Michael V. Pelletier wrote:
> I haven't taken a look at forward.c yet, but I should mention that the
> single address retry was an afterthought - the main purpose of the patch
> is to insure that if a multi-address site such as www.microsoft.com
> refuses a connection, Squid will try the other addresses looking for one
> that's accepting connections before returning an error to the user - so
> setting single addr tries to 1 is a perfectly reasonable approach to take.
forward.c retries on a higher level than comm.c. It walks throught a
list of candidate hosts (Squid 2.2). It always retries 10 times, and if
there is not enought hosts to try then the last host is repeated. Hosts
in this sentence is hostnames of peers and origin site.
Only notable difference I can find apart from the bug in commResetFD is
how the retry connect timeout is supposed to be calculated
(minimum_retry_timeout). But on the other hand this documented connect
retry timeout calculation for multihomed/clustered servers seems to be
absent in the current Squid version, so perhaps this is no difference
after all.
/Henrik
Received on Tue Jul 29 2003 - 13:15:57 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:12:04 MST