Re: Retry patch ,ERR_C0NNECT_FAIL too!

On 7 Apr 1998, Riccardo Vratogna wrote:

> Ah , you are the retry patch creator:)
> Mine compliments for your patch. Squid without this patch
> it is unusabile. Very good work! Thanks for this.

You're very welcome! It's very gratifying to know it's well recieved! If
you could do me the favor of announcing your connection-retry-patched
Squid to the tracker.ircache.net host by enabling the announce_to line in
your squid.conf, I'd appreciate it... :-) I'd like to get one
retry-patched Squid host in each country. ;-)

> I've see in the logs:
>
> www.ilsole24ore.it(194.243.161.219) marked bad
> ERR_CONNECT_FAIL:http://www.ilsole24ore.it/24oreonline/oggi/Prima_pagina/G_RONCHI.htm
>
> So i've made a reload in the clients: and get in the logs:
>
> www.ilsole24ore.it(194.243.161.219) marked good
>
> And the page appear.
>
> Because any try are made the first time? I mistake anythings?

I think when you reloaded, it just got lucky after trying a known-bad
address once again, and got through to the host. This is why the address
was marked good - it used to be refusing connections, and now it's not.

If you know C, it might be helpful for you to look in the "comm.c" module
of the Squid source. The "marked bad" item occurs after the first try
fails, and the CONNECT_FAIL error occurs after the second attempt fails.
It does not log every single attempt to connect a refusing host, otherwise
the log file would grow very large very quickly.

If you set your debug level to "2" for section "5", as in:

debug_options 2,5 1,ALL

that will show you log entries such as:

        www.foobar.com(194.243.161.219) marked bad
        commConnectHandle: FD 17 www.foobar.com conn fail, retrying
        www.foobar.com(194.243.160.10) marked bad
        commConnectHandle: FD 17 www.foobar.com conn fail, retrying

and so on... (from the "commConnectRetry()" function in comm.c.) However,
this "retrying" message will show up for every single time that the host
is retried. If a single-address host has a downed web server, every
single object requested from that host will cause three of these lines to
appear in the log file by default, and that adds up quickly. If all 14
hosts of www.microsoft.com are busy and refusing connections, you'll see
14 of these lines before the CONNECT_FAIL message appears for every single
object requested.

You should turn on this debugging option temporarily, that will give you a
better sense of the functioning of the retry patch.

If you still don't get the "retrying" messages for some hosts, make sure
that your config file option "maximum_single_addr_tries" is set to
something greater than one. This is the number of times a
connection-refusing host with only one IP address will be retried. (I
find that 4 is a good value, particularly for PointCast clients.)

By default, at debug level 1, the only time the retry patch makes its
presence known in the logfile is as you saw - when the state of an address
changes (either OK->BAD or BAD->OK) or when there's a successful
connection to a retrying connection.

        -Mike Pelletier.
Received on Tue Apr 07 1998 - 10:48:02 MDT