We've had outages not just due to a small increase in table size causing 10000x slowdowns when the query planner crosses a threshold, but also similar outages when a small decrease in table size did the same thing! Who would have thought that deleting old/unused data would make your website go down due to 10000x slower database queries?
Performance is a feature, and unpredictable 10000x slowdowns when the database decides to suddenly start doing billion-row table scans in production is just as bad as any other system crash
About a decade ago I interviewed with a (at the time) hot new startup, and during process they told me "we don't do joins here". I thought that was crazy, but here I am a decade later sometimes telling my colleagues "we can't do joins here" as we break up queries, paying the added round trips to help the query planner do the right thing. Oh how the turn tables.