On Sun, May 12, 2013 at 6:00 AM, Amos Jeffries <squ...@treenet.co.nz> wrote:
> The attached patch adds a built-time setting SQUID_PIPELINE_MAX to replace
> the magic numbers previously used for limiting the amount of prefetch Squid
> does when pipeline_prefeth config directive is enabled. The default remains
> at 2 unless altered by using -DSQUID_PIPELINE_MAX=n in CXXFLAGS and this
> adds some documentation about why, and what can be done to improve
> pipelining in future.
On the matter of the TODO:
- it's up tro HTTP/1.0 clients not to send pipelined requests. If a
client behaves like HTTP/1.1, and it's harmless, why deny them the
pleasure?
- it'd probably be best to implement the blocking on NTLM/NEGOTIATE.
However, I doubt that a client would send a type 3 message until it
has received the type 2 back from squid. This is no guarantee and is
probably hard to troubleshoot if it happens, but I suspect that it
won't bite us in the wild. It'd have bitten us by now otherwise.
+1
--
/kinkie
