We have had a lot of faliures on Hudson today. (and I can replicate those on my locan environment also).
It seems those failures happened after the recent changes on ConnectionManagerImpl.
It's being kind of hard to pinpoint the exact problem as it seems to "move" around. But it is a fact that I can get a successful run (with just 1 or 2 failures as it was usual) at r7679.
If you run the testsuite, it will usually fail at WildcardRoutingTest. I could pass that point if reverting the changes on ConnectionManagerImpl. (the close on oldConnections thing).
I will need a little bit more time to understand why... but I wanted to give a heads up on the progress I've made.
Also.. it would help me having some explanation about the change on ConnectionManagerImpl.
Also.. there is another failure on startStop. I at least could improve some of the cascades by ignoring one exception on close.. I will still investigate that other failure.