Re: Inline content modification?

----- Original Message -----
From: "Henrik Nordstrom" <hno@hem.passagen.se>
To: "Robert Collins" <robert.collins@itdomain.com.au>
Cc: "Joe Cooper" <joe@swelltech.com>; "Squid Dev" <squid-dev@squid-cache.org>
Sent: Thursday, January 18, 2001 9:06 AM
Subject: Re: Inline content modification?

> Robert Collins wrote:
>
> > Yes I agree - any changes made coming into squid _have_ to be
> > changes that are wanted for all clients, for all operations. And
> > support has to be thought out for keeping the original metadata
> > so that IMS requests upstream don't miss if things like
> > content-length change.
>
> You also have to consider Range requests, ETag, and other funny twists.
> If you have a unclean object path in a mesh of systems where objects
> sometimes may get rewritten, sometimes not, or sometimes more than once,
> then you have a very messy situation which will be quite unhelthy for
> the contents, and a nighmare to find the cause.
>
> /Henrik
>

From a 'when does content change' point of view , I see three main thigns to consider:

?what effect does the change have on the cacheability of the object
?what effect does the change have on the objects integrity (content size compared to origin,Etag (might be checksum), ...)
?what knock on effects will the change have.

Making the change on the client request or the server response has very little difference if you make the change on all requests.
You may well cause serious headaches to proxy peers if they ever request directly from origin.

Making changes sometimes is nearly worse: It will be hard to diagnose.

For many of these things the best place to make the change and minise disruption is at the edge of the transmission path: WWW
accelerators and or the client's 'local' cache.

The assumption is that www accelerators are under the control of the www server admin, and can be all configured consistently. For
the local cache, again the assumption is that all the caches will be under a single administrative domain, and if they are peering,
won't send altered data to peers outside that domain. But you don't want the mid level cache servers doing the change.. just the
last hop servers.

Rob
Received on Wed Jan 17 2001 - 15:30:22 MST