On Sun, 25 Oct 2009 22:43:53 -0500, Matthew Young <myoung24866@xxxxxxxxx> wrote: > Hello all > > I have a group of proxy users who are not technical at all, and it is > very common for them to complain that the "network is slow" because > there job consists of browsing sites all day and sometimes they are > just lucky that they hit remote servers which are non response, or > initiate the connection but never feed data thus they see their > firefox as loading and loading and they next thing we know they also > think there "computer is slow". They cannot tell the difference > between a local network issue and a remote server issue. > > Id like to drop down the possibilities of this and set a timeout of 30 > seconds, my goal is if the remote site is non responsive (waiting for > data) id like to timeout the connection in 30 seconds tops, if > possible display the timeout message. > > My question is, what are they related time outs within the config that > are safe to modify? I modified the read_timeout (default 15min, yes > some users actually stare 15 min on the screen) to 1 minute but for > some reason it didnt take place. > > Also, does the cache manager offfer a way to list the slowest queries > for inspection? > > Thank you. > > -- Matt When they are finished the native squid format lists the duration of a request in milliseconds (column 2). I'm not aware of any way to find the longest existing current connections. May not be relevant here but SqStat is a cool little utility very good at displaying the current ones in realtime on demand. The timeouts I can think of being relevant to overall transaction delay on failures are request_timeout, read_timeout, connect_timeout, forward_timeout and dns_timeout. Of those request_timeout is the most critical since it limits the total time the sending and receiving of the request may take (read_timeout default allows 15 min for each byte to come in, but request_timeout will still terminate the request at ~4minutes if its not finished by then). It's a bit loose on GET requests (should be very short. 10 seconds etc) but very long for POST (minutes or hours for huge file POSTS). It starts _after_ the connection is successfully opened. Second most critical is a mixture of the other three. AFAICT: read_timeout should be larger than: dns_timeout + forward_timeout forward_timeout should be larger than connect_timeout times the number of IPs each website has. This is based on my observations of a few bugs I suspect of existing because read_timeout or forward-timeout was shorter than the other two settings. Amos
