[erlang-questions] Future of epmd

Dmitry Demeshchuk <>
Wed Nov 7 20:26:18 CET 2012


Hmm. I thought that something like "kill -9" wouldn't inform us that the
socket has been closed until we try to do something with it. But checked –
and yes, you are right, keep-alive is even bad in that case, since without
it we immediately get a {tcp_closed, Socket} message.


On Wed, Nov 7, 2012 at 11:15 PM, Per Hedeland <> wrote:

> Dmitry Demeshchuk <> wrote:
> >
> >1. When we send ALIVE2_REQ and reply with ALIVE2_RESP, we establish a TCP
> >connection. Closing of which is a signal of node disconnection. This
> >approach does have a point, since we can use keep-alive and periodically
> >check that the node is still here on the TCP level.
>
> No, the point is rather the opposite - since this is always a local
> loopback connection, epmd is guaranteed (by the OS/kernel) to
> "immediately" find out that the erlang node died (or disconnected), by
> means of socket close (EOF) - no matter how the death came about. TCP
> keep-alives, that by necessity incur a delay (and the default is
> typically huge) before detection of a problem, are not only inferior but
> pointless in this scenario.
>
> --Per Hedeland
>



-- 
Best regards,
Dmitry Demeshchuk
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20121107/20dced42/attachment.html>


More information about the erlang-questions mailing list