Camille Hodoul

Compose Elastica queries in PHP

December 07, 2019

Elastica is an Elasticsearch client library for PHP.
Among other things, it handles transport and provides a class-based abstraction over the transiting JSON objects.
While Elastica can be used for other tasks such as bulk indexing documents, this post will focus on the Query API.

Query building

As long as you’re building queries whose general structure doesn’t change much, the API is straightforward.
However, I find that when dealing with more complex search techniques such as faceted search, where fields and operators change according to user input, the code can get awkward if written in an imperative way.

As an example, imagine the following index type describing movies:

/**
* This is an oversimplification of
* an actual Elasticsearch mapping
*/
$mapping = [
    "title" => "text",
    "genre" => "keyword",
    "director" => "keyword",
];

And the following search parameters sent from the front-end:

$request = [
    "title" => ["q" => "Apocalypse"],
    "genre" => [ "q" => ["war", "horror"], "operator" => "OR"]
];

A naive imperative implementation could look like this:

use \Elastica\ResultSet;
use \Elastica\Query\AbstractQuery;
use \Elastica\Query\BoolQuery;
use \Elastica\Query\Match;
use \Elastica\Query\SimpleQueryString;

function getQuery($request): AbstractQuery {
    $query = new BoolQuery();
    if (isset($request["title"])) {
        $fullTextQuery = new SimpleQueryString($request["title"]["q"] . "*", [
            "title",
        ]);
        $fullTextQuery->setDefaultOperator(SimpleQueryString::OPERATOR_AND);
        $fullTextQuery->setParam("analyze_wildcard", true);
        $query->addMust($fullTextQuery);
    }
    if (isset($request["genre"])) {
        if ($request["genre"]["operator"] === "OR") {
            $orQuery = new BoolQuery();
            $subQueries = [];
            foreach ($request["genre"]["q"] as $value) {
                $matchQuery = new Match();
                $matchQuery->setFieldQuery("title", $request["title"]["q"]);
                $subQueries[] = $matchQuery;
            }
            $orQuery->addShould($subQueries);
            $orQuery->setMinimumShouldMatch(1);
            $query->addMust($orQuery);
        } elseif ($request["genre"]["operator"] === "AND") {
            foreach ($request["genre"]["q"] as $value) {
                $matchQuery = new Match();
                $matchQuery->setFieldQuery("title", $request["title"]["q"]);
                $query->addMust($matchQuery);
            }
        }
    }
    /** ... */

    return new \Elastica\Query($query);
}

As you can imagine, as you add new searchable fields or field types, it can get difficult to read and maintain.

A touch of functional flavor

Let’s refactor this code to something more manageable using closures.

First, we create a helper for the usual operators AND, OR and NOT.

use \Elastica\Query\BoolQuery;

function getOperatorFn(string $operator): callable {
    $defaultOperator = "AND";
    // one could declare a function for each operator
    // to be more "declarative"
    $operatorFunctions = [
        "AND" => function (BoolQuery $query, array $subQueries): BoolQuery {
            if (count($subQueries) <= 0) {
                return $query;
            }
            $query->addMust($subQueries);
            return $query;
        },
        "OR" => function (BoolQuery $query, array $subQueries): BoolQuery {
            if (count($subQueries) <= 0) {
                return $query;
            }
            $orQuery = new BoolQuery();
            $orQuery->addShould($subQueries);
            $orQuery->setMinimumShouldMatch(1);
            $query->addMust($orQuery);
            return $query;
        },
        "NOT" => function (BoolQuery $query, array $subQueries): BoolQuery {
            if (count($subQueries) <= 0) {
                return $query;
            }
            $query->addMustNot($subQueries);
            return $query;
        }
    ];
    $actualOperator = isset($operatorFunctions[$operator]) ? $operator : $defaultOperator;
    return $operatorFunctions[$actualOperator];
}

Then, we need to figure out what kind of Query to build for a given field:

use \Elastica\Query\SimpleQueryString;
use \Elastica\Query\Match;
use \Elastica\Query\SimpleQueryString;
use \Elastica\Query\AbstractQuery;

function getQueryBuilderForField(string $field, array $mapping): callable {
    $fieldType = $mapping[$field];
    if ($fieldType === "title") {
        return function(string $value) use($field): AbstractQuery {
            $fullTextQuery = new SimpleQueryString(
                $value . "*", 
                [$field]
            );
            $fullTextQuery->setDefaultOperator(SimpleQueryString::OPERATOR_AND);
            $fullTextQuery->setParam("analyze_wildcard", true);
            return $fullTextQuery;
        };
    }
    if ($fieldType === "keyword") {
        return function(string $value) use($field): AbstractQuery {
            $matchQuery = new Match();
            $matchQuery->setFieldQuery($field, $value);
            return $matchQuery;
        }
    }
    throw new \InvalidArgumentException("$fieldType is not supported");
}

Finally, let’s create the actual query.

use \Elastica\Query\AbstractQuery;
use function \Functional\reduce_left;
use function \Functional\map;

function getQuery(array $request, array $mapping): AbstractQuery {
    $query = reduce_left(
        $request,
        function ($search, $field, $_, $query) use($mapping): BoolQuery {
            $operator = getOperatorFn($search["operator"]);
            $q = is_array($search["q"]) ? $search["q"] : [ $search["q"] ];
            return $operator(
                $query,
                map(
                    $q, 
                    getQueryBuilderForField($field, $mapping)
                )
            );
        },
        new BoolQuery()
    );
    return new \Elastica\Query($query);
}

You can of course use the built-in array_* functions instead of functional-php.


Camille Hodoul

I'm a JavaScript and PHP developer living in Marseille, France.
Twitter, Github, Flickr