PQ67826; 4.0.4: so_keepalive option added to the
ORB
Downloadable files
Abstract
Network is unreachable because there is no so_keepalive
option for the object request broker (ORB).
Download Description
PQ67826 resolves the following problem:
USERS AFFECTED:
All WebSphere® Application Server users of Workload Load Managed (WLM)
client for quicker failover.
PROBLEM DESCRIPTION:
WebSphere Application Server client (WLM or non-WLM enabled) is either
slow or unable to detect a broken network when a peer connection is
unreachable. When using two AIX® machines, one running the client, the
other running the server, if the user unplugs the networking cable on the
client box, the client application could hang for a long time. In this
typical failure scenario, the user has a WLM enabled client and multiple
backend EJB™ servers. When one of the EJB servers is unreachable, the WLM
enabled client will hang for a long time before switching to a working
clone. To fix this problem, apply this fix to allow the socket object to
have so_keepalive enabled. In addition to this fix, also set
"tcp_keepidle" to "no" in the JVM™ and TCP/IP layer to allow the socket
timeout to take effect. Also, change the TCP/IP timeout to a smaller
value.
PROBLEM CONCLUSION:
When a remote process is unreachable, WebSphere Application Server could
hang for a long period of time and not be able to switch to a new WLM
server. This fix will add setKeepAlive method for sockets used for
RMI-IIOP connection and allows the socket to throw an IO Exception back to
the ORB component, and the ORB, in turn, throws CORBA.Comm_Failure back to
its caller (WLM).
Prerequisites
None
Installation instructions
Please review the readme.txt for detailed installation
instructions.