Skip to content
This repository has been archived by the owner on May 4, 2023. It is now read-only.
/ couchdb-lucene Public archive

Enables full-text searching of CouchDB documents using Lucene

License

Notifications You must be signed in to change notification settings

rnewson/couchdb-lucene

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

couchdb-lucene Build Status

Version Compatibility

CouchDB-Lucene works with all version of CouchDB from 0.10 upwards.

Issue Tracking

Issue tracking at github.

Minimum System Requirements

Java 1.8 is required; Oracle Java 8 or OpenJDK 8 are recommended.

Build and run couchdb-lucene

If you are on OS X, you might find it easiest to:

brew install couchdb-lucene
  1. Install Maven (2 or 3).
  2. checkout repository
  3. type 'mvn'
  4. cd target
  5. unzip couchdb-lucene-<version>.zip
  6. cd couchdb-lucene-<version>
  7. ./bin/run

The zip file contains all the couchdb-lucene code, dependencies, startup scripts and configuration files you need, so unzip it wherever you wish to install couchdb-lucene.

If you want to run couchdb-lucene on a servlet container like Tomcat, you can build the war file using Maven

mvn war:war

Configure CouchDB

The following settings are needed in CouchDB's local.ini file in order for it to communicate with couchdb-lucene:

Proxy handler (for CouchDB versions from 1.1 onward)

[httpd_global_handlers]
_fti = {couch_httpd_proxy, handle_proxy_req, <<"http://127.0.0.1:5985">>}

Python hook script (for CouchDB versions prior to 1.1)

[couchdb]
os_process_timeout=60000 ; increase the timeout from 5 seconds.

[external]
fti=/path/to/python /path/to/couchdb-lucene/tools/couchdb-external-hook.py

[httpd_db_handlers]
_fti = {couch_httpd_external, handle_external_req, <<"fti">>}

Hook options

You can pass options to the python script like so:

[external]
fti=/path/to/python "/path/to/couchdb-lucene/tools/couchdb-external-hook.py --option-name value"
Option Meaning Default Value
--remote-host The hostname of the couchdb-lucene server localhost
--remote-port The port of the couchdb-lucene server 5985
--local-key The key for the local couchdb instance as known to the couchdb-lucene server local

Configure couchdb-lucene

couchdb-lucene runs in a single, standalone JVM. As such, you can choose to locate your couchdb-lucene server on a different machine to couchdb if you wish, or keep it on the same machine, it's your call.

Start couchdb-lucene

To start couchdb-lucene, run:

bin/run

To stop couchdb-lucene, simply kill the Java process.

Indexing Strategy

Document Indexing

You must supply a index function in order to enable couchdb-lucene as, by default, nothing will be indexed. To suppress a document from the index, return null. It's more typical to return a single Document object which contains everything you'd like to query and retrieve. You may also return an array of Document objects if you wish.

You may add any number of index views in any number of design documents. All searches will be constrained to documents emitted by the index functions.

Here's an complete example of a design document with couchdb-lucene features:

{
    "_id":"_design/foo",
    "fulltext": {
        "by_subject": {
            "index":"function(doc) { var ret=new Document(); ret.add(doc.subject); return ret }"
        },
        "by_content": {
            "index":"function(doc) { var ret=new Document(); ret.add(doc.content); return ret }"
        }
    }
}

Here are some example URL's for the given design document:

Using the Python hook script

http://localhost:5984/database/_fti/_design/foo/by_subject?q=hello
http://localhost:5984/database/_fti/_design/foo/by_content?q=hello

Using the proxy handler

http://localhost:5984/_fti/local/database/_design/foo/by_subject?q=hello
http://localhost:5984/_fti/local/database/_design/foo/by_content?q=hello

A fulltext object contains multiple index view declarations. An index view consists of:

analyzer (optional) The analyzer to use

defaults (optional) The default for numerous indexing options can be overridden here. A full list of options follows.

index The indexing function itself, documented below.

The Defaults Object

The following indexing options can be defaulted:

name description available options default
field the field name to index under user-defined default
type the type of the field date, double, float, int, long, string, text text
store whether the data is stored. The value will be returned in the search result. yes, no no
boost Sets the boost factor hits on this field. This value will be multiplied into the score of all hits on this this field of this document. floating-point value 1.0

String vs Text

There are two supported types that sound equivalent, string and text, but they are very different. A text field will be tokenized into words and is usually what you expect from a full-text index. A string field is not tokenized, only exact matching will work. The advantage to string fields is that they have a meaningful sort order.

