Doesn't happen 100%, but close.  Here are some symptoms.

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fb3eb7fe700 (LWP 680)]
0x00007fb4074ceefb in __cds_list_del (prev=0x21e45b000, next=0xdeadc0de00)
    at /usr/include/urcu/list.h:73
73		next->prev = prev;

(gdb) bt
#0  0x00007fb4074ceefb in __cds_list_del (prev=0x21e45b000, next=0xdeadc0de00)
    at /usr/include/urcu/list.h:73
#1  0x00007fb4074cef33 in cds_list_del (elem=0x7fb3d80120d0)
    at /usr/include/urcu/list.h:81
#2  0x00007fb4074cef4e in cds_list_del_init (elem=0x7fb3d80120d0)
    at /usr/include/urcu/list.h:88
#3  0x00007fb4074d1b4d in glusterd_friend_sm () at glusterd-sm.c:1348
#4  0x00007fb4074c732b in __glusterd_handle_incoming_unfriend_req (
    req=0x7fb3d800f4ac) at glusterd-handler.c:2670

(gdb) p event
$1 = (glusterd_friend_sm_event_t *) 0x7fb3d80120d0
(gdb) p event->list
$2 = {next = 0xdeadc0de00, prev = 0x21e45b000}
(gdb) p tmp
$3 = (glusterd_friend_sm_event_t *) 0xdeadc0de00
(gdb) p *event
$4 = {list = {next = 0xdeadc0de00, prev = 0x21e45b000}, 
  peerid = "\000\241\000\000\000\264\177\000\000\070\000\000\000\000\000", 
  peername = 0x7fb3d80120d000 <error: Cannot access memory at address 0x7fb3d80120d000>, ctx = 0xffffffffffffff00, event = 4294967295}

It looks like list/memory corruption, probably due to improper usage of RCU functions.  I have a patch that makes the problem go away, which I'll post as soon as I get this bug number.