Today I had the job to check out, why the search results of a customer are so weird – and even weirder if you use more concrete search terms.

The consequence is, that if you search for „washing machine“ you get a ton of washing machines, but if you search for „washing machine $brand“, you get all washing machines and all machines from $brand.

To debug this I finally found a way to get an explanation what is happening: The explain query. With this query you can ask ES/OS how the scoring for a search term and a document is calculated and therefore get an impression how to optimise. The linked article is great, two small additions:

1. You don’t need a special field, but can use a simple query for all fields
2. The query is a POST query

POST /$indexName/_explain/$documentId
{
  "query": {
    "query_string": {
      "query": "my search query"
    }
  }
}

Turns out: Elastic search gives great search results and the order is awesome too!

BUT…

But Shopware not only once, but in my opinion twice throws the scoring away:

// \Shopware\Core\Content\Product\SalesChannel\Listing\ProductListingLoader::load
public function load(Criteria $origin, SalesChannelContext $context): EntitySearchResult
{
    $origin->addState(Criteria::STATE_ELASTICSEARCH_AWARE);
    $criteria = clone $origin;

    $this->addGrouping($criteria);
    $this->handleAvailableStock($criteria, $context);

    $ids = $this->productRepository->searchIds($criteria, $context);
    /** @var list<string> $keys */
    $keys = $ids->getIds();
    $aggregations = $this->productRepository->aggregate($criteria, $context);

        // no products found, no need to continue
        if (empty($keys)) {
            return new EntitySearchResult(
                ProductDefinition::ENTITY_NAME,
                0,
                new ProductCollection(),
                $aggregations,
                $origin,
                $context->getContext()
            );
        }

        $mapping = array_combine($keys, $keys);

In line 10 the score is in $ids, but latest in line 27 they are gone.

And in the hydrator, the same happens:

// \Shopware\Elasticsearch\Framework\DataAbstractionLayer\ElasticsearchEntitySearchHydrator::hydrate
public function hydrate(EntityDefinition $definition, Criteria $criteria, Context $context, array $result): IdSearchResult
{
    if (!isset($result['hits'])) {
        return new IdSearchResult(0, [], $criteria, $context);
    }

    $hits = $this->extractHits($result);

    $data = [];
    foreach ($hits as $hit) {
        $id = $hit['_id'];

        $data[$id] = [
            'primaryKey' => $id,
            'data' => array_merge(
                $hit['_source'] ?? [],
                ['id' => $id, '_score' => $hit['_score']]
            ),
        ];
    }

I don’t now exactly why but my understanding is, that $result has greatly ordered products, but after the foreach they are gone. I assume because PHP is handling the explicit key set somehow and orders the keys. MAYBE(!) this is only a debugging artefact!

I hope I am allowed to debug this further and fix it. Stay tuned for more.

Other articles from this category