Re: Cache-DIGEST false hits

From: Alex Rousskov <[email protected]>
Date: Thu, 20 Aug 1998 10:16:55 -0600 (MDT)

On Wed, 19 Aug 1998, Serge Krashakov wrote:

> I have seen not only calamaris-2 report (which supports CACHE_DIGEST tag,
> may be, after my patch) but also did explicit grep from logs of both neighbors.
> And in some periods of time I have observed even 90% of false hits.
> If you need in more information (for example, some additional debug level logs)
> I can extract and send it.

Serge,

        Here is what you can try. Grep the logs for a recent false hit due to
Cache Digest selection algorithm and then (assuming sibling A got a false hit
from sibling B):

        1) purge the object from A using client program
        2) request the same object from A using client program
        3) check that the object indeed came from B (X-Cache)
        4) check that the object is indeed a false hit (X-Cache says MISS)
        5) look at X-Cache-Lookup headers in the reply
           X-Cache-Lookup keeps the result of a store table lookup
           on B _before_ refresh-causing rules are checked:

           - If X-Cache-Lookup says MISS, then A's digest is out of date
             or corrupted OR B purged that object already;
           - If X-Cache-Lookup says HIT, then B had the object, but B's
             refresh-like rules (or A's max age requirements) prevent
             A from getting a hit.

Repeat the procedure above for a few false hits to see if there is a
consistent pattern. Next action will depend on the results in (5).

Good luck,

Alex.
Received on Thu Aug 20 1998 - 09:19:08 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:41:40 MST