You won't need to synchronize the writers yourself as there are locks 
internally, but I would avoid multiple threads sharing a writer for performance 
reasons.

Michael

-----Original Message-----
From: Li Bing [mailto:lbl...@gmail.com] 
Sent: Thursday, October 29, 2009 8:02 PM
To: Michael Garski
Cc: lucene-net-user@incubator.apache.org
Subject: Re: Exceptions When Indexing with Multi-threading

Dear Michael,

Yes, each thread in my system has its own IndexWriter. But all of them
are synchronized by a unique lock. So I think the exception should not
occur, right? If the threads share a single IndexWriter, what about
the performance? I don't need to synchronize by myself if so, right?

I appreciate so much for your help!

LB

On Fri, Oct 30, 2009 at 12:33 AM, Michael Garski
<mgar...@myspace-inc.com> wrote:
> Does each thread have its own IndexWriter?  This could cause the issue
> you describe.
>
> IndexWriter is thread-safe so that multiple threads can use the same
> instance, however you will see a performance degradation compared to
> only having one thread modify the index.  We've done quite a bit of
> testing for that case a while back and only have one thread modifying an
> index at a time.
>
> Michael
>
> -----Original Message-----
> From: Li Bing [mailto:lbl...@gmail.com]
> Sent: Wednesday, October 28, 2009 10:57 PM
> To: lucene-net-user@incubator.apache.org
> Cc: Li Bing - Gmail
> Subject: Exceptions When Indexing with Multi-threading
>
> Dear all,
>
> Today I got exceptions when indexing with multi-threading. It says
> that "deletable.new" or "segments.new" are accessed by other
> processes. So it cannot rename ...
>
> However, I have used "ReaderWriterSlim" lock to avoid the relevant
> folders to be accessed by multiple threads at the same time. For
> example, if the indexing is done in D:\Test\, I use "ReaderWriterSlim"
> to prevent other threads from writing in the folder. I think the above
> exceptions should not happen if I do that. However, it is NOT true.
> Why?
>
> Any other ways to avoid the problem?
>
> Thanks so much!
> LB
>
>
>


Reply via email to