Write operation is to insert new documents into the Elasticsearch. Changes are only written to disks during a Lucene Commit. Lucene commit is an expensive operation so can not be performed after each create or delete operation.

So let's see then how Elasticsearch performs write operation 💥

Write operations are not directly performed on disks but are recorded in Translog which is an in-memory record for create and delete operations.

• Transactions are recorded in Translog
• New documents are added to the memory buffer, which is unindexed in-memory storage.
• A write operation is called successful only if it is recorded in primary translog and all replica translogs in other shards based on Quorum.

Refresh Operation

• During this, the memory buffer content is copied to a newly created segment in the memory and translog is cleared.
• And the newly created segment becomes available for search and indexing. -Refresh operation is executed in every second by default.

Flush Operation

This operation performs writing documents to disk. All new created in-memory segments and contents of memory buffer are merged to a new Lucene Segment and are committed to disk and translog gets empty. The main advantages of having a single merged segment are:

• Discard older version documents and reduce index space on disk.
• The old segments are removed and bigger segments are created which increases the search speed.

Conclusion

Disk write operations are expensive and Elasticsearch is designed to handle a large number of write requests in a production server with minimal latency and large throughput. Write operation in elasticsearch works on the write-ahead mechanism like in SQL databases. To know more about this topic please see 👉 Reading and Writing Documents