[jandrewrogers]

It is fairly common in e.g. multi-terabyte Postgres databases. Beyond a certain scale, it can become impossible for the statistics collector to build an accurate model for large tables. When this happens, you can end up in the strange place where the query plans change every time you run analyze even though the data model and data distribution hasn't changed. I've seen operational issues at multiple companies due to the query plans randomly changing like this on large tables in Postgres.