Re: Magic Markers and MD5

Miguel A.L. Paraz wrote:
>
> Hi,
>
> If we use a Magic Marker sequence of bytes, how do we know that the
> sequence will not occur in real live data? How about adding a
> Content-Length to objects that lack one? Or, compute the MD5 checksum

Main problem (hope I got the question right!):

If cache-parent is receiving data which it doesn't know the Content-Length
for and transferring it to a sibling via http as a "transparent pipe"
we can't simply prepend a "Content-Length" to the beginning without
waiting for the whole connection to complete.

Possible Solutions:
o Some kind of "control" connection between the servers... (mission,
        incompatible with netscape and other proxies, etc etc). Can't really
        use UDP/ICP as it would mean that the close would be unreliable.
o Use out of band data to embed the "complete object recieved" message
        in the actual TCP, so that it is handled by all TCP stacks (and by
        the normal squid error handlers?). I don't
        know how netscape would handle OOB data in a TCP connection if
        netscape is set up with a squid parent though, but from what
        (little) I know it should just treat it as normal data (which would
        currupt files ;)
        Since we (I hope) send a "this is a squid cache" header in the
        "GET" strings we can tell if it's a squid cache connecting to us,
        and can just not send the arbitrary data length/MD5 sum.

        Oskar
Received on Tue Jul 29 2003 - 13:15:41 MDT