[Dxspider-support] questions about 'zombie' spider nodes

[Dirk Koopman]

On Mon, 2006-09-25 at 15:12 -0400, Alan D. Snyder wrote:
> Dirk: the snip below was sent to the mailing list, but it bounced w/o
> telling me why, so am resending it to you directly.  probably a more
> appropriate destination then the list as a whole.
> 
> thanks .... Alan KF3B
> 
> > 
> >   
> > From: alan Snyder KF3B <kf3b at hotmail.com> 
> > To: dxspider-support-request at dxcluster.org 
> > Subject: 'zombie' node 
> > Sent: Monday, September 25, 2006 12:10 PM 
> >                               Hi guys 
> >                                   
> >  Recently changed spider's host pc (to a slightly slower box), and
> >    made other infrastructure changes.  Also wanted to understand
> >             'pinginterval' and 'link' so played a bit. 
> >                                   
> >  Now am in the unenvyable position of having our node 'zombied' at
> >  three of our eight link partners.  That is, they show connected on
> >                       their ends but are not. 
> >                                   
> >   When I issue a 'connect' from my end, get a quick disconnect and
> >   spider says that the node is already connected, because on each
> >        remote end, spider thinks it is connected.  Catch 22. 
> >                                   
> >  This means that each remote sysop needs to disconnect our node at
> >               their end, or at least I think it does. 
> >                                   
> >   First, any ideas on what I did to make this node a 'zombie' so I
> > don't screw up again?  These three nodes have been reliable for many
> > years.  Never have seen this problem before.  And have not upgraded
> >                  spider version in a while either. 
> >                                   
> >   Second, any thoughts on what can be done to 'un-zombie' the node
> >                 without remote sysop intervention? 
> >                                   
> >  Third, shouldn't the fact that the packet does not get accepted on
> >            the remote end indicate something to spider? 
> >                                   
> > If spider's packet initiator doesn't check responses, wouldn't it be
> >           a bit more efficient to change from TCP to UDP? 
> >                                   
> >     Well, it's Monday and I haven't caused trouble in too long. 
> >                                   
> >                        Thanks in advance ... 
> >                                   
> >                            73 Alan KF3B 
> >                                   
> 

I think you need to do a 'set/debug raw conn' and see what is going in
terms of the actual data passing back and forth and also what states the
connection gets itself into.

Spider does (or at least should) check its connection status, except
that this is very difficult thing to do on some of the older perls (perl
< 5.6.1).

What have you done to 'pinginterval' then?

-- 
Dirk Koopman <djk at tobit.co.uk>