The store module allows you to control how index data is stored.

The index can either be stored in-memory (no persistence) or on-disk (the default). In-memory indices provide better performance at the cost of limiting the index size to the amount of available physical memory.

When using a local gateway (the default), file system storage with no in memory storage is required to maintain index consistency. This is required since the local gateway constructs its state from the local index state of each node.

Another important aspect of memory based storage is the fact that ElasticSearch supports storing the index in memory outside of the JVM heap space using the "Memory" (see below) storage type. It translates to the fact that there is no need for extra large JVM heaps (with their own consequences) for storing the index in memory.

Store Level Compressionedit

From version 0.90 onwards, store compression is always enabled.

For versions 0.19.5 to 0.20:

In the mapping, one can configure the _source field to be compressed. The problem with it is the fact that small documents don’t end up compressing well, as several documents compressed in a single compression "block" will provide a considerable better compression ratio. This version introduces the ability to compress stored fields using the index.store.compress.stored setting, as well as term vector using the index.store.compress.tv setting.

The settings can be set on the index level, and are dynamic, allowing to change them using the index update settings API. elasticsearch can handle mixed stored / non stored cases. This allows, for example, to enable compression at a later stage in the index lifecycle, and optimize the index to make use of it (generating new segments that use compression).

Best compression, compared to _source level compression, will mainly happen when indexing smaller documents (less than 64k). The price on the other hand is the fact that for each doc returned, a block will need to be decompressed (its fast though) in order to extract the document data.

Store Level Throttlingedit

(0.19.5 and above).

The way Lucene, the IR library elasticsearch uses under the covers, works is by creating immutable segments (up to deletes) and constantly merging them (the merge policy settings allow to control how those merges happen). The merge process happens in an asynchronous manner without affecting the indexing / search speed. The problem though, especially on systems with low IO, is that the merge process can be expensive and affect search / index operation simply by the fact that the box is now taxed with more IO happening.

The store module allows to have throttling configured for merges (or all) either on the node level, or on the index level. The node level throttling will make sure that out of all the shards allocated on that node, the merge process won’t pass the specific setting bytes per second. It can be set by setting indices.store.throttle.type to merge, and setting indices.store.throttle.max_bytes_per_sec to something like 5mb. The node level settings can be changed dynamically using the cluster update settings API. Since 0.90.1 the default is set to 20mb with type merge.

If specific index level configuration is needed, regardless of the node level settings, it can be set as well using the index.store.throttle.type, and index.store.throttle.max_bytes_per_sec. The default value for the type is node, meaning it will throttle based on the node level settings and participate in the global throttling happening. Both settings can be set using the index update settings API dynamically.

The following sections lists all the different storage types supported.

File Systemedit

File system based storage is the default storage used. There are different implementations or storage types. The best one for the operating environment will be automatically chosen: mmapfs on Solaris/Windows 64bit, simplefs on Windows 32bit, and niofs for the rest.

The following are the different file system based storage types:

Simple FSedit

The simplefs type is a straightforward implementation of file system storage (maps to Lucene SimpleFsDirectory) using a random access file. This implementation has poor concurrent performance (multiple threads will bottleneck). It is usually better to use the niofs when you need index persistence.

NIO FSedit

The niofs type stores the shard index on the file system (maps to Lucene NIOFSDirectory) using NIO. It allows multiple threads to read from the same file concurrently. It is not recommended on Windows because of a bug in the SUN Java implementation.

MMap FSedit

The mmapfs type stores the shard index on the file system (maps to Lucene MMapDirectory) by mapping a file into memory (mmap). Memory mapping uses up a portion of the virtual memory address space in your process equal to the size of the file being mapped. Before using this class, be sure your have plenty of virtual address space.

Memoryedit

The memory type stores the index in main memory with the following configuration options:

There are also node level settings that control the caching of buffers (important when using direct buffers):