lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Erick Erickson <erickerick...@gmail.com>
Subject Re: new to lucene- some questions regarding internals
Date Tue, 11 Aug 2015 17:13:41 GMT
1-3 are really answered by the same explanation:

When you open a searcher, lucene "knows" what all the closed segments
are (i.e., the last commit point). And you can't commit when only part
of a document has been written to the current segment. You can think
of commits as atomic at the document level. At least they'd better be
or it's a serious bug.

Until you open a new searcher that view of the index is kept. All
add/delete/update operations since you opened that searcher do NOT
operate on that view of the index. New updates are written to new
segments which aren't part of that view. Deleted docs are marked as
deleted, but the currently open searcher doesn't have a clue about
those.

When searchers are closed, any segments that have been "merged away"
are deleted from disk.

And since when a new searcher is opened, it only sees closed segments,
there should be no way to get partial docs. In that sense,
add/updated/delete indexing operations are atomic.

4> Sure. In fact all of the 4 possibilities are possible for
stored=T|F, indexed=T|F, although stored=false and indexed=false is
essentially a no-op.

Best,
Erick

On Tue, Aug 11, 2015 at 4:20 AM, Yechiel Feffer <yechielf@gigaspaces.com> wrote:
> Hi
> 1. as I understand Lucene is preparing the documents of the search result in a lazy fashion-
using the docId in the ScoreDoc.  What happens if the document "pointed" by the ScoreDoc is
deleted  meanwhile i.e. the DocId is not relevant (maybe assigned to a different  document)
?
> 2. when a document is added to an index- (addDocument)  I guess there is a period in
which the fields are added to the index one by one. What prevents an indexsearcher (esp' in
NRT mode) to see "half"  document , i.e is it atomic and how?
> 3 The way to update doc is remove the old and add the updated.  Isnt there a problem
of a "hole" between the remove & the addDocument in which no document exists , i.e is
it atomic and how?
> 4 is it possible to store a field in the lucene index without indexing it ? (i.e. just
gettng it back in the result docs)
>
> Thanks,
> Yechiel
>

---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org


Mime
View raw message