Tek Bahadur Limbu <teklimbu@xxxxxxxxxxxx> writes:
>> Timeout related settings:
>>
>> icp_query_timeout 50
>> maximum_icp_query_timeout 50
>
> You can try increasing the icp_query_timeout/maximum_icp_query_timeout
> to 5000 and check if it works.
Thanks a lot! That hasn't helped too much however, as what
_icp_query_timeout_ and _maximum_icp_query_timeout_ affecting are UDP
related queries, and `TCP_MISS/504' is unfortunately a TCP related one.
Anyway, I think we have finally caught it. It is a timeout, but *not* in
the normal sense. When machine A got an `UDP_HIT' response from machine
C, it sent an HTTP request to C to fetch the requested object. That
request (which is sometimes called the `only-if-cached request')
contained a `Cache-Control:' header:
Cache-Control: max-age=259200, only-if-cached\r\n
If the local cached object is older that `max-age', Squid in machine C
would response with a 504 immediately.
I think switching to HTCP would get it fixed.
