Penyimpanan Tembolok dan Konkurensi

Welcome to Caching and Concurrency! In this lesson, we'll explore how Elasticsearch uses caching to speed up searches and how it manages many requests at once.

Caching is like remembering past answers. If you ask the same question repeatedly, it's faster to recall the answer than to figure it out every time.

For search engines, performance is key. Without caching, every query, even identical ones, would require Elasticsearch to re-read data from disk and re-process it.

This leads to higher CPU usage, increased I/O operations, and slower response times. Caching helps reduce this overhead significantly.

Elasticsearch uses several caches. One important one is the Node Query Cache. This cache stores the results of frequently used filter queries.

It operates at the node level and is great for speeding up queries that use common filters, like "status": "active", without re-evaluating them.

The Node Query Cache stores the *bitsets* representing which documents match a filter. When a filter is used again, Elasticsearch can quickly retrieve this bitset instead of scanning all documents.

It's optimized for queries that are small, frequently run, and don't involve complex aggregations or full-text analysis.

Another vital cache is the Request Cache. This cache stores the *entire JSON response* of a search request for a specific shard.

It's useful for queries that are identical, including their aggregations, and are run often. It offers a significant speed boost by returning the pre-computed response.

Historically, Elasticsearch used a 'Field Data Cache' for sorting and aggregations on text fields. This could consume a lot of memory.

Today, Elasticsearch uses Doc Values by default for numeric, boolean, date, IP, and keyword fields. Doc Values are stored on disk in a column-oriented fashion, making them very efficient for aggregations and sorting without heavy memory use.

Caches are great, but they must be up-to-date. When you index, update, or delete a document in an index, Elasticsearch automatically invalidates (clears) the relevant cached entries for that shard.

This ensures that new searches always reflect the latest data, preventing stale results from being served.

Beyond caching, Elasticsearch needs to handle many users querying and indexing data simultaneously. This is called concurrency.

Elasticsearch achieves concurrency by using multiple threads and thread pools, allowing it to process several operations at the same time without waiting for each one to finish sequentially.

Elasticsearch organizes tasks using different thread pools. Each pool handles a specific type of operation:

• Search pool: For executing search queries.
• Index pool: For indexing and updating documents.
• Bulk pool: For handling bulk indexing requests.

These pools prevent one slow operation from blocking others.

When a thread pool is busy, incoming requests are placed into a queue. If the queue becomes full, Elasticsearch will start rejecting new requests for that operation type.

Rejected requests result in an error (e.g., HTTP 429 Too Many Requests). This mechanism is crucial for preventing the cluster from becoming overloaded and unstable.

In this lesson, we explored how Elasticsearch optimizes performance through caching and concurrency:

• Caching: The Node Query Cache and Request Cache store query results to avoid re-computation.
• Doc Values: An efficient, disk-based structure for aggregations and sorting.
• Concurrency: Elasticsearch uses thread pools to manage many simultaneous requests for search, indexing, and bulk operations.
• Queues: Requests are queued when busy, with rejection as a safeguard against overload.

Understanding these mechanisms helps you build faster and more resilient search applications!