SQL Persists

Long ago Oracle once sold features it hadn't built yet, then implemented them afterward. A story from the dawn of databases.

Sybase became SQL Server. PostgreSQL chased Oracle but was overtaken by MySQL on performance. Now, a full cycle later, PostgreSQL is ascendant again. Fortunes shift. What's interesting is that SQL has been there the entire time.

Dialects exist. Every vendor ships its own extensions, and every migration makes you cry. Yet the core grammar holds. Parsing is complex. Optimization is harder. The antithesis to all that weight was the key-value store. Look up a key, return a value. Nothing more. It was clean.

KVS once served as the secret weapon for load distribution in the social-gaming boom. Today it sits quietly as a session store. The revolutionary became municipal infrastructure.

Around the pandemic, LINE reportedly captured a massive survey without touching application servers at all. They recorded HTTP GET access logs and used only the traces of inbound requests. Not processing requests — just noting that requests arrived. The essence of load distribution is to not process at all.

Technology evolves. The problem does not. How do you handle enormous volumes of data? SQL has answered that question for nearly fifty years. Antitheses rise and fall. SQL remains. I was reading sqlc output generated by AI and it struck me. Humans no longer even write the SQL. Code generates the code that generates it. Still, SQL does not change.