Indexes

This article is part of the XQuery Portal. It contains information on the available index structures.

The query compiler tries to optimize and speed up queries by applying the index whenever it is possible and seems promising. To see how a query is rewritten, and if an index is used, you can turn on the Info View in the GUI or use the -V flag on the command line:


 * A message like  indicates that the text index is applied to speed up the search of the shown string. The following message…
 * indicates that a string in a path expression will never yield results. Because of that, the path does not need to be evaluated at all.
 * If you cannot find any index optimization hints in the info output, it often helps if you rewrite and simplify your query.

=Structural Indexes=

Structural indexes are automatically created and cannot be dropped by the user:

Name Index
The name index contains references to the names of all elements and attributes in a database. It contains some basic statistical information, such as the number of occurrence of a name.

The name index is e.g. applied to discard location steps that will never yield results:

 (: will be rewritten to an empty sequence :) /non-existing-name

The contents of the name indexes can be directly accessed with the XQuery functions index:element-names and index:attribute-names.

If a database is updated, new names will be added incrementally, but the statistical information will get out-dated.

Path Index
The path index (which is also called path summary or data guide) stores all distinct paths of the documents in the database. It contains additional statistical information, such as the number of occurrence of a path, its distinct string values, and the minimum/maximum of numeric values. The maximum number of distinct values to store per name can be changed via. Distinct values are also stored for elements and attributes of numeric type.

Various queries will be evaluated much faster if an up-to-date path index is available (as can be observed when opening the Info View):


 * Descendant steps will be rewritten to multiple child steps. Child steps are evaluated faster, as fewer nodes have to be traversed:

 doc('factbook.xml')//province, (: ...will be rewritten to... :) doc('factbook.xml')/mondial/country/province


 * The fn:count function will be pre-evaluated by looking up the number in the index:

 count(doc('factbook')//country)


 * The distinct values of elements or attributes can be looked up in the index as well:

 distinct-values(db:open('factbook')//religions)

The contents of the path index can be directly accessed with the XQuery function index:facets.

If a database is updated, the statistics in the path index will be invalidated.

Document Index
The document index contains references to all document nodes in a database. Once documents with specific paths are requested, the index will be extended to also contain document paths.

The index generally speeds up access to single documents and database paths. It will always be kept up-to-date.

=Value Indexes=

Value indexes can be created and dropped by the user. Four types of values indexes are available: a text and attribute index, and an optional token and full-text index. By default, the text and attribute index will automatically be created.

In the GUI, index structures can be managed in the dialog windows for creating new databases or displaying the database properties. On command-line, the commands and  are used to create and drop index structures. With, you get some insight into the contents of an index structure, and allows you to change the index defaults for new databases:


 * : Open database; create full-text index
 * : Open database; show info on token index
 * : Enable attribute index; only index 'id' and 'name' attributes; create database

With XQuery, index structures can be created and dropped via $$:

 (: Optimize specified database, create full-text index for texts of the specified elements :) db:optimize( 'factbook',  false,  map { 'ftindex': true, 'ftinclude': 'p div' } )

Exact Queries
This index references text nodes of documents. It will be utilized to accelerate string comparisons in path expressions. The following queries will all be rewritten for index access:

 (: example 1 :) //*[text = 'Germany'], (: example 2 :) doc('factbook.xml')//name[. = 'Germany'], (: example 3 :) for $c in db:open('factbook')//country where $c//city/name = 'Hanoi' return $c/name

Before the actual index rewriting takes places, some preliminary optimizations are applied:
 * In example 2, the context item expression . will be replaced with a text step.
 * In example 3, the where clause will be rewritten to a predicate and attached to the first path expression.

The indexed text nodes can be accessed directly with the XQuery function $$. The indexed string values can be looked up via $$.

The option can be enabled to keep this index up-to-date:

 db:optimize( 'mydb',  true,  map { 'updindex':true, 'textindex': true, 'textinclude':'id' } )

Range Queries
The text index also supports range queries based on string comparisons:

 (: example 1 :) db:open('Library')//Medium[Year >= '2011' and Year <= '2016'], (: example 2 :) let $min := '2014-04-16T00:00:00' let $max := '2014-04-19T23:59:59' return db:open('news')//entry[date-time > $min and date-time < $max]

With $$, you can access all text nodes whose values are between a minimum and maximum value.

Please note that the index structures do not support queries for numbers and dates.

Attribute Index
Similar to the text index, this index speeds up string and range comparisons on attribute values. Additionally, the XQuery function fn:id takes advantage of the index whenever possible. The following queries will all be rewritten for index access:

 (: 1st example :) //country[@car_code = 'J'], (: 2nd example :) //province[@* = 'Hokkaido']//name, (: 3rd example :) //sea[@depth > '2100' and @depth < '4000'] (: 4th example :) fn:id('f0_119', db:open('factbook'))

Attribute nodes (which you can use as starting points of navigation) can directly be retrieved from the index with the XQuery functions $$ and $$. The index contents (strings) can be accessed with $$.

