[Sinjo]

Spot on. We ran the migrations against production-sized data beforehand, and everything was fine. We did it again after the downtime, and got the same results.

Most of our queries are fast (tens of milliseconds or lower), which is how we got away without knowing about this for so long. Unsurprisingly, we've been making a bigger effort to eliminate any slow queries we do find lately. ;)

[spotman]

Curious, is the:

  SELECT DISTINCT(email) FROM parent;

The actual query you run in production, or similar enough? That could be a very expensive query with a very large amount of rows, even if indexed. (Large, being in the multi millions, depending on hardware). Might make more sense to run something like that on a reporting slave.

[Sinjo]

That was a fabricated example, rather than something plucked from our codebase. It's the sort of query we avoid running on our production database.

I agree with your approach though. Our data analysis runs elsewhere.