Re: Vacuum ERRORs out considering freezing dead tuples from before OldestXmin

On Mon, Jun 24, 2024 at 4:51 PM Peter Geoghegan <pg(at)bowt(dot)ie> wrote:
>
> On Mon, Jun 24, 2024 at 4:36 PM Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> > I thought the idea was that the GlobalVisTest stuff would force a
> > recalculation now and then, but maybe it doesn't actually do that?
>
> It definitely can do that. Just not in a way that meaningfully
> increases the number of heap tuples that we can recognize as DEAD and
> remove. At least not currently.
>
> > Suppose process A begins a transaction, acquires an XID, and then goes
> > idle. Process B now begins a giant vacuum. At some point in the middle
> > of the vacuum, A ends the transaction. Are you saying that B's
> > GlobalVisTest never really notices that this has happened?
>
> That's my understanding, yes. That is, vistest is approximately the
> same thing as OldestXmin anyway. At least for now.

Exactly. Something has to cause this backend to update its view of the
horizon. At the end of index vacuuming,
GetOldestNonRemovableTransactionId() will explicitly
ComputeXidHorizons() which will update our backend's GlobalVisStates.
Otherwise, if our backend's RecentXmin is updated, by taking a new
snapshot, then we may update our GlobalVisStates. See
GlobalVisTestShouldUpdate() for the conditions under which we would
update our GlobalVisStates during the normal visibility checks
happening during pruning.

Vacuum used to open indexes after calculating horizons before starting
its first pass. This led to a recomputation of the horizon. But, in
master, there aren't many obvious places where such a thing would be
happening.

- Melanie