[prev in list] [next in list] [prev in thread] [next in thread] 

List:       squid-dev
Subject:    Re: Duplicate peers
From:       Mark Nottingham <mnot () yahoo-inc ! com>
Date:       2008-06-10 6:35:32
Message-ID: 3C0A71A0-9326-48AE-ACEA-7F9D9F310AEC () yahoo-inc ! com
[Download RAW message or body]

Yes; however, if the first peer fails but isn't considered down yet,  
it'll be retried. This isn't optimal for an accelerator, esp. if it's  
failing because of a connect timeout.


On 10/06/2008, at 4:17 PM, Henrik Nordstrom wrote:

> On tis, 2008-06-10 at 15:45 +1000, Mark Nottingham wrote:
>> Because peerGetSomeParent and then peerGetAllParents are both called
>> in the NEVER_DIRECT case (e.g., accelerators), the list of peers to
>> use will contain a duplicate; whatever peerGetSomeParent returns will
>> be in the list twice, and if it's the first peer (lexically in the
>> conf file), it'll be twice in a row. This isn't good, especially for
>> round-robin accelerator setups.
>
> It only matters if the first is unreachable or the request otherwise  
> has
> to be retried.
>
> The primary forwarding server is always the first in that list,  
> which is
> the round-robin selected peer if using round-robin.
>
> The other peers is there just to allow Squid to find an alternate path
> if the primary selected one fails.
>
> Regards
> Henrik

--
Mark Nottingham       mnot@yahoo-inc.com


[prev in list] [next in list] [prev in thread] [next in thread]