lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sunil Kumar PK" <pksuni...@gmail.com>
Subject Re: Searching Problem
Date Fri, 27 Oct 2006 06:33:21 GMT
Thanks Mike for the information.

Actually I am using RemoteParallelMultiSearcher with 10 Search Servers, my
crawler program freequently add new documents in to all the Search Servers
in a distributed manner. So in this case, if I add a document in a
particular index, I need to restart the searcher program in that server.
right? Can I do this with a remote call, or I want to add a new method to
Searchable interface?

Thanks,
Sunil


On 10/26/06, Michael McCandless <lucene@mikemccandless.com> wrote:
>
> Sunil Kumar PK wrote:
> > could you please explain?
> >
> > On 10/26/06, Karel Tejnora <karel@tejnora.cz> wrote:
> >> Nope. IndexReader obtains a snapshot of index - not closing and opening
> >> indexreader leads to not deleting files (windows exception, linux will
> >> not free them).
> >> > Is it possible to get all the matching document in the result without
> >> > restarting the Searcher program?
>
> A searcher once created only searches the index as of the "point in
> time" that it was created.  Ie, it's an unchanging snapshot.  So any
> deletes/updates that happen to the index by a writer will not be visible
> until you close and reopen another searcher.
>
> This "point in time" searching relies on certain properties of the
> underlying filesystem in order to work properly.  Windows local and
> remote (SMB) filesystems work because files that are open can't be
> deleted (and Lucene just retries); local UNIX filesystems work because
> the open file handle can still access a deleted file ("delete on last
> close").
>
> However: NFS does not have "delete on last close", so you can't rely on
> "point in time" searching when using NFS across machines (it's possible
> a single machine may work).  If a writer on a different machine has
> committed to the index that a searcher is using over NFS then the
> searcher will eventually hit an IOException with "stale NFS handle".
> See here for details on current known issues with NFS:
>
>    http://issues.apache.org/jira/browse/LUCENE-673
>
> Mike
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-user-help@lucene.apache.org
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message