Skip to the content.

Node.js with SQL Server

Yes, you can use SQL Server with Node.js. I added a Gist to show how it could work. I didn’t get to connection pooling or transactions, just enough to get a general feel.

I’m not convinced that SQL (or an RDBMS in general) is the best persistent store for Node.

Some specific observations:

  • Support for Node with SQL Server seems immature and fragmented. There are many different drivers to choose from, no ADO.NET- or JDBC-like standard interface to wrap all those drivers. So if you run into issues with a driver and need to change to a different one, you may be rewriting code. I ended up picking node-sqlserver-unofficial, which is a precompiled binary version of Microsoft’s msnodesql driver, to get support for Windows integrated/trusted authentication, i.e., single sign-on to DB without a username and password in connect string. Yet, Microsoft’s says their driver is “not production ready” and doesn’t look like it can be cleanly installed via npm.

Tedious looks like a better solution if you can live without Windows integrated authentication. It is pure Javascript and appears to be more stable and widely supported.

  • Async callback hell. Sequential operations, like inserting a set of rows in order, become painful when each individual operation is asynchronous. I tamed this by replacing callbacks with promises (Promise.denodify is helpful), but it’s still harder to read and single-step than a corresponding synchronous version. This is a bigger issue with an RDBMS than with a document store like MongoDB or a key-value store like Redis – in a relational model, an object that contains an array turns into multiple rows that need to be inserted in order (insert parent first, then child), while in a document or key-value store, the same object would be persisted in a single operation.

With these observations in mind, it seems like SQL Server isn’t the best option. It can work, especially if you can use Tedious and live without integrated authentication. PostgreSQL support seems more promising and production-stable, plus it has native JSON support. With native JSON, you can slam a composite object into a single column and query its components, without having to deal with parent-child relationships.

Written on October 1, 2015