On Sun, Dec 24, 2023 at 12:00 PM Murthy Nunna <mnunna@fnal.gov> wrote:
Thanks, Tom.
I am wondering if there is way to obtain same performance as before. Like restoring performance statistics?! Simple analyze is not helping. And Vacuum is a overkill where there are no dead tuples.
I am not setting any hint bits... I can verify this in new and old if you can provide a query or some method to verify this... I still have original copy of same database on a different server for comparison.
Murthy Nunna <mnunna@fnal.gov> writes: > I did pg_dump of a ~20TB database followed by pg_restore. I find simple queries like select count(*) running slow. I did a select count(*) on all tables before pg_dump which took ~4 hours. After pg_restore, same thing took 32 hours.
My bet is that that was setting commit hint bits, and hence incurring a lot of writes. If the data is reasonably stable that's a one-time expense.