Mailing-List: contact java-user-help@lucene.apache.org; run by ezmlm
Precedence: bulk
Reply-To: java-user@lucene.apache.org
Received-SPF: pass (athena.apache.org: domain of xekoukou@gmail.com designates
 74.125.82.53 as permitted sender)
MIME-Version: 1.0
In-Reply-To: <D9315AAD-0323-4E63-A5D9-361E7E4E81D8@gmail.com>
References: 
 <CAOX4E5EZv0NhdcYKexZSWWjpaR1f0va6FiZt-aEwc=khc3ki_g@mail.gmail.com>
	<CAL8PwkZRb_hAeBYfwSeh595OmhvMd3sXq2wE=FBtcUFp0MKEKQ@mail.gmail.com>
	<D9315AAD-0323-4E63-A5D9-361E7E4E81D8@gmail.com>
Date: Tue, 22 Jan 2013 14:08:09 +0200
Message-ID: 
 <CAOX4E5F22+CJ9rjehMc6ESvbDx89vDg9woLDCeBVKNxYs51e8w@mail.gmail.com>
Subject: Re: Is LogByteSizeMergePolicy deterministic?
From: Apostolis Xekoukoulotakis <xekoukou@gmail.com>
To: java-user@lucene.apache.org
Content-Type: multipart/alternative; boundary=f46d044306065f40c004d3df6ec4

--f46d044306065f40c004d3df6ec4
Content-Type: text/plain; charset=ISO-8859-1

I havent started working on this yet. I use the LogByteSizeMergePolicy
because of personal reasons that have nothing to do with replication.
I havent found a lucene replication method. (?)

>From my research, there are 2 ways to replicate an index.

a) You stop indexing. You make a "hard" commit and you replicate the files
of this commit.
or you change the deletion policy and simply replicate an old commit while
you continue to index.

b)You send documents to both replicas which they index independently. This
is done to take advantage of the near-real-time feature of lucene.

when one replica goes offline but is intact, If the missed documents are a
lot you do a), else you replay the missed documents to the replica.

Method a) also seems to not be durable for the new documents which havent
been replicated yet. We have only one replica of them.


2013/1/22 Denis Bazhenov <dotsid@gmail.com>

> Can you explain in more details why is that? We have in-house replication
> for Lucene/3.6 index and use default IndexWriter settings. All works fine
> except sometimes (just after optimization, in fact) index could not be
> opened (segment file is missing on FS). We tolerate this issue by
> replicating index one more time in case of failure. I guess it's somehow
> related to the discussed issue.
>
> On Jan 19, 2013, at 5:16 AM, Michael McCandless <lucene@mikemccandless.com>
> wrote:
>
> > You must also use only a single indexing thread.
> >
> > And you must use SerialMergeScheduler.
> >
> > If you do that, I think it will be deterministic.
> >
> > But don't rely on this ... this is runtime behavior and can suddenly
> > change between releases ...
> >
> > Mike McCandless
> >
> > http://blog.mikemccandless.com
> >
> > On Fri, Jan 18, 2013 at 10:39 AM, Apostolis Xekoukoulotakis
> > <xekoukou@gmail.com> wrote:
> >> I want to replicate an index from multiple replicas at the same time.
> >>
> >> Those replicas have been given the same documents and at the same order.
> >>
> >> Will the files be the same across all replicas?
> >>
> >>
> >>
> >> --
> >>
> >>
> >> Sincerely yours,
> >>
> >>     Apostolis Xekoukoulotakis
> >
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> > For additional commands, e-mail: java-user-help@lucene.apache.org
> >
>
> ---
> Denis Bazhenov <dotsid@gmail.com>
>
>
>
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-user-help@lucene.apache.org
>
>


-- 


Sincerely yours,

     Apostolis Xekoukoulotakis

--f46d044306065f40c004d3df6ec4--