Hi! Would it make sense to have an engine-level option to tell the optimizer to prioritize index use at all times? Something along the line of HA_PRIORITIZE_INDEX.
I'm saying this because the relationship between Cursor::records_in_range() and HA_READ_RANGE seems important to effectively perform range scans at the moment. I tried purposely returning a constant small cost factor in records_in_range() but it seems that the range search will fail if the number of estimated rows (in my case a total lie) is significantly smaller than the number of actual rows in the table. If I don't set, HA_READ_RANGE then it seems Drizzle will resort to a table scan (despite knowing that 'possible_keys' exist in EXPLAIN). Preferably, I don't want to count through the number of rows in a range for the optimizer on every simple query. Although... a completely dishonest implementation of records_in_range() would dirty the EXPLAIN statement result. So there is definitely a downside to what I was experimenting with. It would be great to hear what you all think :) - Toru _______________________________________________ Mailing list: https://launchpad.net/~drizzle-discuss Post to : [email protected] Unsubscribe : https://launchpad.net/~drizzle-discuss More help : https://help.launchpad.net/ListHelp
