Hi Patrick, On Tue, May 23, 2017 at 01:49:42PM -0400, Patrick Hemmer wrote: (...) > haproxy 28856 root 1u IPv4 420797940 0t0 TCP > 10.0.33.145:35754->10.0.33.147:1029 (CLOSE_WAIT) > haproxy 28856 root 2u IPv4 420266351 0t0 TCP > 10.0.33.145:52898->10.0.33.147:1029 (CLOSE_WAIT) > haproxy 28856 root 3r REG 0,3 0 4026531956 net > haproxy 28856 root 4u IPv4 422150834 0t0 TCP > 10.0.33.145:38874->10.0.33.147:1029 (CLOSE_WAIT)
These ones are very interesting. > All those sockets have been sitting there like that for a long time. > The :1029 sockets are "peer" sync connections. That's what I was about to suspect. Fred has started investigating a similar report, but these ones are rare so it's possible that combined with your report it can help spot the root cause. I'm pretty sure we have a bug somewhere in the state machine where we forget to re-enable polling for reads in the peers, but we haven't found it yet :-/ At least your analysis tends to confirm this direction. I'll let Fred contact you if he estimates he needs more info. thanks! Willy