RE: [Condor-users] Bad file descriptors

[Adam Thorn <alt36@xxxxxxxxxxxxxxxx>]

On Fri, 26 Aug 2005, Michael Yoder wrote:

> Wow, that's an interesting one.  That error is way down in the bowels of
> DaemonCore, and might be some sort of internal problem.  Try setting
>
> MASTER_DEBUG = D_ALL

Thanks for the suggestion.  I can't say for sure that this is always true, 
but it seems that I only get this error when the daemons attempt to start 
at boot-up; if I then start them by hand afterwards I think it always 
works.  Anyway, I set the debug flag you suggest and compared the logs 
obtained for a successful and failed start.  The first few hundred lines 
are DaemonCore saying that it's registering commands and then timers. 
Once that's done, that section finishes with

leaving DaemonCore NewTimer id=0
DaemonCore Timeout() Complete, returning 5

It's immediately after that that the error (and the first difference 
between the logs of a succesful and failed startup) is logged:

ERROR "DaemonCore: select() returned an unexpected error: 9 (Bad file 
descriptor)" at line 1840 in file daemon_core.C

There are then a few tens of lines about SECMAN and security policy, at 
the end of which it complains

condor_read(): timeout reading buffer.
Stream::get(int) failed to read padding
SECMAN: negotiation failed. trying the old way...

which I could believe could be related to that error message. I'm happy to 
post the log if anyone could have a quick look at it, but guess it might 
not be appropriate to send such a sizable amount of info to this list....

Adam