PostgreSQL Text Search Index Types (Short Overview)

PostgreSQL full-text search is implemented using three index types:

• GIN (Generalized Inverted Index)
• GiST (Generalized Search Tree)
• RUM (Real-time Universal Metric, extension)

These are the only index types that natively support tsvector / tsquery full-text search.

 1. GIN (Generalized Inverted Index)

Idea

Store a direct mapping from each word (lexeme) to all rows that contain it.

Main Logic

• Convert text → tsvector (set of lexemes)
• For each lexeme:
• store list of row IDs (posting list)
• Query:
• break into lexemes
• fetch row sets
• combine (AND / OR)

Result: exact matches, no recheck needed

2. GiST (Generalized Search Tree)

Idea

Store a compressed signature (fingerprint) of document content instead of exact words.

Main Logic

• Convert text → signature (bit representation of lexemes)
• Store signatures in tree structure
• Query:
• compare query signature with stored signatures
• return candidate rows
• recheck actual data

Result: approximate match → requires verification

3. RUM (Extension Index)

Idea

Extend inverted index by storing extra information (positions, ranking data) together with lexemes.

Main Logic

• Similar to GIN:
• lexeme → row IDs
• Additionally stores:
• positions of words
• frequency / ranking data
• Query:
• retrieve rows
• compute ranking directly from index
• return ordered results

Result: exact match + efficient ranking

Final Conclusion

• GIN → exact inverted index (standard choice)
• GiST → signature-based filtering (approximate)
• RUM → inverted index with ranking support

Core difference:

• GIN = precise lookup
• GiST = probabilistic filter
• RUM = lookup + ranking engine