Re: 2.6 targets?

From: Henrik Nordstrom <hno@dont-contact.us>
Date: Sun, 25 Aug 2002 23:18:26 +0200

On Sunday 25 August 2002 14.46, Robert Collins wrote:

> I propose that we make 2.6 a very small change over 2.5, given that
> it already has the reworked comms code from Adrian.

Agreed.

> I'd like to see the following in 2.6:
> * Some authentication tidyups I intend to do Real Soon Now. (to
> support delay pools w/ all auth types).

Don't forget all the other access lists.. I think the auth identity
should be cached in the request where appropriate.. I have no problem
with proxy_auth being enforced in http_access if to be used in other
access lists, but I have a problem with proxy_auth not working on
other access lists even if already successfully processed in
http_access for the same request.

> * rproxy (if appropriate - Henrik?)

I'll try to get parts of it over at least. Cannot promise all of it
gets there. There is some important plumbing and fixes in the rproxy
branch that really should get into HEAD/2.6 even if rproxy itself
does not..

> * the client side streams I've done for ESI.

OK.

> * Optionally, ESI itself.

OK.

> * Optionally, any other sponsored work that needs to get into
> mainline sooner rather than later.

I have ETag caching.. think we need a little discussion on this topic.
The implementation in the etag branch (etag-2_5.. the head etag
branch is not yet 100% updated with the store client changes) is
working quite reasonable, but it might need a bit of work to apply to
your ESI client side reorganisation.

Some discussion on the current etag branch implementation can be found
at http://devel.squid-cache.org/etag/. Basically it operates by
saving a secondary index of "variants" on the StoreObject of the URL
when detecting Vary headers, and storing the actual objects under
separate keys (ETag driven, or Vary header values if no ETag is
present). Used successfully in production in an environment with lots
of document variants by the project sponsor.

> * If range caching is a showstopper, reinstated range caching.

Some kind of Range processing should be there in a STABLE release. At
least Range processing of cache hits.

> Are there any other near-to-completion projects on
> devel.squid-cache.org that folk would like to see in 2.6?

I have one not yet published (awaiting sponsor).

Regards
Henrik
Received on Sun Aug 25 2002 - 15:21:27 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:16:08 MST