in lih_r(...)
> then
> 
> ClearHostCallbacks_r(hp, 0)
> 
> is called, but this function returns immediately (return value 1)
> if this host entry is locked and the next round in this loop starts.
> 
> So if hostList contains only a few (sometimes only one) entries
> with host->cblist != 0 AND those entries are locked this thread keeps in
> that do-while-loop.
> 
> Do you think, it would be a solution to modify lih_r() a little bit so
> that it doesnt pick up locked host entries?
> 
> Thomas.
> --
> -------------------------------------------------------------------------
> Thomas Mueller, TU Chemnitz, Universitaetsrechenzentrum, D-09107 Chemnitz
> e-mail: Thomas.Mueller@hrz.tu-chemnitz.de
> -------------------------------------------------------------------------
> 
> _______________________________________________
> OpenAFS-devel mailing list
> OpenAFS-devel@openafs.org
> https://lists.openafs.org/mailman/listinfo/openafs-devel

-- 
-----------------------------------------------------------------
Hartmut Reuter                           e-mail reuter@rzg.mpg.de
					   phone +49-89-3299-1328
RZG (Rechenzentrum Garching)               fax   +49-89-3299-1301 
Computing Center of the Max-Planck-Gesellschaft (MPG) and the
Institut fuer Plasmaphysik (IPP)
-----------------------------------------------------------------