This post closes the study of Selinger et al. (1979), covering nested queries. Whether a subquery references columns from the outer query determines if it runs once or thousands of times.

This post continues the study of ‘Access Path Selection in a Relational Database Management System’ (Selinger et al., 1979), now covering chapter 5 and how the optimizer handles joins. PostgreSQL uses hash join most often, which the 1979 paper doesn’t describe.

This post continues the study of ‘Access Path Selection in a Relational Database Management System’ (Selinger et al., 1979), now covering chapter 4 and the cost formulas the optimizer uses to compare access paths for a single relation.

This post is a study dump. I’ve been reading ‘Access Path Selection in a Relational Database Management System’ (Selinger et al., 1979) and stopped at chapter 3 to consolidate what I learned.

A configuration flag changed. A boolean. The side effect opened a database connection pool before fork() and took down every Celery worker in production. This is an account of the investigation, the OS internals behind it, and a proposed solution still waiting for staging validation.