Re: Ranges: ??s and bug(?).

> There is a specific test for the offset limit - you should start there
> in debugging this. IIRC a grep for weDoRanges should find this. (If we
> do ranges, then we get the whole object and give the ranges to the
> client).

Ok - I'll look into this then. I'm definitely not seeing squid 3
prefetch when range_offset_limit is -1 KB, at least in some cases.

> I'd need to check to answer this for you. It's likely that the
> interaction is:
> offset limit result
> * - + suffix = grab everything
> * 0 + suffix = just what the client requests
> * + + suffix = just what the client requests.

I actually looked into this for squid 2.5 and it turned out that that
squid 2.5 always graps just the requested amount when requesting a suffix.
It never grabs everything.

> The squid range logic for 3.0 is:
>
> when the result from the server arrives, extract into canonical
> representation: that is, strip the multipart encoding.
>
> when sending to the client, encode anew.
>
> So: client_side should only ever see the identity data, in offset,length
> blocks. If you see *any* of the multipart encoding in client side, then
> there is a bug in the http.c side logic which parses and removes the
> multipart encoding.

That's actually reassuring. :-) I'll start looking there for the source of
this problem.

Thanks a bunch.
Leeann
Received on Sun May 11 2003 - 17:01:43 MDT