Comment 3 for bug 881983

I'm pretty happy with this code - we've run it in production with large cluster (4,200 apache processes, 18 separate cache servers) and tested server death. Works nicely.