Explanation

An explanation gives the reason behind a design: what the engine does about the limits of object storage, and what that costs a deployment.

All pages

• Architecture

  Why storage is the source of truth, and what nodes are for.

• Generations

  Why an index is a name with generations under it, and why the name is what callers hold.

• Synchronization

  Manifests, epochs, and the leadership table; what keeps two writers from corrupting an index.

• What a write guarantees

  What a client can assume after a document write returns, when the change is durable and visible, and what a failover does to it.

• Separating search and indexing nodes

  Why searching and writing want different nodes, and what a deployment runs out of first as its index count grows.

• Node memory and JVM configuration

  Why the heap is sized against the page cache, and what each JVM flag is there for.

• Trust model

  What a key can reach, why it cannot be narrowed per end user, and why a browser must not hold one.

• How sub-documents are stored

  What a value of an `object` field costs to hold, change, and search.

• Lucene compatibility

  Why an index can outlive the code that can read it, and what the engine does about it.

• Text analysis

  Why one chain serves both sides of a search, where the two sides differ on purpose, and why a change to a chain needs a new generation.

• Where a setting lives

  Which of the index definition, the search settings, and the search request owns a setting, and how fast a change to each one reaches the nodes.

• Relevance

  The layers results are ordered by, what each one is for, and which of them a change reaches without reindexing.

• Signal fields

  Why a number a ranking reads is held apart from the document, what a refresh of it costs, and why refreshes go in large batches.