> (Also, I would push both things as a single commit.) Fine with me. I kept the test separate only because committers sometimes push the fix without the test.
Taking the full lock upfront brings back the deadlock Manu found, where a transaction the worker waits for runs REINDEX on the TOAST relation. That is the same kind of case as DDL on the table itself, so I'm fine with it. Thanks, Shihao
