go-mysql-server icon indicating copy to clipboard operation
go-mysql-server copied to clipboard

[stats] Use floats for stats counters, limit MCVs to distribution outliers

Open max-hoffman opened this issue 1 year ago • 1 comments

Three main changes:

  • stats use floats by default, to avoid uint->float round tripping
  • bucket MCVs are assumed to be sorted
  • MCVs are discarded if they are not at least twice as frequent as the average bucket frequency. The 2x number is arbitrary, but the motivation is that MCVs are expensive and they need to justify their cost by actually being outliers.

max-hoffman avatar Jun 18 '24 17:06 max-hoffman

Additional work is required for integration with Dolt. Additional work is required for integration with DoltgreSQL.

github-actions[bot] avatar Jun 18 '24 18:06 github-actions[bot]