The option can be activated to keep this index up-to-date.

Token Index
In many XML dialects, such as HTML or DITA, multiple tokens are stored in attribute values. The token index can be created to speed up the retrieval of these tokens. The XQuery functions fn:contains-token, fn:tokenize and fn:idref are rewritten for index access whenever possible. If a token index exists, it will e.g. be utilized for the following queries:

 (: 1st example :) //div[contains-token(@class, 'row')], (: 2nd example :) //p[tokenize(@class) = 'row'], (: 3rd example :) doc('graph.xml')/idref('edge8')

Attribute nodes with a matching value (containing at least one from a set of given tokens) can be directly retrieved from the index with the XQuery function $$. The index contents (token strings) can be accessed with $$.

Full-Text Index
The Full-Text index contains the normalized tokens of text nodes of a document. It is utilized to speed up queries with the contains text expression, and it is capable of processing wildcard and fuzzy search operations. Three evaluation strategies are available: the standard sequential database scan, a full-text index based evaluation and a hybrid one, combining both strategies (see XQuery Full Text implementation in BaseX).

If the full-text index exists, the following queries will all be rewritten for index access:

 (: 1st example :) //country[name/text contains text 'and'], (: 2nd example :) //religions[.//text contains text { 'Catholic', 'Roman' } using case insensitive distance at most 2 words]

The index provides support for the following full-text features (the values can be changed in the GUI or via the command):


 * Stemming: tokens are stemmed before being indexed (option: )
 * Case Sensitive: tokens are indexed in case-sensitive mode (option: )
 * Diacritics: diacritics are indexed as well (option: )
 * Stopword List: a stop word list can be defined to reduce the number of indexed tokens (option: )
 * Language: see Languages for more details (option: )

The options that have been used for creating the full-text index will also be applied to the optimized full-text queries. However, the defaults can be overwritten if you supply options in your query. For example, if words were stemmed in the index, and if the query can be rewritten for index access, the query terms will be stemmed as well, unless stemming is not explicitly disabled. This is demonstrated in the following Command Script:

 true true  house  /text[. contains text { 'houses' }] /text[. contains text { 'houses' } using no stemming]

Text nodes can be directly requested from the index via the XQuery function $$. The index contents can be accessed with $$.

Selective Indexing
Value indexing can be restricted to specific elements and attributes. The nodes to be indexed can be restricted via the, , and  options. The options take a list of name patterns, which are separated by commas. The following name patterns are supported:


 * : all names
 * : elements or attributes called, which are in the empty default namespace
 * : elements or attributes called, no matter which namespace
 * : all elements or attributes in the  namespace
 * : elements or attributes called  in the   namespace

The options can either be specified via the command or via XQuery. With the following operations, an attribute index is created for all id and name attributes:

<pre class="brush:xml"> SET ATTRINCLUDE id,name CREATE DB factbook http://files.basex.org/xml/factbook.xml' SET ATTRINCLUDE
 * Commands
 * 1) Restore default

<pre class="brush:xquery"> db:create('factbook', 'http://files.basex.org/xml/factbook.xml', '', map { 'attrinclude': 'id,name' })
 * XQuery

With and $$, new selective indexing options will ba applied to an existing database.

Enforce Rewritings
In various cases, existing index structures will not be utilized by the query optimizer. This is usually the case if the name of the database is not a static string (e.g., because it is bound to a variable or passed on as argument of a function call). Furthermore, several candidates for index rewritings may exist, and the query optimizer may decide for a rewriting that turns out to be suboptimal.

With the option, certain index rewritings can be enforced. While the option can be globally enabled, it is usually better to supply it as Pragma. Two examples:


 * In the query below, 10 databases will be addressed. If it is known in advance that these databases contain an up-to-date text index, the index rewriting can be enforced as follows:

<pre class="brush:xquery"> (# db:enforceindex #) { for $n in 1 to 10 let $db := 'persons' || $n return db:open($db)//person[name/text = 'John'] }


 * The following query contains two predicates that may both be rewritten for index access. If the automatically chosen rewriting is known not to be optimal, another index rewriting can enforced by surrounding the specific expression with the pragma:

<pre class="brush:xquery"> db:open('factbook')//country [(# db:enforceindex #) { @population > '10000000' and @population < '10999999' }] [religions/text = 'Protestant']

The option can also be assigned to predicates with dynamic values. In the following example the comparison of the first comparison will be rewritten for index access. Without the pragma expression, the second comparison is preferred and chosen for the rewriting, because the statically known string allows for an exact cost estimation:

