maria-developers team mailing list archive

[worklog-noreply@xxxxxxxxxxxx]

-----------------------------------------------------------------------
                              WORKLOG TASK
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
TASK...........: Better choice between range and index_merge/intersection options
CREATION DATE..: Wed, 27 May 2009, 15:20
SUPERVISOR.....: Monty
IMPLEMENTOR....: Psergey
COPIES TO......: 
CATEGORY.......: Server-Sprint
TASK ID........: 26 (http://askmonty.org/worklog/?tid=26)
VERSION........: Server-9.x
STATUS.........: Un-Assigned
PRIORITY.......: 60
WORKED HOURS...: 0
ESTIMATE.......: 0 (hours remain)
ORIG. ESTIMATE.: 0

PROGRESS NOTES:

-=-=(Guest - Sat, 20 Jun 2009, 09:39)=-=-
High-Level Specification modified.
--- /tmp/wklog.26.old.21828     2009-06-20 09:39:25.000000000 +0300
+++ /tmp/wklog.26.new.21828     2009-06-20 09:39:25.000000000 +0300
@@ -1 +1,9 @@
+* Not a spec but rather preliminary notes*
 
+User cases
+----------
+
+* BUG#32254: Index merge used unnecessarily
+* BUG#34869: Almost 300% regression in index merge intersect performance
+* BUG#40051: needed way to prevent optimizer from using index_merge on useless
+keys (no testcase) 

-=-=(Guest - Sat, 13 Jun 2009, 06:51)=-=-
Category updated.
--- /tmp/wklog.26.old.26624     2009-06-13 06:51:56.000000000 +0300
+++ /tmp/wklog.26.new.26624     2009-06-13 06:51:56.000000000 +0300
@@ -1 +1 @@
-Server-BackLog
+Server-Sprint

-=-=(Guest - Sat, 13 Jun 2009, 06:06)=-=-
Category updated.
--- /tmp/wklog.26.old.24784     2009-06-13 06:06:28.000000000 +0300
+++ /tmp/wklog.26.new.24784     2009-06-13 06:06:28.000000000 +0300
@@ -1 +1 @@
-Server-RawIdeaBin
+Server-BackLog

-=-=(Psergey - Wed, 03 Jun 2009, 12:09)=-=-
Dependency created: 30 now depends on 26



DESCRIPTION:

The optimizer does a cost-based choice between possible range and
index_merge/intersection scans.  There are some issues with it:

- index_merge/intersection gets chosen even when there is a single 
  multi-part index that covers all keys. Measurements show that this is 
  a poor choice.

- The picked index_merge/intersection can use a redundant set of indexes:
  it will be intersect(idx1, ..., idxN) where all columns in idxN are covered
  by other used indexes.

This WL is to fix these limitations.


HIGH-LEVEL SPECIFICATION:



* Not a spec but rather preliminary notes*

User cases
----------

* BUG#32254: Index merge used unnecessarily
* BUG#34869: Almost 300% regression in index merge intersect performance
* BUG#40051: needed way to prevent optimizer from using index_merge on useless
keys (no testcase) 


ESTIMATED WORK TIME

ESTIMATED COMPLETION DATE
-----------------------------------------------------------------------
WorkLog (v3.5.9)