Fixing Tables

追加: 2026-05-17

[00:00:00]Hi Brent. After you've addressed a performance emergency, are there cases where you recommend table structure changes as a better long-term resolution? If so, what are some examples? Oh, um I'm going to give you the the classic one because normally I can't talk about some client stuff work, especially you can imagine if somebody hears and goes, "Oh my god, those were our tables." Um but there's one that I can talk about publicly and it always is such a good example is Stack Overflow.

[00:00:27]uh at Stack Overflow that you could tag a question with up to five tags CQL server cursor you know you can apply up to five different tags in the beginning those tags were stored in a single column on the post table which is where questions are stored. So if you wanted to find out if a column was tagged with say C sharp, uh you had to scan that whole entire column because it was just one column that had up to five tags in it and the column could be anywhere in that string. Well, SQL Server is notoriously inefficient CPU wise at parsing strings like that. So, not only did it have to shred all the string, it also had to shred through all of the rows. Uh, because you couldn't just jump to the C row. After all, you could be tagged alphabetical or something else like that. Um, so that was a case where after several years, the CPU usage was so bad that they switched out to normalizing that. That's a great example.