As you say it might be a DNS thing. Try configuring the peers using IP
addresses instead.
The connect-timeout argument to cache_peer works once Squid has found
the IP of the peer, and so do the global option.
And if you have never_direct enabled (allow) then Squid-2.4 should add
all of the peers to the list of peers to try before sending an error to
the client.
-- Henrik Nordstrom Squid Hacker Andrei B. wrote: > > I have a satellite reception dish with an Europe Online account. > > As Henrik suggested in a previous discussion : > > cache_peer proxy.113.europeonline.net parent 8080 7 no-query login[...] > cache_peer proxy.114.europeonline.net parent 8080 7 no-query login[...] > cache_peer proxy.115.europeonline.net parent 8080 7 no-query login[...] > cache_peer proxy.103.europeonline.net parent 8080 7 no-query login[...] > > All are activated. He told me that squid will see that if a peer is > down, it will switch to another one with a delay until a dead peer is > detected. > > Due to the nature of the link, only one of these parent caches can be > active at any one time. I wanted to put them all active, so that I > don't have to edit squid.conf and reload config everytime I bring the > link up (modem is used for upload). > > The following thing (and it's annoying) happens: > - client issues a request. > - squid attempts to use the first parent in the above list (which is > down, by the way) > - after a very long time, similar to that of a failed DNS lookup or > failed to connect on a direct path, I get the message "peer cache is > down" > - client issues a reload on the URL > - squid tries the second peer, and if down, the same thing happens. > > Suppose 103 is up, the rest are down. It takes the client a request and > 3 subsequent reloads on an error page until squid gets to use the last > peer in the list and navigation starts to work. Pretty nasty. > > Any suggestions on how to change this behaviour? (If possible) > > I set connect-timeout=5 (seconds) on the cache_peer line, with no > results. > > dead_peer_timeout is not usefull, it's a TCP connection. > > connect_timeout seems to influence this, but also gives timeout errors > on slow servers. > > peer_connect_timeout, although by description it should address this, > seems to have no effect. I set it to 5 seconds, and my connect still > waits about 2 minutes to timeout, then for the next request, peer is > changed. > > ANy help is welcome. Thanks. > > ===== > ing. Andrei Boros > Centrul pt. Tehnologia Informatiei > Societatea Romana de Radiodifuziune > > __________________________________________________ > Do You Yahoo!? > Yahoo! Auctions - buy the things you want at great prices > http://auctions.yahoo.com/Received on Wed May 16 2001 - 16:09:29 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:00:09 MST