Mailing List Archive: PrefixQuery,WildcardQuery,RangeQuery and FuzzyQuery PROBLEM

PrefixQuery,WildcardQuery,RangeQuery and FuzzyQuery PROBLEM

Jul 13, 2005, 3:21 AM

Post #1 of 4 (2628 views)

Hello

I have been readed about "Too many clauses"........... If the max was set too high, the inefficiency would make the search unsable.
I am testing the performance of Lucene and the time that spend Lucene in searching is too high. Moreover I´ve got OutOfMemory error several times.....
I am speaking about an index with 250.000 documents more or less, but in the future will be necessary an index with millions of documents.

These are the kinds of queries:

1. Greater than or lower than request
RangeQuery with Integer.MAX_VALUE for greater than or Integer.MIN_VALUE for lower than

2. RangeQuery

Example:

Field:[minValue to maxValue]

3.WildcardQuery

Example:

Field:value*

ect....

The problem is that PrefixQuery,WildcardQuery,RangeQuery and FuzzyQuery all expand to a series of OR'ed boolean queries.

I have read about BitSetQuery, FilteringQuery, ConstrantScoreQuery.......... I am confused!!!!!!

I can´t use a Filter (DateFilter, QueryFilter ect...) because the client wants to search for all the documents without filter for anything.
I can´t divide a field in subfields to do the query more specific. For example, the user wants the date with format YYYMMDDHHMMSS, not 6 fields, one with the year, one with the month, one with the day, one with de hour ect....
I can´t add more system resources.

My environment is the next:
----LUCENE 1.4.3-------
INDEX ==> 200.000 documents to million of documents
EACH DOCUMENT +- 20 fields (metadatas)
SIZE TEXT DOCUMENT 1k
-----SERVER (dedicated) -------
Red Hat
2 GB Memory
jboss + lucene
JAVA_OPTS -Xmx640M -Xms640M

My question is very simple...... Is it possible to use Lucene like full text search engine with the environment I have explained before, with the server that I have explained before, and doing the queries that I have explained before with an efficient performance and without OutOfMemoryError????

Thanks in advance

Mari Luz

---------------------------------------------------
Mari Luz Elola
Developer Engineer

Caleruega, 67
28033 Madrid (Spain)
Tel.: +34 91 768 46 58
mailto: melola@seinet.es
---------------------------------------------------

Privileged/Confidential Information may be contained in this message and is intended solely for the use of the named addressee(s). Access to this e-mail by anyone else is unauthorised. If you are not the intended recipient, any disclosure, copying, distribution or re-use of the information contained in it is prohibited and may be unlawful. Opinions, conclusions and any other information contained in this message that do not relate to the official business of Seinet shall be understood as neither given nor endorsed by it. If you have received this communication in error, please notify us immediately by replying to this mail and deleting it from your computer.
Thank you.

Re: PrefixQuery,WildcardQuery,RangeQuery and FuzzyQuery PROBLEM [ In reply to ]

erik at ehatchersolutions

Jul 13, 2005, 3:53 AM

Post #2 of 4 (2536 views)

Permalink

On Jul 13, 2005, at 6:21 AM, MariLuz Elola wrote:
> I have been readed about "Too many clauses"........... If the
> max was set too high, the inefficiency would make the search unsable.
> I am testing the performance of Lucene and the time that spend
> Lucene in searching is too high. Moreover I´ve got OutOfMemory
> error several times.....
> I am speaking about an index with 250.000 documents more or
> less, but in the future will be necessary an index with millions of
> documents.
>
> These are the kinds of queries:
>
> 1. Greater than or lower than request
> RangeQuery with Integer.MAX_VALUE for greater than or
> Integer.MIN_VALUE for lower than
>
> 2. RangeQuery
>
> Example:
>
> Field:[minValue to maxValue]

Keep in mind that dealing with numeric information requires some
adjustments both at how you index and how RangeQuerys are formed.
For example, if you index "1" through "10" doing a RangeQuery of [1
TO 5] will also find "10" unless you account for it with a special
QueryParser subclass.

> 3.WildcardQuery
>
> Example:
>
> Field:value*
>
> ect....
>
> The problem is that PrefixQuery,WildcardQuery,RangeQuery and
> FuzzyQuery all expand to a series of OR'ed boolean queries.
>
> I have read about BitSetQuery, FilteringQuery,
> ConstrantScoreQuery.......... I am confused!!!!!!

There certainly are lots of options. The Query classes you mention,
though, are not currently exposed via QueryParser, so you would need
to subclass QueryParser to have them created instead, or create your
own parser, or mix and match some query expression parsing and join
it with some API created Querys via BooleanQuery.

> I can´t use a Filter (DateFilter, QueryFilter ect...) because the
> client wants to search for all the documents without filter for
> anything.

This doesn't make sense to me. Implicitly the user is "filtering"
documents by adding constraints to a query expression using
Field:value* or Field:[min TO max].

> I can´t divide a field in subfields to do the query more specific.
> For example, the user wants the date with format YYYMMDDHHMMSS, not
> 6 fields, one with the year, one with the month, one with the day,
> one with de hour ect....

