System software plays a foundational role for modern society. The reliability of systems impacts various downstream applications including medicine and finance, and critical bugs in them can lead to disastrous consequences (e.g., deaths and financial crisis). The goal of my research is to advance the science and practice of building reliable and secure systems. In this talk, I will present my work toward this goal on improving the reliability of database management systems (DBMSs), one type of widely deployed foundational systems. At both the methodological and technical levels, my research has been guided by the unifying principle that I call granularity shifting. In particular, I show how granularity shifting led to innovative and effective solutions for three core challenges: (1) validating transaction consistency, (2) checking query result correctness, and (3) generating high-quality SQL queries. Impact-wise, my work has collectively exposed hundreds of diverse bugs in production DBMSs such as MySQL, PostgreSQL, and SQLite. I will conclude the talk by discussing my vision and quest toward reliable and secure systems as well as their applications in critical domains.