Where do the performance cliffs come from?

When running queries on a Postgres database, chances are you have
experience with hitting a "performance cliff" - an abrupt change of
performance for a given Postgres query, after only a small change in
data and/or query parameters.

In this talk, you'll learn why performance cliffs happen in Postgres.
After a brief explanation of how cost-based optimizers work and what
assumptions that relies on, I'll show a couple examples why/how it
fails. I'll also share recommendations on how to mitigate these issues.

If time permits, I'll talk about a couple interesting improvements &
possible solutions proposed in recent research projects, focused on
making Postgres query execution more adaptive & robust.