we had been doing some work with this, and had gotten to the
architecture stage on this at $WORK, but the guy who was leading the
charge got put onto other tasks, and left before having a chance to
implement it, and our priorities shifted to other things ;(
From what I remember, there are several issues in it, the biggest two were:
Handling when a replica is down, and getting it back when it comes up again
Keeping the caches current as you would need to do commits which would
ruin the caches (as it forces a reload I think)
doing 3-phase commits to ensure all replicas have identical
information. (we were planning on using a paxos algorithm to help with this)
Hopefully Jason if you are reading this will add further to it.
regards
Ian
Grant Ingersoll wrote:
Hi James,
Can you provide more information about what you are trying to do? By
real time search, do you mean you want indexed documents to be
available immediately? Or is a minute or two acceptable? Do all
users need to see them immediately, or just the current user?
We can better help you if you give us more details on what you are
trying to accomplish.
Thanks,
Grant
On Sep 23, 2007, at 10:38 AM, James liu wrote:
i wanna do it.
Maybe someone did it, if so, give me some tips.
thks
--
regards
jl
--------------------------
Grant Ingersoll
http://lucene.grantingersoll.com
Lucene Helpful Hints:
http://wiki.apache.org/lucene-java/BasicsOfPerformance
http://wiki.apache.org/lucene-java/LuceneFAQ
