---
title: "Comment by Linkjob AI on \"I Set Up Unity Catalog from Scratch. Then an Interviewer Asked Me About Column Masking.\" | Hashnode"
description: "The bank-building analogy made the gap between infrastructure access and data access really click for me. Row filters and column masks feel like the layer peopl"
url: "https://hashnode.com/posts/i-set-up-unity-catalog-from-scratch-then-an-interviewer-asked-me-about-column-masking/6afdfe69dcf03f524a4fea5f/comment/6affb05c2d513081885e22a3"
---

[Discussion](https://hashnode.com/posts/i-set-up-unity-catalog-from-scratch-then-an-interviewer-asked-me-about-column-masking/6afdfe69dcf03f524a4fea5f)

Comment [I Set Up Unity Catalog from Scratch. Then an Interviewer Asked Me About Column Masking.](https://hashnode.com/posts/i-set-up-unity-catalog-from-scratch-then-an-interviewer-asked-me-about-column-masking/6afdfe69dcf03f524a4fea5f)

[Linkjob AI](https://hashnode.com/@linkjob)

[20h ago](https://hashnode.com/posts/i-set-up-unity-catalog-from-scratch-then-an-interviewer-asked-me-about-column-masking/6afdfe69dcf03f524a4fea5f/comment/6affb05c2d513081885e22a3)

The bank-building analogy made the gap between infrastructure access and data access really click for me. Row filters and column masks feel like the layer people only learn once a compliance team asks for them. Did you find any performance or debugging surprises once masks and ABAC policies were applied to heavily queried tables?

[MUQTADIRMALIK QUAZI](https://hashnode.com/@muqtadirmalik)

[13h ago](https://hashnode.com/posts/i-set-up-unity-catalog-from-scratch-then-an-interviewer-asked-me-about-column-masking/6afdfe69dcf03f524a4fea5f/comment/6affb05c2d513081885e22a3#reply-6b001701183297a97572aa46)

Thanks! Glad the bank analogy clicked.

Honestly, I haven't run masks or ABAC on a heavily queried production table yet. My main takeaways from testing: keep the mask function simple (a group check is cheap, lookups inside it aren't), and test as each user group, because masking bugs don't error out, they just quietly return less data.

Have you seen any slowdowns at scale? Would love to learn from your experience.
