Elasticsuite 2.9 unable to load correct spelling type (ES 6.x)

[ahajji06]

Preconditions

Magento Version : CE 2,3.5
Elasticsearch Version: 6.8.8
ElasticSuite Version : 2.9.0
Environment : Developer

Expected result

spelling type should be defined correctly

Actual result

It always returns: self::SPELLING_TYPE_EXACT
(from catch block in src/module-elasticsuite-core/Search/Adapter/Elasticsuite/Spellchecker.php)

Logs:
Request Failure: {"method":"POST","uri":"http://localhost/magento2_de_catalog_product/_termvectors?term_statistics=true&fields=spelling%2Cspelling.whitespace","port":9200,"headers":{"Host":["localhost"],"Content-Type":["application/json"],"Accept":["application/json"],"User-Agent":["elasticsearch-php/7.7.0 (Linux 4.19.0-9-amd64; PHP 7.3.19)"]},"HTTP code":400,"duration":0.000577,"error":"{"error":{"root_cause":[{"type":"illegal_argument_exception","reason":"request [/magento2_de_catalog_product/_termvectors] contains unrecognized parameters: [fields], [term_statistics]"}],"type":"illegal_argument_exception","reason":"request [/magento2_de_catalog_product/_termvectors] contains unrecognized parameters: [fields], [term_statistics]"},"status":400}"} []
Response ["{"error":{"root_cause":[{"type":"illegal_argument_exception","reason":"request [/magento2_de_catalog_product/_termvectors] contains unrecognized parameters: [fields], [term_statistics]"}],"type":"illegal_argument_exception","reason":"request [/magento2_de_catalog_product/_termvectors] contains unrecognized parameters: [fields], [term_statistics]"},"status":400}"] []
curl -XPOST 'http://localhost/magento2_de_catalog_product/_termvectors?term_statistics=true&fields=spelling%2Cspelling.whitespace' -d '{"doc":{"spelling":"fnster"}}' [] []

Any idea how i can fix that?
Thank you

[ahajji06]

the solution is to add the type to the request since it is needed for ES 6.8.8.
It should be added to this function:
(in src/module-elasticsuite-core/Search/Adapter/Elasticsuite/Spellchecker.php)

    private function getTermVectors(RequestInterface $request)
    {
        $termVectorsQuery = [
            'index'           => $request->getIndex(),
            'term_statistics' => true,
            'type'            => '_doc',
            'fields'          => [
                MappingInterface::DEFAULT_SPELLING_FIELD,
                MappingInterface::DEFAULT_SPELLING_FIELD . "." . FieldInterface::ANALYZER_WHITESPACE,
            ],
        ];

        $termVectorsQuery['body']['doc'] = [MappingInterface::DEFAULT_SPELLING_FIELD => $request->getQueryText()];

        return $this->client->termvectors($termVectorsQuery);
    }

the only problem that remains now is that this function is private, so we can´t modify its behavior via plugin :-(

Any ideas how to tackle that without risking update problems?

[rbayet]

Hello @ahajji06,

Edit: confirmed with Elasticsearch 6.8.3

The easiest way for you is to use a composer patch, using cweagans/composer-patches.
See also https://devdocs.magento.com/guides/v2.3/comp-mgr/patching.html#composer
It will fail by design to apply the patch when the bug is fixed and a new 2.9.x release of Elasticsuite is published.

In your case, for ES 6.8.x, the patch would be

diff --git a/src/module-elasticsuite-core/Search/Adapter/Elasticsuite/Spellchecker.php b/src/module-elasticsuite-core/Search/Adapter/Elasticsuite/Spellchecker.php
index 44ae906d..919f4d88 100644
--- a/src/module-elasticsuite-core/Search/Adapter/Elasticsuite/Spellchecker.php
+++ b/src/module-elasticsuite-core/Search/Adapter/Elasticsuite/Spellchecker.php
@@ -142,6 +142,7 @@ class Spellchecker implements SpellcheckerInterface
     {
         $termVectorsQuery = [
             'index'           => $request->getIndex(),
+            'type'            => '_doc',
             'term_statistics' => true,
             'fields'          => [
                 MappingInterface::DEFAULT_SPELLING_FIELD,

On our side, we will have to find a satisfying approach for both ES 6 and ES 7.

Regards,

[rbayet]

Hello @ahajji06,

I've just checked, Elasticsearch 7.x still accepts the _doc parameter for termVectors even if it is not present in the documentation and the returned data, even if not strictly speaking the same (fields ordered differently), is consistent with or without the parameter that is required for Elasticsearch 6.x.

Long story short, feel free to submit us a PR if you wish.

Regards,

[ahajji06]

Thank you 😊

[rbayet]

Re-opening because we need to fix that for everyone :)