Indexing Basics

Elasticsearch documents are stored in "indexes", each which contain a "type".

It's easy to think of Elasticsearch indexes as the RDBMS equivalent of a database, and types as the equivalent of a table, however there are notable differences in the underlying architecture.

​Elasticsearch engineer Adrien Grand on the comparisons:

the way data is stored is so different that any comparisons can hardly make sense, and this ultimately led to an overuse of types in cases where they were more harmful than helpful.

An index is stored in a set of shards, which are themselves Lucene indices. This already gives you a glimpse of the limits of using a new index all the time: Lucene indices have a small yet fixed overhead in terms of disk space, memory usage and file descriptors used. For that reason, a single large index is more efficient than several small indices: the fixed cost of the Lucene index is better amortized across many documents.

On types:

types are a convenient way to store several types of data in the same index, in order to keep the total number of indices low for the reasons exposed above... One nice property of types is that searching across several types of the same index comes with no overhead compared to searching a single type: it does not change how many shard results need to be merged

• From Index vs. Type on the Elastic blog.

In short, indexes have a higher overhead and make the aggregation of search results between types very more expensive. If it is desired that your application search interfaces return multiple entity or domain types, then those should respresent distinctive types within a single index, allowing them to be aggregated, sorted, and ordered in search results.

Retreiving information on Indices

To retreive a list of all indices on the connected cluster, use the client getIndices method:

var indexMap = getInstance( "[email protected]" ).getIndices();

This will return a struct of all indexes ( with the names as keys ), which will provide additional information on each index, such as:

• Any assigned aliases

• The number of documents in the index

• The size of the storage space used for the index in bytes

Creating and Mapping an Index

The IndexBuilder model assists with the creation and mapping of indexes. Mappings define the allowable data types within your documents and allow for better and more accurate search aggregations. Let's say we have a book model that we intend to make searchable. We are storing this in our bookshop index, under the type of book. Let's create the index (if it doesn't exist) and map the type of book:

var indexBuilder = getInstance( "[email protected]" ).new(
    "bookshop",
    {
        "books" = {
            "_all" = { "enabled" = false },
            "properties" = {
                "title" = { "type" = "string" },
                "summary" = { "type" = "string" },
                "description" = { "type" = "string" },
                // denotes a nested struct with additional keys
                "author" = { "type" = "object" },
                // date with specific format type
                "publishDate" = {
                    "type" = "date",
                    // our format will be = yyyy-mm-dd
                    "format" = "strict_date"
                },
                "edition" = { "type" = "integer" },
                "ISBN" = { "type" = "integer" }
            }
        }
    }
).save();

If you use Elasticsearch > 6, replace above "type":"string" with "type":"text" (Elasticsearch has dropped the string type and is now using text).

We can also add mappings after the new() method is called:

// instantiate the index builder
var indexBuilder = getInstance( "[email protected]" ).new( "bookshop" );
// our mapping struct
var booksMapping = {
    "_all" = { "enabled" = false },
    "properties" = {
        "title" = { "type" = "string" },
        "summary" = { "type" = "string" },
        "description" = { "type" = "string" },
        // denotes a nested struct with additional keys
        "author" = { "type" = "object" },
        // date with specific format type
        "publishDate" = {
            "type" = "date",
            // our format will be = yyyy-mm-dd
            "format" = "strict_date"
        },
        "edition" = { "type" = "integer" },
        "ISBN" = { "type" = "integer" }
    }
};

Deprecation notice: The index "type" ( e.g. "books" ) has now been deprecated in recent versions of Elasticsearch, and should no longer be used. Only a single type will be accepted in future releases.

Note that, in the above examples, we are applying the index and mappings directly from within the object, itself, which is intuitive and fast. We could also pass the IndexBuilder object to the [email protected] instance's applyIndex( required IndexBuilder indexBuilder ) method, if we wished.

If an explicit mapping is not specified when the index is created, Elasticsearch will assign types when the first document is saved.

We've also passed a simple struct in to the index properties. If we wanted to add additional settings or configure replicas and shards, we could pass a more comprehensive struct, including a range of settings to the new() method to do so:

indexBuilder.new(
    "bookshop",
    {
        "settings" = {
            "number_of_shards" = 10,
            "number_of_replicas" = 2,
            "auto_expand_replicas" = true,
            "shard.check_on_startup" = "checksum"
        },
        "mappings" = {
            "books" = {
                "_all" = { "enabled" = false },
                "properties" = {
                    "title" = { "type" = "string" },
                    "summary" = { "type" = "string" },
                    "description" = { "type" = "string" },
                    // denotes a nested struct with additional keys
                    "author" = { "type" = "object" },
                    // date with specific format type
                    "publishDate" = {
                        "type" = "date",
                        // our format will be = yyyy-mm-dd
                        "format" = "strict_date"
                    },
                    "edition" = { "type" = "integer" },
                    "ISBN" = { "type" = "integer" }
                }
            }
        }
    }
​
);

Additional Reading:

• ​Elasticsearch Mapping Guide​

• ​Index Settings Reference​

Retreiving information on Aliases

The client's getAliases method allows you to retreive a map containing information on aliases in use in the connected cluster.

var aliasMap = getInstance( "[email protected]" ).getAliases();

The corresponding object will have two keys: aliases and unassgined. The former is a map of aliases with their corresponding index, the latter is an array of indexes which are unassigned to any alias.

