[英]Materialized view fast refresh - unexpected behavior in Oracle 12c
I've faced unexpected behavior of materialized view fast refresh. 我遇到了物化视图快速刷新的意外行为。 It turns to be that at a background database queries unrelated m.view logs. 事实证明,在后台数据库查询不相关的m.view日志。
Imagine we have two tables with m.view logs and one m.view for one of the tables: 想象一下,我们有两个带有m.view日志的表,一个表的m.view:
Initially fast refresh for table1 works fast. 最初,对table1的快速刷新工作很快。 But if we fill table2 in another session with some big amount of data in one transaction, then fast refresh of m.view for table1 degradates. 但是,如果我们在另一个会话中用一个事务中的大量数据填充table2,则对table1的m.view的快速刷新会降低性能。 Most of the time is spent on select over m.view log for table2, which has no relations to refreshing m.view . 大部分时间都花在选择table2的m.view日志上,这与刷新m.view无关 。
My database version: 我的数据库版本:
Oracle Database 12c Enterprise Edition Release 12.1.0.2.0 - 64bit Production PL/SQL Release 12.1.0.2.0 - Production "CORE 12.1.0.2.0 Production" TNS for Linux: Version 12.1.0.2.0 - Production NLSRTL Version 12.1.0.2.0 - Production
drop materialized view mv_table1;
drop table table1 purge;
drop table table2 purge;
create table table1(
a integer primary key,
b date,
c varchar2(16)
);
create table table2(
a integer primary key,
b date,
c varchar2(16)
);
create materialized view log on table1
with rowid, sequence, commit scn ( b, c ) including new values;
create materialized view log on table2
with rowid, sequence, commit scn ( b, c ) including new values;
create materialized view mv_table1 (b, c, cnt)
refresh fast on demand
as
select b, c, count(*) cnt
from table1
group by b, c;
SQL> set timing on
SQL> exec dbms_mview.refresh('MV_TABLE1','F');
PL/SQL procedure successfully completed
Executed in 0,25 seconds
insert into table2(a,b,c)
with get_data(a,b,c,i) as (
select 1 a,trunc(sysdate) b,to_char(sysdate,'yyyy-mm-dd')||'-'||1 c, 2 i from dual
union all
select i, trunc(sysdate)+mod(i,10000),to_char(sysdate+mod(i,10000),'yyyy-mm-dd')||'-'||mod(i,10000), i + 1
from get_data
where i<= 3000000
)
select a,b,c
from get_data;
SQL> set timing on
SQL> exec dbms_mview.refresh('MV_TABLE1','F');
PL/SQL procedure successfully completed
Executed in 5,367 seconds
In av$session we see that Session 1 spends most of time for selecting from m.view log for table2: 在av $ session中,我们看到会话1花费大部分时间从m.view日志中选择table2:
select 1 from "SCOTT"."MLOG$_TABLE2" where rownum=1
This is information from the trace file (separate run): 这是来自跟踪文件的信息(单独运行):
SQL ID: 89s2c53j480vz Plan Hash: 1479141652
select 1
from
"SCOTT"."MLOG$_TABLE2" where rownum=1
call count cpu elapsed disk query current rows
------- ------ -------- ---------- ---------- ---------- ---------- ----------
Parse 1 0.00 0.00 0 4 0 0
Execute 1 0.00 0.00 0 0 0 0
Fetch 1 5.88 5.89 0 3058867 0 0
------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 3 5.88 5.89 0 3058871 0 0
Misses in library cache during parse: 1
Optimizer mode: ALL_ROWS
Parsing user id: SYS (recursive depth: 1)
Number of plan statistics captured: 1
Rows (1st) Rows (avg) Rows (max) Row Source Operation
---------- ---------- ---------- ---------------------------------------------------
0 0 0 COUNT STOPKEY (cr=3058867 pr=0 pw=0 time=5891187 us)
0 0 0 TABLE ACCESS FULL MLOG$_TABLE2 (cr=3058867 pr=0 pw=0 time=5891181 us cost=7990 size=0 card=1)
Additionally, in the trace file I see that there are queries to ALL the m.view logs in the db, even those that are located in anoter schemas. 此外,在跟踪文件中,我看到对数据库中的所有 m.view日志都有查询,即使是位于注释模式中的查询。
Has anyone else faced such unexpected behavior ? 还有其他人面对过这样的意外行为吗? How did you solve it ? 您是如何解决的? Do you have ideas for workaround ? 您有解决方法的想法吗?
Use timestamp based
mat.view logs instead of commit scn
based ones. 使用timestamp based
mat.view日志,而不是基于commit scn
的日志。
解决方法:基于时间戳的mat.view日志快速刷新基于提交scn的mat.view日志的怪异行为看起来像一个Oracle错误。
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.