Anand Avati <anand.avati@xxxxxxxxx> wrote: > Unfortunately event.c's poll() implementation is very little tested Is it an event-poll story? When the backtrace hits even-pool.c, glusterfs has already decided to destroy the transport because its refcount droped to zero. It looks more like a refcount tracking issue, which is always done outside of event-poll code. -- Emmanuel Dreyfus http://hcpnet.free.fr/pubz manu@xxxxxxxxxx
