On 6/08/2013 6:39 p.m., x-man wrote:
Hi Amos,
thanks for the valuable input from your side.
I did further testing and I finally got to the problem.
Really with the new version of squid the problem is happening, but it's only
happening when I use workers > 1, if I keep single worker, then everything
is ok, once I have multiple squid workers, somehow the Responses from the
Cache peer are not reaching back probably 1 of the workers, OR all responses
are going back to only 1 worker, that's why there is this
TIMEOUT_FIRSTUP_PARENT happening.
Now the questions is why it's happening like this, because my firewall
config is kept unchanged. I'm using tproxy for both squid and my cache peer.
The squid and my cache peer are running on same machine.
Can you provide some insights based on the above?
Oh one educated guess...
When SMP workers are enabled they all share the port sockets and first
worker to read the reply back will be the one handling it. Even if that
worker had not sent it. So its possible the worker sending it will never
get a reply and timeout.
I have a port upgrade patch still working its way done QA that should
fix that as a side effect of its main change.
As a workaround for now you can try allocating each worker a unique
icp_port.
Amos
