Elasticsearch for Bitrix: 10x faster search

When Bitrix built-in search stops working

With 100 thousand products, the built-in MySQL-based search slows down: queries to b_search_content take 300–500 ms, and on a million-record database — timeout. Clients complain about long search times, price sorting doesn't work, facets overload the database. For example, a recent project with a catalog of 250,000 products: standard search returned results in 2–3 seconds, complex property filters (price, brand, size) caused 30-second pauses. Elasticsearch solves this radically: response time 5–30 ms, aggregations on the fly, typos and morphology out of the box. Our experience — 30+ ES integrations with Bitrix, we guarantee a turnkey result. We use licensed components and are 1C-Bitrix certified. To understand if Elasticsearch is right for your project, order a free audit — we will analyze the load and data structure.

Problems that Elasticsearch solves

Elasticsearch replaces the standard Bitrix search engine and solves three key problems:

1. Full-text search speed. MySQL FULLTEXT starts slowing down at 50–100 thousand documents. ES handles millions.
2. Faceted search (aggregations). Bitrix built-in filters are separate queries for each property. ES returns aggregations in one request, giving a 10–20x gain on complex filters.
3. Search with typos and synonyms. Without third-party modules, MySQL does not support fuzzy search. ES has built-in fuzziness and synonyms.

If you face similar issues, contact us for an audit — it will help identify bottlenecks.

Why Elasticsearch is faster than MySQL FULLTEXT?

Elasticsearch is 10–50 times faster than MySQL FULLTEXT on large data.

Integration architecture

The integration consists of three parts:

1. Indexer — a component that reads data from Bitrix (infoblocks, users, pages) and writes documents to Elasticsearch index.
2. Search gateway — replaces standard requests to b_search_content with requests to Elasticsearch API. The gateway is implemented as a PHP proxy: it receives a request from the standard bitrix:search.page component, transforms it into Elasticsearch query DSL, and returns results in the format expected by Bitrix.
3. Event handlers — update the index when entities are modified or deleted.

Index structure for product catalog

The index is created via Elasticsearch Mapping API. Example mapping for products:

PUT /bitrix_catalog
{
  "mappings": {
    "properties": {
      "id":          { "type": "integer" },
      "iblock_id":   { "type": "integer" },
      "name":        { "type": "text", "analyzer": "russian" },
      "description": { "type": "text", "analyzer": "russian" },
      "sku":         { "type": "keyword" },
      "price":       { "type": "float" },
      "active":      { "type": "boolean" },
      "section_id":  { "type": "integer" },
      "properties":  { "type": "object" },
      "updated_at":  { "type": "date" }
    }
  },
  "settings": {
    "analysis": {
      "analyzer": {
        "russian": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": ["lowercase", "russian_stop", "russian_stemmer"]
        }
      },
      "filter": {
        "russian_stemmer": { "type": "stemmer", "language": "russian" },
        "russian_stop":    { "type": "stop", "stopwords": "_russian_" }
      }
    }
  }
}

The russian analyzer with stemmer is a key difference from MySQL FULLTEXT, which without additional dictionaries does not understand morphology.

PUT /bitrix_catalog/_settings
{
  "analysis": {
    "filter": {
      "russian_synonyms": {
        "type": "synonym",
        "synonyms": [
          "брюки, штаны, джинсы => trousers",
          "смартфон, телефон, мобила => mobile"
        ]
      }
    },
    "analyzer": {
      "russian_with_synonyms": {
        "tokenizer": "standard",
        "filter": ["lowercase", "russian_stop", "russian_stemmer", "russian_synonyms"]
      }
    }
  }
}

How to set up automatic index update?

Subscribe to infoblock events:

// local/php_interface/init.php
AddEventHandler('iblock', 'OnAfterIBlockElementUpdate', 'esUpdateProduct');
AddEventHandler('iblock', 'OnAfterIBlockElementDelete', 'esDeleteProduct');

function esUpdateProduct(array &$arFields): void
{
    $client = getEsClient();
    $productId = (int)$arFields['ID'];
    // Re-index a single document
    $client->index([
        'index' => 'bitrix_catalog',
        'id'    => $productId,
        'body'  => buildProductDocument($productId),
    ]);
}

function esDeleteProduct(int $productId): void
{
    getEsClient()->delete(['index' => 'bitrix_catalog', 'id' => $productId]);
}

The OnAfterIBlockElementUpdate event also triggers on API changes (1C import), which is important for index freshness.

