https://bugzilla.redhat.com/show_bug.cgi?id=1104843
--- Comment #6 from John Eckersberg <jeckersb(a)redhat.com> ---
I think Lennart described in the fedora-devel thread why we can't really shut
down epmd after it has been started via socket activation.
The problem here is that epmd, however it gets launched, is going to be shared
among all of the erlang VMs running on a server. So you have one of two cases:
(1) is what we have now, where each service as part of booting its VM will
check for epmd, and start it if it is not present. That's a problem though.
Consider you have two erlang services, A and B. You start A, it starts epmd as
part of its startup. Now the epmd process is running under the cgroup of
service A. Then you start service B, it detects epmd is already running and
starts using it. Then service A is stopped and all of the processes under the
cgroup, including epmd. Now process B doesn't have an epmd to talk to, and you
will have to restart it in order for a new one to get spawned.
(2) is this proposed case, where epmd is its own standalone service, running in
its own cgroup, under socket activation. This avoids the mess from (1), with
the only downside that I see is that it does not automatically stop (it can't).
IMO, the downside to (1) is much worse than the downside to (2). If the
user/administrator does not want epmd to be running after rabbitmq-server is
stopped, I don't think it's unreasonable to require it to be explicitly
stopped.
--
You are receiving this mail because:
You are on the CC list for the bug.