Alias Builder

cbElasticSearch can add or remove aliases in bulk using the applyAliases method on the cbElasticSearch client.

var removeAliasAction = getWireBox().getInstance( "[email protected]" )
    .remove( indexName = "testIndexName", aliasName = "aliasNameOne" );
var addNewAliasAction = getWireBox().getInstance( "[email protected]" )
    .add( indexName = "testIndexName", aliasName = "aliasNameTwo" );
​
variables.client.applyAliases(
    // a single alias action can also be provided
    aliases = [ removeAliasAction, addNewAliasAction ]
);

These operations will be done in the same transaction, so it's safe to use for switching the alias from one index to another.

Mapping Builder

Introduced in v1.0.0 the MappingBuilder model provides a fluent closure-based sytax for defining and mapping indexes. This builder can be accessed by injecting it into your components:

component {
    property name="builder" inject="[email protected]";
}

The new method of the IndexBuilder also accepts a closure as the second (properties) argument. If a closure is passed, a MappingBuilder instance is passed as an argument to the closure:

indexBuilder.new( "elasticsearch", function( builder ) {
    return {
        "_doc" = builder.create( function( mapping ) {
            mapping.text( "title" );
            mapping.date( "createdTime" ).format( "date_time_no_millis" );
        } )
    };
} );

The MappingBuilder has one primary method: create. create takes a callback with a MappingBlueprint object, usually aliased as mapping.

Mapping Blueprint

The MappingBlueprint gives a fluent api to defining a mapping. It has methods for all the ElasticSearch mapping types:

builder.create( function( mapping ) {
    mapping.text( "title" );
    mapping.date( "createdTime" ).format( "date_time_no_millis" );
    mapping.object( "user", function( mapping ) {
        mapping.keyword( "gender" );
        mapping.integer( "age" );
        mapping.object( "name", function( mapping ) {
            mapping.text( "first" );
            mapping.text( "last" );
        } );
    } );
} )

As seen above, object expects a closure which will be provided another MappingBlueprint. The results will be set as the properties of the object call.

Parameters

Parameters can be chained on to a mapping type. Parameters are set using onMissingMethod and will use the method name (as snake case) as the parameter name and the first argument passed as the parameter value.

builder.create( function( mapping ) {
    mapping.text( "title" ).fielddata( true );
    mapping.date( "createdTime" ).format( "date_time_no_millis" );
} )

You can also add parameters using the addParameter( string name, any value ) or setParameters( struct map ) methods.

The only exception to the parameters functions is fields which expects a closure argument and allows you to create multiple field definitions for a mapping.

builder.create( function( mapping ) {
    mapping.text( "city" ).fields( function( mapping ) {
        mapping.keyword( "raw" );
    } );
} );

Partials

The Mapping Blueprint also has a way to reuse mappings. Say for instance you have a user mapping that gets repeated for managers as well.

The partial method accepts three different kinds of arguments: 1. A closure 1. A component with a getPartial method 1. A WireBox mapping to a component with a getPartial method

var partialFn = function( mapping ) {
    return mapping.object( "user", function( mapping ) {
        mapping.integer( "age" );
        mapping.object( "name", function( mapping ) {
            mapping.text( "first" );
            mapping.text( "last" );
        } );
    } );
};
​
builder.create( function( mapping ) {
    mapping.partial( "manager", partialFn );
    mapping.partial( definition = partialFn ); // uses the partial's defined name, `user` in this case
} );

The first approach is great for partials that are reused in the same index. The second two approaches work better for partials that are reused across indexes.

Reindexing

On occassion, due to a mapping or settings change, you will need to reindex data from one index to another. You can do this by calling the reindex method on the Client.

getInstance( "[email protected]" )
    .reindex( "oldIndex", "newIndex" );

If you want the work to be done asynchronusly, you can pass false to the waitForCompletion flag. When this flag is set to false the method will return a Task instance, which can be used to follow up on the completion status of the reindex process.

getInstance( "[email protected]" )
    .reindex(
        source = "oldIndex",
        destination = "newIndex"
        waitForCompletion = false
    );

If you have more settings or constraints for the reindex action, you can pass a struct containing valid options to source and destination.

getInstance( "[email protected]" )
    .reindex(
        source = {
            "index": "oldIndex",
            "type": "testdocs",
            "query": {
                "term": {
                    "active": true
                }
            }
        },
        destination = "newIndex"
    );

You may also pass a script in to the reindex method to transform objects as they are being transferred from one index to another:

getInstance( "[email protected]" )
    .reindex(
        source = {
            "index": "oldIndex",
            "type": "testdocs",
            "query": {
                "term": {
                    "active": true
                }
            }
        },
        destination = "newIndex",
        script = {
            "lang" : "painless",
            "source" : "if( ctx._source.foo != null && ctx._source.foo == 'baz' ){ ctx._source.foo = 'bar'; }"
        }
    );

If you waitForCompletion and the reindex action fails, a cbElasticsearch.JestClient.ReindexFailedException will be thrown. You can disable the exception by passing false to the throwOnError parameter.