Concepts

Collection

A named group of documents sharing one schema — roughly equivalent to a table or an index in other systems. See Collections.

Document

A single searchable record: a flat JSON object. Every document has an id, implicit and never declared as a field. Field names starting with _ are reserved. See Documents.

Field

A named, typed value on a document, declared on the collection's schema — text, keyword, int, float, bool, or date. A field is only filterable, sortable, or facetable if the schema says so explicitly; text is always searched.

Inverted index

The data structure that maps tokens to the documents containing them, built as documents are added — what makes full-text search fast, because a query looks up a short list of matching documents instead of scanning every one. See Inverted Index for how Tachyon's is laid out on disk.

Token

The unit the tokenizer produces from text: roughly, a run of alphanumeric characters, lowercased and Unicode-normalized. Tachyon does not stem words or strip stop words — "running" and "run" are different tokens. See Tokenization for the exact rules.

Ranking

Search results are scored with BM25, a standard lexical relevance function that rewards term frequency in a document while discounting terms that are common across the whole collection. See Relevance & BM25.

Block-max WAND

The query execution strategy that skips postings blocks provably incapable of ranking in the top results, without changing which documents end up there. See Block-max WAND.

Filter

A boolean expression over fields declared filter: true (or facet/sort) that narrows a search to matching documents without affecting ranking. See Filtering.

Facet

A count of how many matching documents hold each value of a field declared facet: true — the data behind a "Brand (12) / Category (8)" style sidebar. See Faceting.

Write-ahead log

The append-only, checksummed log every write hits before it's acknowledged, replayed on startup to recover anything not yet flushed to a segment. See Write-Ahead Log.

Segment

An immutable, memory-mapped chunk of a collection's index on disk. New documents accumulate in memory (a memtable) and are periodically flushed into a new segment; segments are later folded together by a background merge. See Segment Merging and Persistence.