The Analyzer Option

Lucene has numerous ways of converting free-form text into tokens, these classes are called Analyzer's. By default, the StandardAnalyzer is used which lower-cases all text, drops common English words ("the", "and", and so on), among other things. This processing might not always suit you, so you can choose from several others by setting the "analyzer" field to one of the following values:

  • brazilian
  • chinese
  • cjk
  • czech
  • dutch
  • english
  • french
  • german
  • keyword
  • perfield
  • porter
  • russian
  • simple
  • snowball
  • standard
  • thai
  • whitespace
  • ngram
The Snowball Analyzer

This analyzer requires an extra argument to specify the language (see here for details):

"analyzer":"snowball:English"

Note: the argument is case-sensitive and is passed directly to the SnowballAnalyzer's constructor.

The Per-field Analyzer

The "perfield" option lets you use a different analyzer for different fields and is configured as follows:

"analyzer":"perfield:{field_name:\"analyzer_name\"}"

Unless overridden, any field name not specified will be handled by the standard analyzer. To change the default, use the special default field name:

"analyzer":"perfield:{default:\"keyword\"}"
The Ngram Analyzer

The "ngram" analyzer lets you break down the output of any other analyzer into ngrams ("foo" becomes "fo" and "oo").

"analyzer":"ngram:{analyzer:\"simple\",min:2,max:3}"

If not specified, the delegated analyzer is "standard" and min and max ngram sizes are 1 and 2 respectively.

Configuring additional analyzers

There are many other analyzers included in Lucene and there are also occasions where custom analyzers not included in Lucene are needed. There is now support for configuring additional analyzers without needing to further modify couchdb-lucene.

The Document class

You may construct a new Document instance with:

var doc = new Document();

Data may be added to this document with the add method which takes an optional second object argument that can override any of the above default values.

// Add with all the defaults.
doc.add("value");

// Add a numeric field.
doc.add(35, {"type":"int"});

// Add a date field.
doc.add(new Date("1972/1/6 16:05:00"),        {"type":"date"});
doc.add(new Date("January 6, 1972 16:05:00"), {"type":"date"});

// Add a date field (object must be a Date object

// Add a subject field.
doc.add("this is the subject line.", {"field":"subject"});

// Add but ensure it's stored.
doc.add("value", {"store":"yes"});

// Add but don't analyze.
doc.add("don't analyze me", {"index":"not_analyzed"});

// Extract text from the named attachment and index it to a named field
doc.attachment("attachment field", "attachment name");

// log an event (trace, debug, info, warn and error are available)
if (doc.foo) {
  log.info("doc has foo property!");
}

Example Index Functions

Index Everything
function(doc) {
    var ret = new Document();

    function idx(obj) {
	for (var key in obj) {
	    switch (typeof obj[key]) {
	    case 'object':
		idx(obj[key]);
		break;
	    case 'function':
		break;
	    default:
		ret.add(obj[key]);
		break;
	    }
	}
    };

    idx(doc);

    if (doc._attachments) {
	for (var i in doc._attachments) {
	    ret.attachment("default", i);
	}
    }

    return ret;
}
Index Nothing
function(doc) {
  return null;
}
Index Select Fields
function(doc) {
  var result = new Document();
  result.add(doc.subject, {"field":"subject", "store":"yes"});
  result.add(doc.content, {"field":"subject"});
  result.add(new Date(), {"field":"indexed_at"});
  return result;
}
Index Attachments
function(doc) {
  var result = new Document();
  for(var a in doc._attachments) {
    result.attachment("default", a);
  }
  return result;
}
A More Complex Example
function(doc) {
    var mk = function(name, value, group) {
        var ret = new Document();
        ret.add(value, {"field": group, "store":"yes"});
        ret.add(group, {"field":"group", "store":"yes"});
        return ret;
    };
    if(doc.type != "reference") return null;
    var ret = new Array();
    for(var g in doc.groups) {
        ret.push(mk("library", doc.groups[g].library, g));
        ret.push(mk("method", doc.groups[g].method, g));
        ret.push(mk("target", doc.groups[g].target, g));
    }
    return ret;
}

Attachment Indexing

Couchdb-lucene uses Apache Tika to index attachments of the following types, assuming the correct content_type is set in couchdb;

Supported Formats

  • Excel spreadsheets (application/vnd.ms-excel)
  • HTML (text/html)
  • Images (image/*)
  • Java class files
  • Java jar archives
  • MP3 (audio/mp3)
  • OpenDocument (application/vnd.oasis.opendocument.*)
  • Outlook (application/vnd.ms-outlook)
  • PDF (application/pdf)
  • Plain text (text/plain)
  • Powerpoint presentations (application/vnd.ms-powerpoint)
  • RTF (application/rtf)
  • Visio (application/vnd.visio)
  • Word documents (application/msword)
  • XML (application/xml)

Searching with couchdb-lucene

You can perform all types of queries using Lucene's default query syntax.