Data indexing

Initial indexing is run via a cron script. Data is read in batches using CIBlockElement::GetList() with nTopCount = 100 and offset to avoid memory overload:

\Bitrix\Main\Loader::includeModule('iblock');

$client = \Elasticsearch\ClientBuilder::create()
    ->setHosts(['localhost:9200'])
    ->build();

$offset = 0;
$batchSize = 100;

do {
    $res = \CIBlockElement::GetList(
        [],
        ['IBLOCK_ID' => CATALOG_IBLOCK_ID, 'ACTIVE' => 'Y'],
        false,
        ['nTopCount' => $batchSize, 'nPageSize' => $batchSize, 'iNumPage' => floor($offset / $batchSize) + 1],
        ['ID', 'NAME', 'DETAIL_TEXT', 'IBLOCK_ID', 'IBLOCK_SECTION_ID']
    );

    $bulk = [];
    while ($item = $res->GetNext()) {
        $bulk[] = ['index' => ['_index' => 'bitrix_catalog', '_id' => $item['ID']]];
        $bulk[] = [
            'id'         => (int)$item['ID'],
            'iblock_id'  => (int)$item['IBLOCK_ID'],
            'name'       => $item['NAME'],
            'description'=> strip_tags($item['DETAIL_TEXT']),
            'section_id' => (int)$item['IBLOCK_SECTION_ID'],
            'active'     => true,
            'updated_at' => date('c'),
        ];
        $offset++;
    }

    if (!empty($bulk)) {
        $client->bulk(['body' => $bulk]);
    }
} while ($res->SelectedRowsCount() === $batchSize);

Bulk API allows sending up to 1000 documents per request. Do not use individual index requests for initial indexing — it is 10–50 times slower.

Search query

Replace the standard bitrix:search.page component with a custom one that queries Elasticsearch:

$response = $client->search([
    'index' => 'bitrix_catalog',
    'body'  => [
        'query' => [
            'multi_match' => [
                'query'     => $searchQuery,
                'fields'    => ['name^3', 'description', 'sku'],
                'type'      => 'best_fields',
                'fuzziness' => 'AUTO',
            ],
        ],
        'sort'  => ['_score' => ['order' => 'desc']],
        'from'  => ($page - 1) * $pageSize,
        'size'  => $pageSize,
    ],
]);

The fuzziness: AUTO parameter provides typo search: for words up to 5 characters, 1 substitution is allowed; for longer words, 2 substitutions.

What is included in our work?

• Audit of current search and load.
• Setting up Elasticsearch cluster (version, configuration, monitoring).
• Creating mapping according to your data structure.
• Developing indexer and search gateway.
• Configuring events for automatic update.
• Performance and accuracy testing.
• Documentation and training for your team.
• Support during the warranty period.
• Monitoring and alerts for indexing failures.

Implementation timeline

Contact us for a consultation. Get an accurate project estimate and architectural recommendations — with no obligation.

80% of Bitrix sites slow down due to one table

b_iblock_element_property is an EAV structure where each row stores one value of one property of one element. A catalog of 50,000 products with 30 properties yields 1.5 million rows. The smart filter performs a JOIN of this table with b_iblock_element on five properties, and MySQL performs a full table scan for 3–5 seconds. Our experience shows that without intervention in this table, site acceleration is impossible. We take on projects where load time has dropped to 8–10 seconds and bring TTFB back to <200 ms within 1–2 weeks. Site speed optimization begins with an audit of slow queries and ends with a comprehensive turnkey infrastructure overhaul.

Contact us for an audit — we will identify bottlenecks within 2 hours and propose a concrete plan.

How to achieve TTFB below 200 ms?

Server optimization is the first step. Nginx configuration goes beyond simple gzip. Specifically:

• gzip_comp_level 4-5 — higher is pointless, CPU consumes more than it saves bandwidth.
• brotli on with brotli_static on for precompressed files.
• HTTP/2 with http2_max_concurrent_streams 128.
• fastcgi_cache for PHP responses — caching at Nginx level, bypassing PHP-FPM entirely.
• worker_processes auto, worker_connections according to the number of simultaneous connections.

PHP-FPM tuning: choose between pm = dynamic and pm = static. Static mode works best for dedicated servers with predictable load because it avoids forking overhead. Dynamic saves RAM under low traffic. Calculate pm.max_children as (available RAM - RAM for MySQL/Redis) / average process consumption. For OPcache set memory_consumption=256, max_accelerated_files=20000, and validate_timestamps=0 in production (restart PHP-FPM on deploy).

MySQL/MariaDB: the main bottleneck is almost always the database. Enable slow_query_log with a threshold of 0.5 sec and analyze every query via EXPLAIN. Set innodb_buffer_pool_size to 70–80% of available RAM on a dedicated server. Create composite indexes for faceted search: (IBLOCK_ID, IBLOCK_PROPERTY_ID, VALUE) on b_iblock_element_property. Run OPTIMIZE TABLE b_iblock_element_property after mass operations.

How to configure three-level caching?

Managed component cache. Set TTL individually for each component. Catalog — 3600 sec, news feed — 300 sec, banners — 86400. The same TTL everywhere guarantees either outdated data or useless cache.

