[kroolik]

The locking part is often forgotten when discussing zero downtime migrations in postgres. As more and more operations become fast, more engineers will get bitten by this.

Ive telling this to engineers in my org for such a long time, now I also have a nice article to share :)

BTW, you don't need the pg_sleep and hurry with the execution. Simple BEGIN and COMMIT/ROLLBACK will give you infinite wait time :)