I wrote:
> So there's no longer any doubt that something is holding back OldestXmin.
> I will go put some instrumentation into the code that's computing that.
The something is the logical replication launcher. In the failing runs,
it is advertising xmin = 724 (the post-initdb NextXID) and continues to
do so well past the point where tenk1 gets vacuumed.
Discuss.
regards, tom lane