Further to my last message, I have been looking more closely at the cache log
and it would appear the problem did not start immediately after upgrade from
1.0.12 to 1.0.18. It took nearly two days to start happening. We reboot our
server every Tuesday at 4am and HUP it every other weekday at 4am. It took
one reboot and two HUPS of contented running before errors like the following
started showing up in the cache.log file (immediately after the second HUP):
[23/Oct/1996:04:01:35 +1000] main.c:460: Ready to serve requests.
[23/Oct/1996:04:01:37 +1000] store.c:1384: 4096 Lines read so far.
[23/Oct/1996:04:01:39 +1000] store.c:1384: 8192 Lines read so far.
[23/Oct/1996:04:01:40 +1000] store.c:1384: 12288 Lines read so far.
[23/Oct/1996:04:01:42 +1000] store.c:1384: 16384 Lines read so far.
[23/Oct/1996:04:01:44 +1000] store.c:1384: 20480 Lines read so far.
[23/Oct/1996:04:01:46 +1000] store.c:1384: 24576 Lines read so far.
[23/Oct/1996:04:01:46 +1000] store.c:1140: storeSwapInHandle: WARNING! Obj
ect size mismatch.
[23/Oct/1996:04:01:46 +1000] store.c:1141: --> <URL:http://home.netscape
.com/escapes/images/i-site_sampler.gif>
[23/Oct/1996:04:01:46 +1000] store.c:1142: --> Expecting 1673 bytes from
file: /var/cache/harvest1/32/134928
[23/Oct/1996:04:01:46 +1000] store.c:1144: --> Only read 4939 bytes
[23/Oct/1996:04:01:48 +1000] store.c:1384: 28672 Lines read so far.
These messages continued over the next day or so in bursts, sometimes every
minute, sometimes none for an hour or so. They stopped immediately we
downgraded to 1.0.12. It wasn't just gifs either...
Rob.
>We have squid 1.0.12 running on a Sparc20, Solaris2.5, 8Gbyte cache, 250,000
>accesses per day. Last week we upgraded to 1.0.18. For the next two days
>we started getting isolated reports about the wrong gif showing up on a page
>where it ought not be when pages were accessed through the cache. This
>culminated with a pornographic image apearing on the home page for a primary
>school :-(. Needless to say we immediately reverted to 1.0.12 and everything
>has been fine.
>
>Looking through the logs for the cases we could pin down (we only had about
>6 reported cases over the two days out of half a million accesses) there is
>evidence in the cache.access log that the size of the gifs are different to
>what they oughta be:
>
>incorrect size:
>24/Oct/1996:09:04:15 +1000 5615 203.62.0.59 TCP_HIT/200/NONE 4131 GET http://f
>reenet.actein.edu.au/PLMP/pic1/grball.GIF
>24/Oct/1996:09:07:19 +1000 189596 203.62.0.59 TCP_HIT/200/NONE 8192 GET http://f
>reenet.actein.edu.au/PLMP/pic1/grball1.GIF
>
>correct size:
>31/Oct/1996:12:56:00 +1100 206 150.203.4.30 TCP_HIT/200/NONE 1538 GET http://freenet.actein.edu.au/PLMP/pic1/grball.GIF
>31/Oct/1996:12:56:00 +1100 194 150.203.4.30 TCP_HIT/200/NONE 1278 GET http://freenet.actein.edu.au/PLMP/pic1/grball1.GIF
>
>Another interesting thing to note is the large amount of time the incorrect
>gifs took to arrive from the cache.
>
>Anyone else noticed this problem? - I haven't been brave enough to try 1.0.19...
>
>Rob.
>
-- * Robert Davy * .* Robert.Davy@anu.edu.au Network Services ph:06 2492978 fax:06 2798199 Australian National University * Canberra, ACT 0200, AustraliaReceived on Wed Oct 30 1996 - 22:50:22 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:33:24 MST