Well, one of the principal arguments for having VACUUM at all is that it
off-loads required maintenance effort from foreground transaction code
paths. I'm not really going to be in favor of solutions that put more
work into the transaction code paths (HOT already did more of that than
I would like :-(). OTOH, I agree that scanning the WAL log doesn't
really sound like something well-matched to this problem either.
Tom, Don't you like the idea of building some more structures around WAL, like Asynchronous Materialized views. Indexes, if implemented as stated, would remove the HOT code in the path of the transaction(as you may know). I am also slightly doubtful of the argument, that doing full-table scans and full index scans for Vacuum is efficient. Can you please advise me on why we should not use a read only operation on WAL log ?
Thanks,
Gokul.
