To SQL or Not To SQL: That Is Not the Question

The convergence of SQL and NoSQL combines the simplicity of JSON document development and the power of relational databases within a single database.

May 7th, 2025 11:00am by Hermann Baer

Featued image for: To SQL or Not To SQL: That Is Not the Question

Featured image by rangizzz on Shutterstock.

The ongoing debate between SQL and NoSQL is as old as relational database systems. Ever since Edgar F. Codd introduced the relational database concept in 1970, there has been an ongoing deep tension between structure and flexibility in database systems. SQL databases claim superiority through decades of proven reliability, strong ACID guarantees and powerful query capabilities ideal for complex, relational data models. In contrast, proponents of NoSQL databases argue that rigid schemas can’t keep pace with modern development, so they offer schema flexibility, simple application development and better performance for high-velocity or unstructured data. Each camp claims dominance — SQL for consistency and integrity, NoSQL for speed and agility — yet the real-world trend is convergence, as both camps increasingly work on borrowing features from the other.

SQL and NoSQL: Converging Into Multimodel Databases

Multimodel databases like MySQL, PostgreSQL and Oracle are at the forefront of integrating non-relational capabilities into traditionally relational systems. Recent innovations include vector processing, document store features and support for collections of JSON documents. These new capabilities allow developers to:

Work with JSON documents using familiar document APIs
Query and process both structured and semistructured data using SQL
Combine relational normalization with document-centric development

Multimodel Databases Support Native Document Store APIs

All these multimodel databases have introduced a binary format for storing JSON data natively and optimized. This includes Oracle’s OSON format, which has shown advantages over BSON in research for performance and efficiency of incremental updates. For developers seeking a document-first experience, vendors offer APIs that expose native JSON stores, such as MySQL’s X DevAPI, PostgreSQL-based FerretDB or Oracle’s Database API for MongoDB. These APIs provide the simplicity of NoSQL development on top of relational infrastructure. Developers enjoy the simplicity and agility of schema-flexible development without needing a special-purpose document store database. However, unlike SQL, document store APIs are not standardized and remain vendor specific. Efforts to define a common standard are underway but are still in the early stages.

Querying JSON With SQL

With native JSON support, developers can treat document collections as first-class citizens in the database. Using ANSI SQL/JSON (introduced in SQL:2016), you can extract values, unnest arrays, filter documents and apply SQL functions to JSON content. This bridges relational and semistructured data through a unified language, opening the door to benefit from the power of SQL for analytics and reporting. Here’s an example that queries a JSON collection of movies, extracts attributes, aggregates gross revenue by year and calculates each year’s revenue share:

WITH revenue AS (

   SELECT   

      m.data.year

      , round(sum(JSON_VALUE(data,'$.gross' RETURNING NUMBER NULL ON ERROR NULL ON EMPTY ))/1000000) as millions

   FROM movies m 

   WHERE m.data.gross IS NOT NULL 

   GROUP BY m.data.year

)

SELECT year

       , millions 

       , ROUND((RATIO_TO_REPORT(millions) OVER ())*100,2) pct_revenue

FROM revenue r

WHERE year > 2000

ORDER BY year DESC;

This example uses both the ANSI SQL/JSON operator JSON_VALUE and Oracle’s simplified dot notation, a straightforward SQL-style way to extract JSON scalar values from a JSON document (m.data.year). Developers extract and ”relationalize” attributes stored in JSON documents, join this information with relational tables, and use nested subqueries or other constructs — the same way they use any available SQL operator or function for relational objects.

Exposing Relational Data as JSON Documents

Conversely, relational data can be exposed as JSON documents using JSON collection views, making it accessible to document-centric applications. The following example creates a JSON collection view on top of Oracle’s EMP table, probably one of the first relational tables ever created for demonstration purposes:

CREATE OR REPLACE JSON COLLECTION VIEW emp_collection AS

   SELECT

      JSON{'_id' : empno,

           'employeName' : ename,

           'jobRole' : job}

     FROM emp;

Using this collection view through a document store API would return data similar to:

jason> db.emp_collection.findOne()

{ _id: 7369, employeName: 'SMITH', jobRole: 'CLERK' }

This bidirectional flexibility blurs the boundaries between SQL and NoSQL and decouples how data is processed from how it is stored.

Beyond Dual Access: JSON-Relational Duality

While native binary JSON storage, document APIs and SQL/JSON functionality represent strong progress, JSON-relational duality takes things a step further. This new capability offers the best of both relational and JSON documents without the trade-offs of either model. In short, duality views store the JSON documents internally in a highly efficient normalized format, using relational and JSON constructs. At the same time, developers interact with JSON documents.

Duality views: stored as rows, exposed as documents

Whether the application uses REST or document-centric APIs, developers benefit from the simplicity of retrieving and working with all the data needed for a single application-tier object: the JSON document. Creating a JSON duality view is pretty straightforward. Like a JSON collection view, developers define the Object Document Model for the business object as metadata in the database. Here’s a simple example that defines a duality view to model a conference schedule using GraphQL. The schedule object is comprised of:

The attendee information
The schedule for an individual attendee
The schedule representation for one or more sessions that an attendee is planning to attend
The detailed information about a session, including the speaker information

create or replace JSON Duality view ScheduleV 

AS 

attendee  

{ _id : aid name : aname 

  schedule : schedule @insert @update @delete 

       { scheduleId : schedule_id 

         sessions @unnest 

                   { sessionId : sid 

                     name : name 

                     location : room 

                     speaker @unnest 

                              { speakerId : sid 

                                speaker : name 

                                } 

                        }  

            } 

} ;

JSON-relational duality, therefore, provides the storage, consistency and efficiency benefits of the relational model while applications retrieve and manipulate deeply nested JSON structures.

The Best of Both Worlds

The convergence of SQL and NoSQL enables developers to benefit from the simplicity of JSON document development and the power of relational databases with a single database. Multimodel databases eliminate the need to choose one paradigm over another, taking over the single-purpose NoSQL database systems market. Developers don’t have to face the choice between SQL and NoSQL systems and lock themselves into one programming paradigm. They enjoy the flexibility and freedom to pick the best approach for individual applications without sacrificing functionality. Oracle’s JSON-relational duality views go beyond coexistence by fusing the strengths of relational and document models into a unified architecture. As multimodel systems continue to evolve, this approach sets a compelling precedent — and others are likely to follow. Begin innovating today with JSON-relational duality, either locally with Oracle Database 23ai Free or in the cloud on OCI with Oracle Autonomous Database.

Hermann is a senior director of product management in the Oracle Database organization. He and his team focus on Oracle's core functionality, such as Oracle Partitioning, the Oracle Optimizer, and analytical SQL. His team also looks after semi-structured data processing,...