The information provided by event 10032 can be used to check what the average size of the reads is. The trace file will contain a section such as the following:Dazu wieder ein sehr übersichtlicher Test in 11.1.0.7, in dem ich nur das Event aktiviere und eine harmlose CTAS-Operation mit Analytics durchführe
Size of read slots for merge phase 57344 Number of read slots for merge phase 26 Size of read slots for output 57344 Number of read slots for output 21 Number of direct sync reads 1557 Number of blocks read synchronously 10063
The number of blocks read can be divided by the number of reads to get the number of blocks per read operation. If it is low (close to 1 block per read), then you may be encountering this problem.
alter session set events '10032 trace name context forever, level 1'; drop table test_analytics_1; create table test_analytics_1 as select col1 , col2 , col3 , col4 , col_day_of_month , median (val1) over ( partition by col1 ) median_val1 , val1 from test_analytics;
Im Trace-File finden sich dazu folgende Angaben:
---- Sort Parameters ------------------------------ sort_area_size 2949120 sort_area_retained_size 2949120 sort_multiblock_read_count 15 max intermediate merge width 4 ---- Direct Read Statistics ------ Size of read slots for merge phase 245760 Number of read slots for merge phase 8 Size of read slots for output 49152 Number of read slots for output 24 Number of direct sync reads 12141 Number of blocks read synchronously 12185
Ich verstehe den MOS-Hinweis so, dass die Number of blocks read synchronously/Number of direct sync reads knapp über 1 liegt, wenn man diesen Bug trifft - und das ist dann hier offenbar der Fall.
Keine Kommentare:
Kommentar veröffentlichen