On 9/25/2010 2:23 AM, Sim IJskes - QCG wrote:
On 09/25/2010 11:05 AM, Patricia Shanahan wrote:
I needed to do a general reset before starting a new set of tests. I
completely shut down (virtual power off) all my VMs and restarted the
system. This happened during the first build I did after restarting the
VM in question. Before the build, I had run gedit but no tests. If it is
a testing framework problem, it has to be in the junit testing that is
done during the build.
The operating system is not without state, so a test might be influenced
by the state changes in the OS from the previous test.
For instance when a server socket is opened, and closed, you cannot
immediately bind on that same server socket, unless you specify
SO_REUSEADDR with setsockopt(). This is because after closing the socket
stays in TIME_WAIT2 state for some time.
Across a reboot? I did a complete shut down of the VM, and restart of
the hosting WindowsXP, between the last test and this build failure.
I have had problems in the past in which killing a test left a socket
tied up, but rebooting has always cleared it.
So if you forget to specify SO_REUSEADDR on your server sockets, and run
the same test in rapid succession, it might fail.
I did a small codereview on the TcpServerEndpoint, and no mention of the
ServerSocket.setReuseAddress() call, so that is a potential cause.
Gr. Sim
