Conference noted::dnu_osi

2158.0. "dnsclerk core dump" by AUBER::DORNANO () Tue Apr 08 1997 04:02

Hello


Don't want to be rude by posting such elements without many clues,
but I browsed the whole conference without finding a solution to
my customer's problem.

His dnsclerk core dumps but I really don't know why. 
Here are some infos on his version and the result of a quick
analyse under dbx :



Node 0
AT 1997-03-26-11:47:43.303+00:00I-----

Characteristics

    Implementation                    =
        {
            [
                Type ........  = DECnet/OSI for Digital UNIX ,
                Implementation = "DECnet/OSI for Digital UNIX V3.2B-0 (Rev. 23.4
7); Mon Sep  9 17:05:24 EDT 1996"
            ] ,
            [
                Type ........  = Digital UNIX ,
                Implementation = "Digital UNIX V3.2G (Rev. 62); Tue Dec 24 09:30
:28 WET 1996"
            ] ,
            [
                Type ........  = Alpha ,
                Implementation = ""DEC2100_A50""
            ]
        }




C-shell> dbx dnsclerk core.1
dbx version 3.11.8
Type 'help' for help.
Core file created by program "dnsclerk"

warning: dnsclerk has no symbol table -- very little is supported without it


thread 0xfffffc00035c59a0 signal Floating point exception at >*[__fp_class_f, 0x
3ff8051b0f8]    cmovne  r16, 0x5, r0
(dbx) tstack

Thread 0xfffffc00035c59a0:
>  0 __fp_class_f(0x0, 0x0, 0x0, 0x0, 0x3ff80570244) [0x3ff8051b0f8]
   1 exc_raise(0x0, 0x0, 0x1, 0x80000000, 0x3ff80581e54) [0x3ff8057003c]

Thread 0xfffffc00035c5d60:
>  0 msg_receive_trap(0x140000c90, 0x1, 0x8, 0x100, 0x3ff80528f94) [0x3ff8052e57
0]
   1 msg_receive(0x0, 0x0, 0x0, 0x0, 0x0) [0x3ff80524dd0]
   2 cma__vp_sleep(0x2801000000, 0x0, 0x6, 0x0, 0xaceface) [0x3ff805885ec]
   3 cma__dispatch(0x6, 0x0, 0xaceface, 0x0, 0x3ff80567e34) [0x3ff8056e958]
   4 cma__int_wait(0x100000000000004, 0x4ac18, 0x3ffc01de9f8, 0x47c08, 0x12001a1
e0) [0x3ff80567e30]
   5 pthread_cond_wait(0x12001a270, 0x47f18, 0x12001a228, 0x90, 0x4ab18) [0x3ff8
0576a30]

Thread 0xfffffc00035c5ea0:
>  0 __bcopy(0x14000b9d0, 0x2, 0x12001b14c, 0x0, 0x12001b11c) [0x3ff8051a5c0]

Thread 0xfffffc00035c45a0:
>  0 memmove(0x12001c930, 0x47408, 0xa0000, 0x300, 0x3ff80576ba8) [0x3ff8051c010
]
(dbx) quit



He is using a LOCAL: namespace.
I had a look at the last patches on netrix, but I think nothing
relates to this problem.

Help appreciated

Pascal d'ORNANO

Hi Pascal,

	What is your customer trying to do when the DNS Clerk crashes?
Is this behavior repeatable?  There is a new patch which we are putting
in this afternoon which replaces one of the DNS libraries which might help.
It fixes some thread problems with the clerk.  It should be available to you
tomorrow.

		                                             Regards,

                                                             Ted Pettis
[Posted by WWW Notes gateway]

Hello Ted, 

Thank you for your answer.


What the customer is doing :

From this UNIX machine he starts communication procedures on 9 other UNIX
machines and 1 VMS machine, with tell commands.

Then he starts his DECnet application (socket layer), which talks to the
procedures ran by the tell commands.

It seems to my customer that problem arises if his socket application gets
nut or dies. DECnet is still up and running in this case, and they rerun the
whole process (initialisations with tell commands and start of the appli),
without having to stop/start DECnet.

Often, about 50 cores will appear very very quickly. It is not constant
and not everyday. 


note: they always invoke remote machines by their names (not adresses).


Regards
Pascal d'Ornano

Hello pascal,

I know that on a Multi-CPU machine the DECdns clerk will crash very often.


Hello Ted,

Which problem do you fix in the DECdns library ?
What is the library concerned : libdns_r.so libcache_r.so, or libmisc_r.so ?


Regards,

	Alain.