I have some other problems probably with this patch.
Currently trying to use a parent cache does not work for me.
I am testing the configuration with older squid trees and I have not any
problem.
Did the configuration parameters changed? Are there changes I should
take care about?
On 06/21/2011 07:29 PM, Alex Rousskov wrote:
> On 06/21/2011 06:09 AM, Amos Jeffries wrote:
>> On 21/06/11 21:46, Tsantilas Christos wrote:
>>> On 06/19/2011 05:19 AM, Amos Jeffries wrote:
>>>> Comm no longer contains implicit DNS resolution. It needs to be given
>>>> the remote IP address before opening.
>>>
>>>> Since ServiceRep::getConnection() is strictly synchronous and assumes
>>>> that there is only one IP for the service.
>>>>
>>>> Xaction::openConnection() can become async so now contains:
>>>> "TODO: find the IPs and attempt each one if this is a named service"
>>>> marks the spot.
>>>>
>>>> For now it converts IP-addressed services using raw-IP text conversion
>>>> "remote= s.cfg().host.termedBuf()". Which is synchronous and
>>>> non-blocking but not resolving FQDN.
>>>>
>>>> When FQDN resolution is required do we ...
>>>>
>>>> a) split openConnection() into two async steps and only use one of the
>>>> possibly multiple IPs returned?
>>>>
>>>> OR
>>>>
>>>> b) lookup the DNS at some far earlier point and store the IPA in
>>>> ServiceRep for some sort of load balancing selection of just one by
>>>> getConnection()?
>>>>
>>
>> This just needs:
>> openConnection step 1 to call ipcache_nbgethostbyname()
>> openConnection step 2 to do something similar to
>> src/peer_select.cc:peerSelectDnsResults() but tuned a bit for ICAP.
>
> For a complete fix, we also need to modify the callers to handle a
> two-step (i.e., async) openConnection(). AFAICT, both callers call
> openConnection() at the end of an already async call, so it is not a
> major problem, just more work.
>
>
> Thank you,
>
> Alex.
>
Received on Thu Jun 23 2011 - 17:23:23 MDT
This archive was generated by hypermail 2.2.0 : Fri Jun 24 2011 - 12:00:05 MDT