GIN (Generalized Inverted Index) — Full Description

Idea

GIN is based on the concept of an inverted index.

Instead of storing rows → values, it stores:
value (lexeme) → rows that contain it

This makes it extremely efficient for searching text, where the goal is to find which rows contain specific words.

Main Logic

1. Text Transformation

• Input text is converted into a tsvector
• This process:
• tokenizes text into words
• normalizes them (stemming, lowercasing)
• removes stop words

2. Index Structure

For each lexeme, GIN stores:

• Posting List → list of row identifiers (TIDs)
• If large → converted into a Posting Tree

Structure conceptually:

3. Search Execution

• Query is converted into tsquery (lexemes + operators)
• For each lexeme:
  • retrieve matching row sets
• Combine results:
  • AND → intersection
  • OR → union

Result: exact matching rows

4. No Recheck Needed

Unlike GiST:

• GIN returns only valid matches
• No additional verification step required

Key Characteristics

✔ Strengths

• Very fast for full-text search
• Scales well with large datasets
• Precise results (no false positives)
• Efficient for multi-term queries

Weaknesses

• Slower inserts/updates (index maintenance overhead)
• Larger index size
• Uses pending list (delayed indexing for performance)

Conceptual Summary

GIN works like a reverse dictionary:

• Instead of asking “what words are in this row?”
• It answers “which rows contain this word?”

Final Insight

GIN is the default and most important index for PostgreSQL full-text search.

If your goal is:

• accurate search
• large-scale text querying
• predictable performance