On Thu, Jul 31, 2008 at 5:53 PM, Dick Davies <[EMAIL PROTECTED]> wrote:
> On Thu, Jul 31, 2008 at 6:36 PM, Allen Gilliland
> <[EMAIL PROTECTED]> wrote:
>
>> Dave wrote:
>>>
>>> On Wed, Jul 30, 2008 at 6:05 AM, Dick Davies <[EMAIL PROTECTED]> wrote:
>
>>>> * turn off search
>>>> * have a shared uploads folder
>>>> * shared planet cache.dir directory
>>>> * arrange for each instance to have a unique tasks.clientId attribute in
>>>> roller-custom.properties
>
>> You also need some form of cluster safe cache implementation as well unless
>> you plan to disable all of Roller's caches. Without that you'll get
>> mismatching data between nodes.
>
> Hmm, that's a bit of a gotcha :)
> Might explain some weirdness we've been seeing though, thanks for the warning.
There is a memcached implementation of the Roller cache system at
roller.dev.java.net. Here's a HOWTO that explains how to use it:
http://rollerweblogger.org/roller/entry/howto_memcached_with_roller
- Dave
