Outliers can trick the optimizer into the wrong plan

[nalgeon]

https://hakibenita.com/postgresql-correlation-brin-multi-minmax

[tedunangst]

> The example in this article is contrived - it was engineered to inject an outlier in every range so that the single range BRIN would perform as worst as possible. However, this article was motivated by a similar scenario we reached naturally in our systems.

I was kinda skeptical that you'd ever have so many outliers as to blow up the whole index, but I guess their scenario is plausible. Still curious what their reorder rate is.

Maybe if you wanted to make the single index work, you'd treat the table as an append log, and insert duplicate entries instead of trying to update.

[jrm4]

I confess I don't much understand the code here, but this seems the exact point of Nassim Taleb's "The Black Swan?"

[vore → jrm4]

When did Taleb talk about in depth techniques to optimize indexing in Postgres?

[tedunangst → vore]

If you take a step back and cross your your eyes, you'll see that Taleb has been correct about predicting everything.

[mhh__ → jrm4]

Not quite but it does probably rhyme.

[cosmojg → jrm4]

Perhaps! Properly dealing with outliers is important in many fields, and I can see the connection, if only just barely.