Re: overlapping stateful helpers

On Sun, 2003-04-20 at 09:01, Henrik Nordstrom wrote:
> On Sunday 20 April 2003 00.13, Robert Collins wrote:
>
> > > I'd also like to see the "overlapping" helper request support in
> >
> > Ditto. BTW I think the stateful overlapping concept may be uhm,
> > difficult. (We can start a separate thread to discuss that).
>
> I did not see any major difficulties in the concept when doing the
> patch, just a bit more code changes than I had time for. With the
> overlapping concept the item to bind to the state is a helper request
> channel/slot, not a full helper. There is no need to change any
> higher level helper protocols because of this, the patch just adds a
> layer below the current protocols, indicating which channel/slot the
> request is sent on. This changes a bit of code as the concept of this
> indirection needs to be added so the stateful "clients" have
> something to bind to..
>
> Note: The current "singlethreaded" helpers only have a single unnamed
> request channel, but it still is a channel/slot in the internals of
> Squid with the patch.

Agreed. The area I see potential difficulties with is that at the
moment, unbound helper requests are guaranteed to be sent to an 'ok'
helper, whereas once you start overlapping, this is not guaranteed.
The point being that stateful helpers change whether they will accept
new requests on the response of each request - which is logically before
the pipelined request can be sent.

Rob

-- 
GPG key available at: <http://users.bigpond.net.au/robertc/keys.txt>.