The index structure needs to be a bit more abstracted from the user
in your case, it seems. The user does not need to know explicitly
that the index is split into multiple fields for dates in order to
make searching more efficient. If the user is not doing queries down
to the second level, but rather always at the day level, then you
can build the index to account for that type of usage and improve the
experience.

I encourage you to reconsider your "can't"'s and investigate
alternative approaches. Such considerations might be - does the user
really need FuzzyQuery? Are WildcardQuery's desired? If so, what
types of wildcard queries are needed? (this can affect how you index
and construct queries - a WildcardQuery literally is not the only way
to achieve the same sort of thing, as has been mentioned using a
PhraseQuery for numeric information) Can the user interface be
crafted to be more structured rather than just a Google-like search
box where the user has to enter field selectors and know QueryParser
voodoo? (perhaps the date field constraint can use a date picker
rather than a textual expression?)

> My question is very simple...... Is it possible to use Lucene like
> full text search engine with the environment I have explained
> before, with the server that I have explained before, and doing the
> queries that I have explained before with an efficient performance
> and without OutOfMemoryError????

Short answer: yes.

Longer answer: see above for some techniques to consider

Erik

Re: PrefixQuery,WildcardQuery,RangeQuery and FuzzyQuery PROBLEM [ In reply to ]

paul.elschot at xs4all

Jul 13, 2005, 11:00 AM

Post #3 of 4 (2540 views)

Permalink

On Wednesday 13 July 2005 12:53, Erik Hatcher wrote:
>
> On Jul 13, 2005, at 6:21 AM, MariLuz Elola wrote:
> > I have been readed about "Too many clauses"........... If the
> > max was set too high, the inefficiency would make the search unsable.
> > I am testing the performance of Lucene and the time that spend
> > Lucene in searching is too high. Moreover I´ve got OutOfMemory
> > error several times.....
> > I am speaking about an index with 250.000 documents more or
> > less, but in the future will be necessary an index with millions of
> > documents.
> >
> > These are the kinds of queries:
> >
> > 1. Greater than or lower than request
> > RangeQuery with Integer.MAX_VALUE for greater than or
> > Integer.MIN_VALUE for lower than
> >
> > 2. RangeQuery
> >
> > Example:
> >
> > Field:[minValue to maxValue]
>
> Keep in mind that dealing with numeric information requires some
> adjustments both at how you index and how RangeQuerys are formed.
> For example, if you index "1" through "10" doing a RangeQuery of [1
> TO 5] will also find "10" unless you account for it with a special
> QueryParser subclass.
>
> > 3.WildcardQuery
> >
> > Example:
> >
> > Field:value*
> >
> > ect....
> >
> > The problem is that PrefixQuery,WildcardQuery,RangeQuery and
> > FuzzyQuery all expand to a series of OR'ed boolean queries.
> >
> > I have read about BitSetQuery, FilteringQuery,
> > ConstrantScoreQuery.......... I am confused!!!!!!
>
> There certainly are lots of options. The Query classes you mention,
> though, are not currently exposed via QueryParser, so you would need
> to subclass QueryParser to have them created instead, or create your
> own parser, or mix and match some query expression parsing and join
> it with some API created Querys via BooleanQuery.
>
> > I can´t use a Filter (DateFilter, QueryFilter ect...) because the
> > client wants to search for all the documents without filter for
> > anything.
>
> This doesn't make sense to me. Implicitly the user is "filtering"
> documents by adding constraints to a query expression using
> Field:value* or Field:[min TO max].
>
> > I can´t divide a field in subfields to do the query more specific.
> > For example, the user wants the date with format YYYMMDDHHMMSS, not
> > 6 fields, one with the year, one with the month, one with the day,
> > one with de hour ect....
>
> The index structure needs to be a bit more abstracted from the user
> in your case, it seems. The user does not need to know explicitly
> that the index is split into multiple fields for dates in order to
> make searching more efficient. If the user is not doing queries down
> to the second level, but rather always at the day level, then you
> can build the index to account for that type of usage and improve the
> experience.

One can also index all of these (or even more) and hide them from the user:

YYY
YYYMM
YYYMMDD
YYYMMDDHH
YYYMMDDHHMM
YYYMMDDHHMMSS

With this, searching ranges would require subclassing the QueryParser
with classes that implement the range search using as few terms as possible.
That should bring down the number of terms used to a some logarithm
of the total range size.

Regards,
Paul Elschot

RE: PrefixQuery,WildcardQuery,RangeQuery and FuzzyQuery PROBLEM [ In reply to ]

james at ryley

Jul 13, 2005, 11:10 AM

Post #4 of 4 (2533 views)

Permalink

> One can also index all of these (or even more) and hide them from the
> user:
>
> YYY
> YYYMM
> YYYMMDD
> YYYMMDDHH
> YYYMMDDHHMM
> YYYMMDDHHMMSS
>
> With this, searching ranges would require subclassing the QueryParser
> with classes that implement the range search using as few terms as
> possible.
> That should bring down the number of terms used to a some logarithm
> of the total range size.

Yep, that's what we do for large date ranges and it works fine. Just a
little logic to determine the way to expand the user input to the least
possible terms.

Sincerely,
James