On 24.10.2013 13:02, Pavan Deolasee wrote:
> Another difference AFAICS is that checksum feature needs the block to be
> backed up only after the first time a hint bit is updated after checkpoint.
> But for something like pg_rewind to work, we will need to WAL log every
> hint bit update on a page. So we would want to keep it as short as possible.
To fix that, pg_rewind could always start the rewinding process from the
last checkpoint before the point that the histories diverge, instead of
the exact point of divergence. That would make the rewinding more
expensive as it needs to read through a lot more WAL, but I think it
would still be OK.
- Heikki