PromQL mistakes that give you wrong p95s and blank error rates | The Platform Radar
The p95 on your dashboard can be wrong, and the graph won't warn you. Learn how to tell whether the data or the query is at fault, before an incident depends on that number.
You will learn what a histogram bucket really records, how rate and quantile get computed, and where those numbers go wrong. By the end, you can tell whether a wrong p95 or a strange error rate comes from the data or the query. You will also know what to change on Monday in your paging and dashboard queries so the number holds up during an incident.
Chapters
00:00 Intro
01:20 Why the graph can't warn you
03:40 What a…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.