MQSeries.net :: View topic

mhubbard · Posted: Tue Oct 25, 2005 5:11 pm Post subject:

Hello -

I am running WSMQ 5.3 with service pack 10.
I have a client application that is written in C++ using the C++ API.
This applicaton resides on about 1500 client machines and connects to one of several Queue Managers that I have configured in a cluster.
These clients are connecting via SSL across the internet.

For the most part things are working very well. But on occasion, I find that a client seems to be hung after issuing a Connect to the Queue Manager. Using GDB, I have come up with this consistant stack trace on any box that has the problem:

#0 0x405253c4 in read () from /lib/libc.so.6
#1 0x408f5f00 in SSLAcceleratorCryptoTable () from /opt/mqm/lib/amqcctca_r
#2 0x408eb93d in cciTcpSslReadCallback () from /opt/mqm/lib/amqcctca_r
#3 0x40940c2b in gsk_attribute_get_data () from /usr/local/ibm/gsk6/lib/libgsk6ssl.so
#4 0x40958874 in gsk_attribute_get_data () from /usr/local/ibm/gsk6/lib/libgsk6ssl.so
#5 0x4094e50a in gsk_attribute_get_data () from /usr/local/ibm/gsk6/lib/libgsk6ssl.so
#6 0x4094e3af in gsk_attribute_get_data () from /usr/local/ibm/gsk6/lib/libgsk6ssl.so
#7 0x40919bfc in gsk_secure_soc_init () from /usr/local/ibm/gsk6/lib/libgsk6ssl.so
#8 0x408f9ec7 in ccigsk_secure_soc_init () from /opt/mqm/lib/amqccgsk_r
#9 0x408ebf67 in cciTcpSslInitConn () from /opt/mqm/lib/amqcctca_r
#10 0x408ec3b3 in cciSslSecureAllocConv () from /opt/mqm/lib/amqcctca_r
#11 0x4083ffde in ccxSecureAllocConv () from /opt/mqm/lib/libmqiz_r.so
#12 0x408619ea in rriInitSess () from /opt/mqm/lib/libmqiz_r.so
#13 0x40867bab in DoConnect () from /opt/mqm/lib/libmqiz_r.so
#14 0x40866edd in zstMQCONNX () from /opt/mqm/lib/libmqiz_r.so
#15 0x4082a9fd in MQCONNX () from /opt/mqm/lib/libmqic_r.so
#16 0x405a8855 in ImqMgr::connect() () from /usr/lib/libimqc23gl_r.so
#17 0x08062a7c in MQQueueManager::Connect() ()
#18 0x08055c7d in GetQueueManagerConnection(int&) ()
#19 0x08056b89 in MainLoop() ()
#20 0x080593e5 in main ()
#21 0x4047b14f in __libc_start_main () from /lib/libc.so.6

It seems to me that this tends to happen to some boxes at a time just after something goes wrong on a server - or perhaps I do something intentionally to a server like bring it down for some maintenance. When that happens, the client apps detect that the server is gone and get a 2059 error on the MQGet call that they were probably issuing. The app then tries to reconnect - maybe to a different server - maybe to the same one, and it will consistantly get the 2059 until the server comes back up. But in these problem cases, it appears that it gets into some state where it believes there server to be up, but is not getting a response from it.

Has anyone seen this type of behaviour before - or does anyone know of some connection option that I could set that might limit the amount of time that I sit on this read? - it will stay on that read forever without intervention.

Thanks very much
Mike Hubbard
_________________
Michael J. Hubbard