Expand description
The SurrealDB index engines.
Each engine here owns one index structure: the full-text index and its
analyzer, the HNSW and DiskANN vector graphs, the count index, the
unique/non-unique entry index, and the document-id mapping they share. An
engine is reached downward — through a transaction and an env::IndexEnv
handle — and answers from the keyspace alone.
The pure computations those engines and the executor share — index key
encoding and decoding, full-text term resolution and scoring, vector
scoring — are kernels: functions over values and bytes that touch no
storage, named in a registry. An executor that does its own reads hands
what it read to them.
Choosing an index for a query, and iterating one on behalf of a statement, is not here: that is the planner’s job, and it lives above this crate with the executor that drives it.
This crate is an internal implementation detail of SurrealDB. Its API is unstable and changes without notice.
Re-exports§
pub use surrealdb_catalog as catalog;
Modules§
- adjacency
- Packed graph adjacency: the merged reader over blocks and per-edge keys.
- config
- The index and graph-adjacency engines’ configurable resource limits.
- count
- docids
- Table-level document-ID (surrogate) space shared by all indexes on a table.
- entry
- The value stored under a b-tree index entry.
- env
- The environment the index engines run against.
- expr
- The type definitions for the computation format of the surreaql executor.
- file
- The allowlist every file an index engine opens has to resolve inside.
- ft
- index
- This module applies index mutations for a single document across different index types (UNIQUE, regular, search, fulltext, Hnsw). Index keys are constructed via key::index and field values are encoded using key::value::Array.
- inline_
cache - Inline sidecar caches: one key holding a record’s complete adjacency in
one direction (
!ce), or its complete incoming reference set (!cr). - kernels
- Kernels: the pure computations an index read or write is made of.
- places
- The node’s worker places, as the index engines take them.
- spatial
- Hierarchical Hilbert-curve spatial cell decomposition and region covering over the WGS84
(lon, lat)plane. - trees
- val
Structs§
- Index
KeyBase - Previous
BpSpelling - A previous release’s primary-appending keys for one build generation, with the generation’s prefix encoded once for however many records.
- Previous
Build Primary Key - A primary-appending marker’s key as a previous release spelled it: the generation’s prefix, then the record id under the index format.
Enums§
- Compaction
Apply - What applying one compaction plan did.
- Error
- A failure in the index layer.
Functions§
- check_
index_ kernel - Fails when this build left out the kernel an index of kind
indexneeds. - index_
exists_ record - The record a unique-index conflict names, if
erroris one. - wipe_
index_ data - Clears everything an index has stored, and advances its compaction generation so that starting from empty does not unfence a compactor.
- wipe_
index_ page - Deletes, in
tx, at mostlimitof the keys the indexikbnames holds, other than the compaction generationswipe_index_datareads and writes back, and answers how many it deleted.