Mailing List Archive

> From: Ype Kingma [mailto:ykingma@xs4all.nl]
>
> Suppose I would like to retrieve all docs that are resulting
> from a query.
> I should then use the search() call with the HitCollector argument
> which is called back with collect(docNr, score)
>
> Would it be wise to sort by docNr when using IndexReader.doc(docNr) to
> get to the stored fields?

Yes, that would be the most efficient way. Note that HitCollector.collect()
is called with increasing document ids, so you don't actually have to
sort--they arrive in order.

> And another question. I looked at the source for searcher.close() and
> found that it closes its reader. Does that close the index reader used
> to perform the search? That would interfere with a strategy to keep
> an index reader open as long as possible and share it between threads.

You shouldn't close the Searcher. Reuse it too.

I may have confused things in an earlier message. One should keep a single
Searcher which uses a single IndexReader. Thus the code I posted earlier
for caching an IndexReader should really be caching a Searcher. The
IndexReader will be cached implicitly by the IndexSearcher. Thus the
correct code for caching the IndexReader in "latest" should be:

private Searcher searcher;
private long lastModified;
public synchronized Searcher getSearcher() throws IOException {
if (lastModified != IndexReader.lastModified("latest")) {
// there's a new index: open it
lastModified = IndexReader.lastModified("latest");
searcher = new IndexSearcher("latest");
}
return searcher;
}

Doug

--
To unsubscribe, e-mail: <mailto:lucene-user-unsubscribe@jakarta.apache.org>
For additional commands, e-mail: <mailto:lucene-user-help@jakarta.apache.org>

Are you implying ( ... public synchronized Searcher getSearcher()....) to
use this synchronized method in a servlet/jsp thread as well? Is that a
good idea? Your jhtml example doesn't appear to synchronzied. Maybe I'm
missing something though.




Doug Cutting
<DCutting@grandce To: "'Lucene Users List'"
ntral.com> <lucene-user@jakarta.apache.org>
cc:
01/31/2002 02:11 Subject: RE: Obtaining all results
PM efficiently. Closing a searcher.
Please respond to
"Lucene Users
List"






> From: Ype Kingma [mailto:ykingma@xs4all.nl]
>
> Suppose I would like to retrieve all docs that are resulting
> from a query.
> I should then use the search() call with the HitCollector argument
> which is called back with collect(docNr, score)
>
> Would it be wise to sort by docNr when using IndexReader.doc(docNr) to
> get to the stored fields?

Yes, that would be the most efficient way. Note that HitCollector.collect
()
is called with increasing document ids, so you don't actually have to
sort--they arrive in order.

> And another question. I looked at the source for searcher.close() and
> found that it closes its reader. Does that close the index reader used
> to perform the search? That would interfere with a strategy to keep
> an index reader open as long as possible and share it between threads.

You shouldn't close the Searcher. Reuse it too.

I may have confused things in an earlier message. One should keep a single
Searcher which uses a single IndexReader. Thus the code I posted earlier
for caching an IndexReader should really be caching a Searcher. The
IndexReader will be cached implicitly by the IndexSearcher. Thus the
correct code for caching the IndexReader in "latest" should be:

private Searcher searcher;
private long lastModified;
public synchronized Searcher getSearcher() throws IOException {
if (lastModified != IndexReader.lastModified("latest")) {
// there's a new index: open it
lastModified = IndexReader.lastModified("latest");
searcher = new IndexSearcher("latest");
}
return searcher;
}

Doug

--
To unsubscribe, e-mail: <
mailto:lucene-user-unsubscribe@jakarta.apache.org>
For additional commands, e-mail: <
mailto:lucene-user-help@jakarta.apache.org>





--
To unsubscribe, e-mail: <mailto:lucene-user-unsubscribe@jakarta.apache.org>
For additional commands, e-mail: <mailto:lucene-user-help@jakarta.apache.org>

> From: Jonathan_Wasson@dom.com [mailto:Jonathan_Wasson@dom.com]
>
> Are you implying ( ... public synchronized Searcher
> getSearcher()....) to
> use this synchronized method in a servlet/jsp thread as
> well?

Yes.

> Your jhtml example doesn't appear to
> synchronzied. Maybe I'm missing something though.

That's because it uses a Hashtable, which is already synchronized. But in
the method I sent today, when writing the long lastModified, explicit
synchronization is required. (Java does not guarantee that 64-bit stores
are atomic.) It also wouldn't hurt to synchronize in the jhtml example,
although I don't think its required. It would keep multiple threads from
potentially re-opening the same index several times when it is modified,
improving efficiency.

Doug

--
To unsubscribe, e-mail: <mailto:lucene-user-unsubscribe@jakarta.apache.org>
For additional commands, e-mail: <mailto:lucene-user-help@jakarta.apache.org>