Numeric range queries

In addition to normal text-based range searches (using the "field:[lower TO upper]" syntax), couchdb-lucene also supports numeric range searches for the following types: int, long, float, double and date. The type is specified after the field name, as follows:

type example
int field:[0 TO 100]
long field:[0 TO 100]
float field:[0.0 TO 100.0]
double field:[0.0 TO 100.0]
date field:[from TO to] where from and to match any of these patterns: "yyyy-MM-dd'T'HH:mm:ssZ", "yyyy-MM-dd'T'HH:mm:ss", "yyyy-MM-ddZ", "yyyy-MM-dd", "yyyy-MM-dd'T'HH:mm:ss.SSSZ", "yyyy-MM-dd'T'HH:mm:ss.SSS". So, in order to search for articles published in July, you would issue a following query: published_at<date>:["2010-07-01T00:00:00"+TO+"2010-07-31T23:59:59"]

An example numeric range query for spatial searching.

?q=pizza AND lat<double>:[51.4707 TO 51.5224] AND long<double>:[-0.6622 TO -0.5775]

Numeric term queries

Fields indexed with numeric types can still be queried as normal terms, couchdb-lucene just needs to know the type. For example, ?q=age<long>:12 will find all documents where the field called 'age' has a value of 12 (when the field was indexed as "type":"int".

Search methods

You may use HTTP GET or POST. For POST, use application/x-www-form-urlencoded format.

Search parameters

The following parameters can be passed for more sophisticated searches:

analyzer Override the default analyzer used to parse the q parameter

callback Specify a JSONP callback wrapper. The full JSON result will be prepended with this parameter and also placed with parentheses."

debug Setting this to true disables response caching (the query is executed every time) and indents the JSON response for readability.

default_operator Change the default operator for boolean queries. Defaults to "OR", other permitted value is "AND".

force_json Usually couchdb-lucene determines the Content-Type of its response based on the presence of the Accept header. If Accept contains "application/json", you get "application/json" in the response, otherwise you get "text/plain;charset=utf8". Some tools, like JSONView for FireFox, do not send the Accept header but do render "application/json" responses if received. Setting force_json=true forces all response to "application/json" regardless of the Accept header.

include_docs whether to include the source docs

include_fields By default, all stored fields are returned with results. Use a comma-separate list of field names with this parameter to refine the response

highlights Number of highlights to include with results. Default is 0. This uses the fast-vector-highlighter plugin.

highlight_length Number of characters to include in a highlight row. Default and minimum is 18.

limit the maximum number of results to return. Default is 25.

q the query to run (e.g, subject:hello). If not specified, the default field is searched. Multiple queries can be supplied, separated by commas; the resulting JSON will be an array of responses.

skip the number of results to skip

sort the comma-separated fields to sort on. Prefix with / for ascending order and \ for descending order (ascending is the default if not specified). Type-specific sorting is also available by appending the type between angle brackets (e.g, sort=amount). Supported types are 'float', 'double', 'int', 'long' and 'date'.

stale=ok If you set the stale option to ok, couchdb-lucene will not block if the index is not up to date and it will immediately return results. Therefore searches may be faster as Lucene caches important data (especially for sorting). A query without stale=ok will block and use the latest data committed to the index. Unlike CouchDBs stale=ok option for views, couchdb-lucene will trigger an index update unless one is already running.

All parameters except 'q' are optional.

Special Fields

_id The _id of the document.

Dublin Core

All Dublin Core attributes are indexed and stored if detected in the attachment. Descriptions of the fields come from the Tika javadocs.

_dc.contributor An entity responsible for making contributions to the content of the resource.

_dc.coverage The extent or scope of the content of the resource.

_dc.creator An entity primarily responsible for making the content of the resource.

_dc.date A date associated with an event in the life cycle of the resource.

_dc.description An account of the content of the resource.

_dc.format Typically, Format may include the media-type or dimensions of the resource.

_dc.identifier Recommended best practice is to identify the resource by means of a string or number conforming to a formal identification system.

_dc.language A language of the intellectual content of the resource.

_dc.modified Date on which the resource was changed.

_dc.publisher An entity responsible for making the resource available.

_dc.relation A reference to a related resource.

_dc.rights Information about rights held in and over the resource.

_dc.source A reference to a resource from which the present resource is derived.

_dc.subject The topic of the content of the resource.

_dc.title A name given to the resource.

_dc.type The nature or genre of the content of the resource.

