spec/support/client_forker.rb in zk-1.5.0 vs spec/support/client_forker.rb in zk-1.5.1

@@ -10,10 +10,11 @@
 
   def initialize(cnx_args, base_path)
     @cnx_args  = cnx_args
     @base_path = base_path
     @pids_root = "#{@base_path}/pid"
+    @child_latch = Latch.new
   end
 
   LBORDER = ('-' * 35) << '< '
   RBORDER = ' >' << ('-' * 35) 
 
@@ -61,10 +62,17 @@
     proc do |s,dt,pn,msg|
       "#{CLEAR}#{YELLOW}#{orig_formatter.call(s,dt,pn,msg)}#{CLEAR}"
     end
   end
 
+  def start_child_exit_thread(pid)
+    @child_exit_thread ||= Thread.new do
+      _, @stat = Process.wait2(pid)
+      @child_latch.release
+    end
+  end
+
   def run
     before
     mark 'BEGIN TEST'
 
     logger.debug { "Process.pid of parent: #{Process.pid}" }
@@ -100,11 +108,10 @@
 
     mark 'FORK'
 
     @pid = fork do
       Thread.abort_on_exception = true
-      ::Logging.reopen
 
       @zk.wait_until_connected
 
       child_pid_path = "#{@pids_root}/#{$$}"
 
@@ -167,10 +174,12 @@
       else
         exit! 1
       end
     end # forked child
 
+    start_child_exit_thread(@pid)
+
     # replicates deletion watcher inside child
     child_pid_path = "#{@pids_root}/#{@pid}"
 
     delete_latch = Latch.new
 
@@ -186,12 +195,13 @@
       end
     end
 
     delete_latch.await if @zk.exists?(child_pid_path, :watch => true)
 
-    _, @stat = Process.wait2(@pid)
+    @child_latch.await(30) # if we don't get a response in 30 seconds, then we're *definately* hosed
+    
+    raise "Child did not exit after 30 seconds of waiting, something is very wrong" unless @stat 
 
-#     $stderr.puts "#{@pid} exited with status: #{stat.inspect}"
   ensure
     mark "END TEST"
     kill_child!
     tear_down
   end