On 29/10/11 00:23, Justin Lawler wrote:
Hi,
We're using ICAP on Solaris for squid 3.1.16 and we're seeing some strange behavior.
When cache is disabled and/or using 'ufs' caching policy, all works ok, but if we change to 'aufs' and enable caching, we're seeing REQMOD's coming through, but no 'RESPMOD's.
Note that REQMOD is asking about the client request. Before it gets
satisfied by Squid. It gets "called" for all requests.
Whereas, RESPMOD is asking about some reply, which has *already* been
satisfied by some server. It _never_ gets "called" on TCP_HIT.
http://wiki.squid-cache.org/Features/ICAP#Squid_Details
Most of what you describe is expected behaviour (cache enabled == less
RESPMOD).
Having same/more RESPMOD with UFS despite caching enabled sounds like
a bug.
We see the same issue on 3.0.15.
Is this a known issue - or anybody see this before? Could it be a configuration option? Should we consider sticking with ufs, or changing to 'diskd'?
It has not been mentioned in this way before. Though maybe was described
in other terms.
Note that the I/O system functions used to read/write data onto HDD in
no way affects the decision about whether to pass traffic over the
network to ICAP.
However, if one method of I/O were corrupting the objects during load it
could cause Squid to abandon the cached object and fetch a new one. This
would appear as similar behaviour to what you describe for UFS with
caching enabled.
Amos
--
Please be using
Current Stable Squid 2.7.STABLE9 or 3.1.16
Beta testers wanted for 3.2.0.13
