trunk patch: http://svn.apache.org/viewvc?view=revision&revision=1542379
2.4.x patch: trunk patch works mod CHANGES
+1: jim, covener
+ trawick: [I guess I'm just confused but] I thought that the scenario as
+ described by the reporter was that this code was running while
+ a bunch (nearly all) of httpd children were zombies, leading to
+ a connect() timeout since no processes were available to empty
+ the TCP queue. IIUC, the scoreboard state will not yet have
+ been changed to SERVER_DEAD and the pid will not have been
+ cleared for those processes, since parent code that can't run
+ while we're stuck in this function takes care of that, so we'll
+ still block.
* event fdqueue: Use a normalized offset point for idlers via atomics
trunk patch: http://svn.apache.org/viewvc?view=revision&revision=1545286