What is 'rank error' of aggregate functions?

[linter-purgatory]

We are looking at using approx_percentile to compute some statistics. I know that underneath those functions are implemented with the q-digest data structure and have read some resources that attempt to characterize it, but I still don't have a very good understanding of what 'rank error' is without going too into the weeds/implementation.

Could anyone help me understand what 'rank error' is, or point me to something that explains it?

Appreciate the help!

[rongrong]

Can you share a failure stack trace?

[linter-purgatory]

It isn't a crash, it's understanding what 'rank error' means in the context of approx_percentile. That docs for that function says:

As approx_percentile(x, percentage), but with a maximum rank error of accuracy. The value of accuracy must be between zero and one (exclusive) and must be constant for all input rows. Note that a lower “accuracy” is really a lower error threshold, and thus more accurate. The default accuracy is 0.01.

I'm trying to understand what they mean by 'rank error'