<pre class="brush:xquery"> for $name in ('Germany', 'Italy') for $country in db:open('factbook')//country where (# db:enforceindex #) { $country/name = $name } where $country/religions/text = 'Protestant' return $country

Please note that:


 * The option should only be enabled if the addressed databases exist, have all required index structures and are up-to-date (otherwise, you will be given an error message).
 * If you address the full-text index, and if you use non-default indexing options, you will have to specify them in your query (via using stemming, using language 'de', etc).
 * If you have more than one enforce pragma in a single path expression, only the first will be considered.
 * In general, there are always expressions that cannot be rewritten for index access. If you enforce rewritings, you will have no guarantee that an index will be used.

=Custom Index Structures=

With XQuery, it is comparatively easy to create your own, custom index structures. The following query demonstrate how you can create a factbook-index database, which contains all texts of the original database in lower case:

<pre class="brush:xquery"> let $db := 'factbook'

let $index := { for $nodes in db:open($db)//text group by $text := lower-case($nodes) return <text string='{ $text }'>{ for $node in $nodes return <id>{ db:node-id($node ) }</id> } }

return db:create($db || '-index', $index, $db || '-index.xml')

In the following query, a text string is searched, and the text nodes of the original database are retrieved:

<pre class="brush:xquery"> let $db := 'factbook' let $text := 'italian' for $id in db:open($db || '-index')//*[@string = $text]/id return db:open-id($db, $id)/..

With some extra effort, and if is enabled for both your original and your index database (see below), your index database will support updates as well (try it, it’s fun!).

=Performance=

If main memory runs out while creating a value index, the current index structures will be partially written to disk and eventually merged. If the memory heuristics fail for some reason (i.e., because multiple index operations run at the same time, or because the applied JVM does not support explicit garbage collections), a fixed index split sizes may be chosen via the option.

If is enabled, the command-line output might help you to find a good split size. The following example shows the output for creating a database for an XMark document with 1 GB, and with 128 MB assigned to the JVM:

> basex -d -c"SET FTINDEX ON; SET TOKENINDEX ON; CREATE DB xmark 1gb.xml" Creating Database... ................................ 76559.99 ms (29001 KB) Indexing Text... ....|...|...|.....|. 9.81 M operations, 18576.92 ms (13523 KB). Recommended SPLITSIZE: 20. Indexing Attribute Values... .........|....... 3.82 M operations, 7151.77 ms (6435 KB). Recommended SPLITSIZE: 20. Indexing Tokens... .......|..|.....|.. 3.82 M operations, 9636.73 ms (10809 KB). Recommended SPLITSIZE: 10. Indexing Full-Text... ..|.|.|.|...|...|..|.|..| 116.33 M operations, 138740.94 ms (106 MB). Recommended SPLITSIZE: 12.

The output can be interpreted as follows:


 * The vertical bar  indicates that a partial index structure was written to disk.
 * The mean value of the recommendations can be assigned to the option. Please note that the recommendation is only a vague proposal, so try different values if you get main-of-memory errors or indexing gets too slow. Greater values will require more main memory.
 * In the example, the full-text index was split 12 times. 116 million tokens were indexed, processing time was 2,5 minutes, and final main memory consumption (after writing the index to disk) was 76 MB. A good value for the split size option could be 15.

=Updates=

Generally, update operations are very fast in BaseX. By default, the index structures will be invalidated by updates; as a result, queries that benefit from index structures may slow down after updates. There are different alternatives to cope with this:


 * After the execution of one or more update operations, the command or the $$ function can be called to rebuild the index structures.
 * The option can be activated before creating or optimizing the database. As a result, the text, attribute and token indexes will be incrementally updated after each database update. Please note that incremental updates are not available for the full-text index and database statistics. This is also explains why the UPTODATE flag, which is e.g. displayed via  or $$, will be set to false until the database will be optimized again (various optimizations won’t be triggered. For example, count(//item) can be extremely fast if all meta data is up-to-date.
 * The option can be enabled before creating or optimizing the database. All outdated index structures and statistics will then be recreated after each database update. This option should only be done for small and medium-sized databases.
 * Both options can be used side by side: will take care that the value index structures will be updated as part of the actual update operation.  will update the remaining data structures (full-text index, database statistics).

=Changelog=


 * Version 9.1


 * Updated: Enforce Rewritings, support for comparisons with dynamic values.


 * Version 9.0


 * Added: Enforce Rewritings


 * Version 8.4


 * Updated: Name Index, Path Index


 * Version 8.4


 * Added: Token Index


 * Version 8.3


 * Added: Selective Indexing
 * Version 8.0


 * Added: AUTOOPTIMIZE option


 * Version 7.2.1


 * Added: string-based range queries