Re: IMS replies

Not even metadata can be updated on-disk today. Only what is saved in swap.state can be updated. Once a cache object has been
written to disk it stays as it is until it gets removed.

But yes, having the headers as meta-data would most likely make it easier to manage updates.

/Henrik

Robert Collins wrote:

> I think implementing this would fall out as a freebie if we turn the headers into metadata rather than part of the datastream.
>
> Rob
>
> ----- Original Message -----
> From: "Henrik Nordstrom" <hno@safecore.se>
> To: <squid-dev@squid-cache.org>
> Sent: Friday, January 19, 2001 6:32 AM
> Subject: IMS replies
>
> > I think there is many bugs/misfeatures in how we handle IMS replies to
> > refreshes.
> >
> > a) The new headers are never reflected back to the client. This is
> > especially a problem due to the Date header. Not updating the Date
> > header will upset age calculation in downstream caches badly, but there
> > might be other headers which should be updated as well.
> >
> > b) New headers are stored in the memory object, but it seems there is no
> > pruning of old headers, or at least the array where the headers are
> > stored keeps on growing for each received reply, and the array item
> > counts gets very large.
> >
> > c) The new headers are never stored on disk.
> >
> > Ideas are welcome on how to handle the object headers on refreshes, but
> > until we have found a way I think the in-memory header update should be
> > disabled to at least not waste memory on not doing it correctly......
> >
> > /Henrik
> >
> >
Received on Thu Jan 18 2001 - 13:56:49 MST