On Wednesday 19 February 2003 18.37, Andre Kajita wrote:
> I used to have the 'no-query' option for each sibling on every
> machine but I noticed that, despite having heavy usage to the same
> basic sites they weren't sharing cached information so I configured
> all the ICP ports and options and removed the 'no-query' from
> everyone and restarted.
A no-query sibling is effectively a dead sibling as it will never be
queried to see if it has an object.. unless you are also using cache
digests.
> Things are working fine and dandy with no noticable differences,
> just as they were before but now I'm picking up hundreds of
> "TIMEOUT_DIRECT" messages on each of the proxies. I added the
> 'no-query' option again and the messages stopped but so did the ICP
> traffic between the machines (that I logged for a while and were,
> as I could tell, working).
It is possible that the ICP timeout calculation gets the roundtrip
calculation wrong.. try manyally configuring the ICP timeout to a
reasonable value for your network in squid.conf.
> First of all, what the heck does "TIMEOUT_DIRECT" mean?
That Squid did not see any ICP replies in a timely manner (the
TIMEOUT_ part) and went direcly to the origin (the DIRECT part).
> Second, what does removing 'no-query' from the peer settings have
> to do with TIMEOUT_DIRECTs?
By specifying no-query you disable ICP. If ICP is not used there is no
possibility for a timeout as there is nothing to wait for..
Regards
Henrik
Received on Wed Feb 19 2003 - 14:21:13 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:13:29 MST