named dying pollfd

[Jim Reid]

>>>>> ">" == dchench  <dchench at cablespeed.com> writes:

    >> My dns service has died twice and the only error message that I
    >> have is in the /var/adm/messages file....  
    >> "dns10 named{24767}: pollfd realloc(1634734080) failed" 
    >> I am on a sun netra t1 ultra
    >> sparcsolaris 5.6 running bind4.9.4 I can't seem to find out
    >> what is causing the problem..  pollfd chunk is related to bind
    >> 8 named.conf file -option my secondary is running bind 8 could
    >> this be causing the problem.  Thanks for any info.

Your message makes no sense. If you're running BIND4, why are you
talking about BIND8 and named.conf? BIND4.9.4 is very, very dead. It's
also got security holes. Upgrade to 8.2.3 or 9.1.1. The ISC's web site
has the details. The name server implementation on the slave
(secondary) server is irrelevant. They are all supposed to speak the
same zone transfer protocol. Even so a zone transfer should never
crash another name server.

I've just done a quick grep of the BIND4.9.7 source code and there's
no reference to pollfd or any syslog messages about failing realloc()
calls. It's unlikely, but not impossible, that this error message was
taken out in BIND4.9.[56]. If the message was never in the ISC code,
why your server is reporting it is a mystery. Maybe it's an
enhancement by Mr. Solaris to the base code? Why not ask Sun?