I think it is still is useful to provide a mechanism to gracefully shutdown
an update stream. If the client just closes the stream (or crashes), the
server keep on sending updates for quite a while before the TCP stack times
out. This obviously eats up resources.

But I think it should be considerably simpler than the current draft. So how
about this approach:

* A server offers one or more update-stream resources. Each such resource
provides updates for a set of resources selected by the server. The client
request specifies a subset of those resources, and the server responds by
pushing updates to the client. This is the only type of request an
update-stream resource accepts, so we do not need to call it start-updates
or get-updates.

* We define another resource type, called close-stream, with a different
media-type. The client request is a list of stream-ids for active
update-streams. The server responds by closing its end of those streams. A
close-stream can close a stream obtained from any update-stream resource.
Hence a close-stream resource is not tied to a specific update-stream
resource, and a server does not need to offer more than one close-stream
resource.

I think that is simple & unambiguous. What does everyone think of that
modification?

- Wendy

From:  EXT Gao Kai <[email protected]>
Date:  Wed, March 9, 2016 at 01:05
To:  Wendy Roome <[email protected]>, <[email protected]>
Subject:  Re: [alto] State of the WG

On 08/03/16 22:14, Wendy Roome wrote:
> A few points:
> 
> * The original proposal did not provide an explicit stop-updates request.
> The server just pushed updates until the client closed the stream. Then
> after implementing that, I discovered that it took a long time for the
> server to discover that the client had closed the stream. That is why I
> added stop-updates. At first that was just stop-everything; then it seemed
> like a simple extension to allow the client to stop a subset of the
> resources. If adding a control interface causes controversy, we could
> drop stop-updates altogether.
For simplicity, I agree on dropping the stop-updates.  And the stream-id is
also no longer needed with the removal of control interfaces.
> 


_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto

Reply via email to