Manik Surtani wrote:
Looking at using the thread counts and parallelizing test execution
in
TestNG (as Bela has done with JGroups), one of my biggest stumbling
blocks is that quite a few tests end up starting and stopping
replicated JBoss Cache instances. If these tests were to be run in
parallel, they would detect each other and produce unexpected results.
Question for any of the JGroups guys on this list: how did you get
around this with creating JGroups channels, as I expect you have the
same problem? Or do you not run tests that create/start channels in
parallel?
I haven't done this yet, but the idea is to change ChannelTestBase
(superclass for all stack-dependent tests) to use a resource pool which
manages multicast addresses and multicast ports (or TCP ports): each
test removes a multicast address:port combo and returns it when done.
Same for TCP ports.
Once we have a valid multicast address/port, the superclass will modify
the transport properties *before* the channel is started.
I'll keep you posted when I'm done implementing this (should be end of
next week).
--
Bela Ban
Lead JGroups / Clustering Team
JBoss - a division of Red Hat