Collect statistics at the following phases of table development:
- Newly created, unpopulated tables.
Collect statistics on unpopulated tables to set up the interval histogram used in internal processing. This initial collection makes subsequent statistics collections faster. Make sure to recollect statistics after data is added.
- Prototype phase, newly populated tables.
- Production phase, after changing ~10% rows of the table or partition. For high volumes of nonunique values, such as dates or timestamps, you may want to recollect at 7%.
Recommendation: Collect production phase statistics after you have created users and applied real world query loads to the database (up to about 3 months of querying).
- Collect statistics weeks after an upgrade or migration during periods of low CPU usage.