[Dxspider-support] Cluster Lock-Up

[Dirk Koopman]

Danilo Brelih wrote:
> Anthony (N2KI) pravi:
> 
>> Unfortuneately this is an intermittant problem. The only good lead is that
>> it happens at the same time (2351Z) when it does.
> 
> We crashed tonight same time seems same problem !
> 
> http://s50clx.infrax.si:41115/statistika/msg.html
> 
> DX Spider Cluster version 1.54 (build 0.164) on Linux
> Copyright (c) 1998-2007 Dirk Koopman G1TLH
> S50U de S50CLX  9-Oct-2007 1225Z 2.3522  dxspider >
> 

I am embarrassed to say that there appears to be a long standing error 
in the PC9x sentence deduping code that caused a enormous loop last 
night. This caused a huge spike in the amount of traffic. Something that 
I have not seen before, somewhat surprisingly.

As some of you may know, I am using the number of seconds in a day as a 
deduping mechanism. This means that at midnight, the number rolls over 
from 86399 -> 0. Last night we had a loop because we were getting PC92, 
from an (italian) station whose clock was / is wildly out, PC92 C 
records with a time of 85967 and an A record of 518. Now the time 
checking that was being done only worked for clocks that were nearer UTC 
than this station.

I believe that I have fixed this so I don't rely on other nodes being 
syncronised to UTC (although you should really all be).

This means, sadly, and rather embarrassingly, I have ask people to 
upgrade again. However I would wait until tomorrow (or at least 02:00Z) 
to see whether, this time, I really have fixed it...

Mea culpa

Dirk

PS Please just upgrade using CVS or CVSlatest.zip/tgz silently. You 
don't need to tell the list that you have done it. I will know (or at 
least I can find out by looking at the PC92s you send) anyway.