Simon Riggs <simon@2ndquadrant.com> writes:
> All of those commands produce WAL for critical changes to data blocks,
> so those are passed through correctly.
> Hint bits are not set following recovery, so a full database VACUUM may
> help later performance, depends upon your access patterns.
That's only partially true. Full-page images in WAL will propagate hint
bits, so in practice you could expect that only about one checkpoint
cycle's worth of hint information is missing. If you'd been running
with full_page_writes = off then this doesn't hold, and in that case
I'd agree that a forced vacuum might be worth the trouble.
regards, tom lane