Examples

Using the Python hook script

http://localhost:5984/dbname/_fti/_design/foo/view_name?q=field_name:value
http://localhost:5984/dbname/_fti/_design/foo/view_name?q=field_name:value&sort=other_field
http://localhost:5984/dbname/_fti/_design/foo/view_name?debug=true&sort=billing_size<long>&q=body:document AND customer:[A TO C]

Using the proxy handler

http://localhost:5984/_fti/local/dbname/_design/foo/view_name?q=field_name:value
http://localhost:5984/_fti/local/dbname/_design/foo/view_name?q=field_name:value&sort=other_field
http://localhost:5984/_fti/local/dbname/_design/foo/view_name?debug=true&sort=billing_size<long>&q=body:document AND customer:[A TO C]

Search Results Format

The search result contains a number of fields at the top level, in addition to your search results.

etag An opaque token that reflects the current version of the index. This value is also returned in an ETag header to facilitate HTTP caching.

fetch_duration The number of milliseconds spent retrieving the documents.

limit The maximum number of results that can appear.

q The query that was executed.

rows The search results array, described below.

search_duration The number of milliseconds spent performing the search.

skip The number of initial matches that was skipped.

total_rows The total number of matches for this query.

The search results array

The search results arrays consists of zero, one or more objects with the following fields:

doc The original document from couch, if requested with include_docs=true

fields All the fields that were stored with this match

id The unique identifier for this match.

score The normalized score (0.0-1.0, inclusive) for this match

Here's an example of a JSON response without sorting:

{
  "q": "+content:enron",
  "skip": 0,
  "limit": 2,
  "total_rows": 176852,
  "search_duration": 518,
  "fetch_duration": 4,
  "rows":   [
        {
      "id": "hain-m-all_documents-257.",
      "score": 1.601625680923462
    },
        {
      "id": "hain-m-notes_inbox-257.",
      "score": 1.601625680923462
    }
  ]
}

And the same with sorting:

{
  "q": "+content:enron",
  "skip": 0,
  "limit": 3,
  "total_rows": 176852,
  "search_duration": 660,
  "fetch_duration": 4,
  "sort_order":   [
        {
      "field": "source",
      "reverse": false,
      "type": "string"
    },
        {
      "reverse": false,
      "type": "doc"
    }
  ],
  "rows":   [
        {
      "id": "shankman-j-inbox-105.",
      "score": 0.6131107211112976,
      "sort_order":       [
        "enron",
        6
      ]
    },
        {
      "id": "shankman-j-inbox-8.",
      "score": 0.7492915391921997,
      "sort_order":       [
        "enron",
        7
      ]
    },
        {
      "id": "shankman-j-inbox-30.",
      "score": 0.507369875907898,
      "sort_order":       [
        "enron",
        8
      ]
    }
  ]
}

Content-Type of response

The Content-Type of the response is negotiated via the Accept request header like CouchDB itself. If the Accept header includes "application/json" then that is also the Content-Type of the response. If not, "text/plain;charset=utf-8" is used.

Fetching information about the index

Calling couchdb-lucene without arguments returns a JSON object with information about the index.

http://127.0.0.1:5984/<db>/_fti/_design/foo/<index>

returns:

{"current":true,"disk_size":110674,"doc_count":397,"doc_del_count":0,
"fields":["default","number"],"last_modified":"1263066382000",
"optimized":true,"ref_count":2}

Index Maintenance

For optimal query speed you can optimize your indexes. This causes the index to be rewritten into a single segment.

curl -X POST http://localhost:5984/<db>/_fti/_design/foo/<index>/_optimize

If you just want to expunge pending deletes, then call:

curl -X POST http://localhost:5984/<db>/_fti/_design/foo/<index>/_expunge

If you recreate databases or frequently change your fulltext functions, you will probably have old indexes lying around on disk. To remove all of them, call:

curl -X POST http://localhost:5984/<db>/_fti/_cleanup

Authentication

By default couchdb-lucene does not attempt to authenticate to CouchDB. If you have set CouchDB's require_valid_user to true, you will need to modify couchdb-lucene.ini. Change the url setting to include a valid username and password. e.g, the default setting is:

[local]
url=http://localhost:5984/

Change it to:

[local]
url=http://foo:bar@localhost:5984/

and couchdb-lucene will authenticate to couchdb.

Other Tricks

A couple of 'expert' options can be set in the couchdb-lucene.ini file;

Leading wildcards are prohibited by default as they perform very poorly most of the time. You can enable them as follows:

[lucene]
allowLeadingWildcard=true

Lucene automatically converts terms to lower case in wildcard situations. You can disable this with:

[lucene]
lowercaseExpandedTerms=false

CouchDB-Lucene will keep your indexes up to date automatically but this consumes resources (network sockets). You can ask CouchDB-Lucene to stop updating an index after a timeout with:

[lucene]
changes_timeout = 60000