On Wed, Jun 16, 2021 at 1:21 PM Peter Geoghegan <pg@bowt.ie> wrote:
> Oh yeah, I think that I get it now. Tell me if this sounds right to you:
>
> It's not so much that HeapTupleSatisfiesVacuum() "disagrees" with
> heap_prune_satisfies_vacuum() in a way that actually matters to
> VACUUM. While there does seem to be a fairly mundane bug in
> GetOldestNonRemovableTransactionId() that really is a matter of
> disagreement between the two functions, the fundamental issue is
> deeper than that. The fundamental issue is that it's not okay to
> assume that the XID horizon won't go backwards. This probably matters
> for lots of reasons. The most obvious reason is that in theory it
> could cause lazy_scan_prune() to get stuck in about the same way as
> Justin reported, with the GetOldestNonRemovableTransactionId() bug.
Any update on this, Andres?
--
Peter Geoghegan