tag 7389026f6ca01f81261b52ce0aacc15e6f6f59f4
tagger zhangbaowen <zhangbaowen@hashdata.cn> Mon Mar 11 17:31:58 2024 +0800
object 2ffc7e60dd4e3c7270ae1412d60abc604c186aec

Stamp Cloudberry Database 1.5.0

commit 2ffc7e60dd4e3c7270ae1412d60abc604c186aec
author Zhang Mingli <avamingli@gmail.com> Tue Nov 14 10:54:41 2023 +0800
committer Max Yang <79948451+my-ship-it@users.noreply.github.com> Fri Dec 01 15:52:55 2023 +0800
tree 1696c942dc4beffd4b3ec7383c964cd924087e04
parent 2bf926fe45aa550f8ad717fa5d28743ec1a83fa1

Answer Query Using Materialized Views.

AQUMV for short, is used to compute part or all of a Query from
materialized views during planning.
It could provide massive improvements in query processing time,
especially for aggregation queries over large tables[1].

AQUMV usually uses Incremental Materialized Views(IMV) as candidates,
as IMV usually have real time data when there are writable operations
on related tables.

AQUMV is actually a Equivalent Transformation on Query tree.
A materialized view(MV) could be use to compute a Query if:
1.The view contains all rows needed by the query expression.
  If MV has more rows than query wants, additional filter may be added
  if possible.
2.All output expressions can be computed from the output of the view.
  The output expressions could be fully or partially matched from MV's
  target list.
3.Cost-based.
  There may be multiple valid MV candidates, or select from MV is not
  better than select from origin table(ex: has an index and etc),
  let planner decide the best one.

Construct rows by splitting MV query quals(mv_query_quals) and Query quals
(origin_query_quals) to difference set and intersection set.
And post_quals formed by:{origin_query_quals - mv_query_quals} will be
processed by MV query's target list, and rewritten to MV relation's
target list expressions.

Construct columns expressions using a Greedy Algorithm.
Sort the MV query's target list by complexity, and try to rewrite expressions
by that order.
Expressions that have no Vars are kept to upper(Const Expressions) or be
rewritten if there were corresponding expressions.

Reference:
   [1] Optimizing Queries Using Materialized Views: A Practical,
Scalable Solution.
   https://courses.cs.washington.edu/courses/cse591d/01sp/opt_views.pdf

Authored-by: Zhang Mingli avamingli@gmail.com