How does pgvector handle NULL values in vector columns?

-- Create table allowing NULLs (default behaviour)
CREATE TABLE documents (
    id        BIGSERIAL PRIMARY KEY,
    content   TEXT,
    embedding VECTOR(1536)   -- nullable by default
);

-- Insert without embedding (embedding is NULL)
INSERT INTO documents (content) VALUES ('Article not yet embedded');

-- NULL rows are automatically excluded from similarity queries
-- (they cannot have a meaningful distance)
SELECT id, content FROM documents
ORDER BY embedding <-> '[0.1,0.2,...]'  -- NULLs won't appear in results
LIMIT 5;

-- Find rows missing embeddings (need to be processed):
SELECT id, content FROM documents WHERE embedding IS NULL;

-- Count unembedded documents:
SELECT COUNT(*) FROM documents WHERE embedding IS NULL;

-- Prevent NULLs if all rows must have embeddings:
CREATE TABLE documents_required (
    id        BIGSERIAL PRIMARY KEY,
    content   TEXT,
    embedding VECTOR(1536) NOT NULL  -- enforce non-null
);

-- Partial index to cover only non-NULL rows
-- (useful for variable-dimension or partially-embedded tables):
CREATE INDEX ON documents USING hnsw (embedding vector_cosine_ops)
WHERE embedding IS NOT NULL;