SQL: Schemeless Data Storage in PostgreSQL

schemeless.sql

/*
 * Approach: Schemeless Data Storage in PostgreSQL
 *
 * This implementation utilizes the JSONB data type to achieve a schemeless (document-oriented) 
 * architecture within a relational database.
 *
 * Advantages:
 * - Schema Flexibility: Allows storing diverse data structures without migrations.
 * - Development Speed: Faster iteration during early-stage development.
 * - Dynamic Attributes: Easily handle entities with unpredictable or sparse fields.
 *
 * Disadvantages:
 * - Validation Overhead: Requires explicit CHECK constraints or triggers to enforce data types.
 * - Query Complexity: Accessing nested fields requires specific JSON operators.
 * - Performance: Slightly higher CPU/storage cost compared to strictly typed columns.
 * - Tooling: Some ORMs and BI tools have limited support for deep JSON structures.
 */

CREATE TABLE IF NOT EXISTS domain_data (
    id UUID PRIMARY KEY DEFAULT uuidv7(),
    is_active BOOLEAN DEFAULT TRUE, -- For logical deletion.
    created_at TIMESTAMPTZ DEFAULT now(),
    updated_at TIMESTAMPTZ DEFAULT now(),
    entity VARCHAR(25) NOT NULL,
    medadata JSONB NOT NULL
);

-- Option 1: Add checks directly.
ALTER TABLE domain_data ADD CONSTRAINT check_metadata_structure CHECK (
    (entity = 'user' AND medadata ? 'email' AND medadata ? 'username') OR
    (entity = 'product' AND medadata ? 'sku' AND medadata ? 'price') OR
    (entity NOT IN ('user', 'product')) -- Allow others without validation
);

ALTER TABLE domain_data ADD CONSTRAINT check_metadata_types CHECK (
    CASE 
        WHEN entity = 'product' THEN 
            jsonb_typeof(medadata -> 'price') = 'number' AND
            jsonb_typeof(medadata -> 'sku') = 'string'
        ELSE TRUE
    END
);

-- Option 2: Use a function.
CREATE OR REPLACE FUNCTION validate_entity_metadata(ent VARCHAR, data JSONB) 
RETURNS BOOLEAN AS $$
BEGIN
    IF ent = 'user' THEN
        RETURN (data ? 'email') AND (data ->> 'email' ~* '^[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}$');
    ELSIF ent = 'product' THEN
        RETURN (data ? 'sku') AND (data -> 'price' != 'null');
    END IF;
    RETURN TRUE; -- Unknown entities pass by default
END;
$$ LANGUAGE plpgsql;

ALTER TABLE domain_data ADD CONSTRAINT constraint_entity_validation 
CHECK (validate_entity_metadata(entity, medadata));

-- This index will only take up space for users, ignoring products.
CREATE UNIQUE INDEX idx_unique_user_email 
ON domain_data ((medadata->>'email')) 
WHERE (entity = 'user');

-- Option 3: partitions.
-- pg_partman extension. It automatically creates the "next month" partition so you never run out of space to insert.
CREATE TABLE IF NOT EXISTS domain_data (
    id UUID PRIMARY KEY DEFAULT uuidv7(),
    is_active BOOLEAN DEFAULT TRUE, -- For logical deletion.
    created_at TIMESTAMPTZ DEFAULT now(),
    updated_at TIMESTAMPTZ DEFAULT now(),
    entity VARCHAR(25) NOT NULL,
    medadata JSONB NOT NULL,
    -- The partition key must be part of the Primary Key in Postgres
    PRIMARY KEY (id, created_at) 
) PARTITION BY RANGE (created_at);

-- Partition for January 2026
CREATE TABLE domain_data_2026_01 PARTITION OF domain_data
    FOR VALUES FROM ('2026-01-01') TO ('2026-02-01');

-- Partition for February 2026
CREATE TABLE domain_data_2026_02 PARTITION OF domain_data
    FOR VALUES FROM ('2026-02-01') TO ('2026-03-01');

-- 1. GIN index for general searches within any JSONB field
CREATE INDEX idx_metadata_gin ON domain_data USING GIN (metadata);

-- 2. B-Tree index for the entity_type column (useful for fast filters)
CREATE INDEX idx_entity_type ON domain_data (entity_type);

-- 3. Expression Index (Example: if searching frequently by product 'sku')
CREATE INDEX idx_metadata_sku ON domain_data ((metadata->>'sku')) 
WHERE (entity_type = 'product');