<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html;charset=ISO-8859-1" http-equiv="Content-Type">
</head>
<body bgcolor="#ffffff" text="#000000">
Ray, Timothy J. (GSFC-583.0) wrote:<br>
<span style="white-space: pre;"><br>
> Dear WG members,<br>
> <br>
> The purpose of this email is bring up a possible minor tweak. The<br>
> tweak is not needed, and I don’t have a strong attachment to it –<br>
> just want to mention it...<br>
> <br>
> If a configuration server dies and quickly restarts, the registrars<br>
> will not miss enough heartbeats to impute its death, and will<br>
> continue sending heartbeats. If the Meta-AMS transport mechanism
is<br>
> udp and the restarted server uses the same udp port, then these<br>
> heartbeats will reach the restarted server.<br>
> <br>
> Would it be better if the server had a grace period at startup
during<br>
> which it would ignore heartbeats from unregistered registrars
(rather<br>
> than sending them a ‘you-are-dead’ message)? The grace period
would<br>
> last long enough to allow the registrars to impute the death of the<br>
> old registrar and re-register with the restarted registrar.</span><br>
<br>
I see what you're saying, Tim, and I agree that it would be more
graceful. Would the same behavior be appropriate for a registrar that
restarts quickly, before its cell's nodes have imputed its death?<br>
<br>
On the one hand this is kind of a corner case and I am reluctant to add
any more complexity to MAMS unnecessarily. On the other hand, I can
imagine operational scenarios where this situation might come up and
result in some confusion. I'll take a crack at modifying the spec if
it's the consensus of the WG that this is a worthwhile enhancement.
Any thoughts?<br>
<br>
Scott<br>
</body>
</html>