Composite cache. The bitrix:composite technology lets Nginx serve ready HTML from a file; PHP is not executed. Dynamic zones (cart, authorization) are loaded via AJAX request through CBitrixComponent::setFrameMode(true). TTFB drops below 50 ms. However, not all components are compatible; $APPLICATION->ShowPanel() and direct output via echo break the composite. We check every page through the panel 'Performance → Composite Site'. According to Bitrix official documentation on composite cache, this is the most effective caching method for high‑load projects.

Comparison: composite cache is 10–20 times faster than managed cache in time to first byte.

Memcached / Redis. Transfer cache from the file system: sessions go to Redis (session.save_handler = redis) — 10–50 times faster than files, plus cluster support. Component cache goes to Memcached via .settings.php: 'cache' => ['type' => 'memcache']. Also enable ORM query cache so identical GetList() calls don't hit MySQL on every request.

What is the fastest way to optimize Bitrix database?

Default MySQL settings are insufficient. Indexes — composite for faceted search, covering for frequent queries. MySQL responds from the index without accessing the data. Partial indexes (MariaDB) for filtering by ACTIVE = 'Y'. Audit unused indexes — each slows down INSERT/UPDATE.

Partitioning. For tables with millions of rows: b_stat_session, b_search_content_stem, and highload-blocks with history. Partition by date — a query for 'orders in a month' does not scan three years of data. Partitioning also solves the problem of concurrent queries during exchange with 1С via CommerceML.

Real case: a catalog of 200,000 products, 50 properties. Filtering by 10 properties took 12 seconds. After creating composite indexes on (IBLOCK_ID, IBLOCK_PROPERTY_ID, VALUE) and partitioning b_iblock_element_property by IBLOCK_ID, execution time dropped to 0.3 seconds. MySQL load decreased by 40 times.

Cleanup. Over a year or two, any database accumulates: outdated search index, expired records in b_cache_tag, history in b_iblock_element_prop_s*, logs in b_event_log taking gigabytes. We set up regular cleanup via agents.

Frontend and CDN

Images account for 60–80% of page weight. Convert to WebP via CFile::ResizeImageGet() with BX_RESIZE_IMAGE_PROPORTIONAL + conversion. Use srcset + sizes — never load a 3000px image into a 400px block. Add loading="lazy" for everything below the fold. AVIF offers another 20–30% savings vs WebP.

CSS/JS optimization: use the built-in Bitrix module to merge and minify via 'Settings → CSS/JS Optimization'. Apply PurgeCSS / UnCSS — in a typical Bitrix project, 60–70% of CSS is unused. Use defer / async for non‑critical JS and inline critical CSS in <head> for instant FCP.

Fonts: add <link rel="preload" as="font" crossorigin> for the main font. Set font-display: swap — text visible immediately. Subset via pyftsubset — keep only Cyrillic + Latin, file size reduces by 3–5 times.

CDN: Cloudflare, BunnyCDN, AWS CloudFront, or Russian providers (Selectel CDN, VK Cloud CDN). Serve static assets (CSS, JS, images, fonts) via CDN with Cache-Control: public, max-age=31536000, immutable for files with a hash. Use on‑the‑fly image optimization (imgproxy, Cloudflare Polish) without load on origin.

Why is load testing necessary?

Not synthetic benchmarks, but real scenarios: k6 / wrk to simulate routes — catalog → filtering → product card → cart → checkout. Measure RPS, response time (p50, p95, p99), error rate. Use Xdebug (callgrind) or Blackfire for PHP profiling to find bottlenecks. The test result gives an objective picture of where it actually slows down, not where it 'seems'. After optimization, run again to record improvements.

Results

What is included in the work?

1. Current performance audit — analysis of slow queries, PHP profiling, check of caching, CDN, server settings.
2. Server configuration — Nginx, PHP-FPM, MySQL, Redis/Memcached, OPcache.
3. Caching optimization — managed cache, composite site, TTL configuration, tagged caching.
4. Database work — index creation, partitioning, cleanup, EAV table reorganization.
5. Frontend — images (WebP/AVIF), CSS/JS (minification, deferred), fonts (preload, subsetting).
6. CDN — connection, caching rule setup.
7. Load testing — real user scenarios, metric report.
8. Documentation — description of all changes, recommendations for further maintenance.
9. Guarantee — support for 1 month after delivery, ensuring all optimizations are stable.

Monitoring

Without monitoring, everything degrades in six months. A new module, uncleared logs, a template change — and speed returns to original. Use web-vitals API for Real User Monitoring from actual visitors. Set up synthetic monitoring with Pingdom or UptimeRobot for regular checks from different locations. Configure alerts — TTFB > 500 ms or LCP > 3 sec triggers notification.

Timelines and cost

Cost is calculated individually after the audit. Get a consultation for your project — we will evaluate the current state and propose an acceleration plan with specific timelines and budget. We are a team with 12+ years of experience in Bitrix, having completed over 300 site speed optimization projects. Contact us to start the performance audit today.