检查数据倾斜分布

从传统数据库迁移到GP中一个重要的且经常被开发人员忽略的概念是数据分布，没有良好的设计表的分布键会导致严重的性能问题，以下函数将给开发人员及DBA检测一个表的数据倾斜情况。 -- Function: gpmg.data_skew(character varying) -- DROP FUNCTION gpmg.da

-- Function: gpmg.data_skew(character varying)
 
-- DROP FUNCTION gpmg.data_skew(character varying);
 
CREATE OR REPLACE FUNCTION gpmg.data_skew(tablename character varying)
  RETURNS text AS
$BODY$
--2014-05-26,Gtlions,收集和统计数据倾斜情况
declare
  v_func character varying(200)=&＃39;gpmg.data_skew()&＃39;;
  v_begin_time timestamp;
  v_end_time timestamp;
  v_status int=0;
  v_msg text=&＃39;Done.&＃39;;
  v_record record;
 
  v_id integer;
  v_rq timestamp;  
  v_segs integer=64;
  v_totalnums bigint=0;
  v_maxskew numeric=0.0;
  v_minskew numeric=0.0;
  v_maxskew_seg varchar(20);
  v_minskew_seg varchar(20);
  v_maxrows bigint=0;
  v_minrows bigint=0;   
  v_result varchar(2000);
 
begin
  v_id=nextval(&＃39;gpmg.commonseq&＃39;);
  v_rq=now();
  v_begin_time=clock_timestamp();
  v_result = &＃39;GP hava &＃39;;
  select into v_segs count(*) segs from gp_segment_configuration where role=&＃39;p&＃39; and content<>-1;
  v_result = v_result||v_segs||&＃39; instances, Standard skew is &＃39;||1.0/v_segs||&＃39;. &＃39;;
  -- bg1 segid, bg2 节点记录数量
  execute &＃39;insert into gpmg.commontab(seq,tabname,bg1,bg2) select &＃39;||v_id||&＃39;,&＃39;&＃39;&＃39;||$1||&＃39;&＃39;&＃39;,gp_segment_id,count(*) segrownums from &＃39;||$1||&＃39; group by rollup(( gp_segment_id)) order by gp_segment_id&＃39;;
  select into v_segs,v_totalnums v_segs,max(bg2) from gpmg.commontab where seq=v_id and tabname=$1;
  --nm1 标准倾斜率, nm2 节点倾斜率, nm3 标准-节点倾斜率绝对值
  update gpmg.commontab set nm1=1::numeric/v_segs,nm2=bg2::numeric/v_totalnums,nm3=abs(1::numeric/v_segs-bg2::numeric/v_totalnums) where seq=v_id and tabname=$1;
  select into v_maxskew,v_minskew max(nm2),min(nm2) from gpmg.commontab where seq=v_id and tabname=$1 and bg1 is not null;
 
  select into v_maxskew_seg hostname from gp_segment_configuration where role=&＃39;p&＃39; and content in (select bg1 from gpmg.commontab where seq=v_id and tabname=$1 and bg1 is not null and nm2=v_maxskew limit 1);
  select into v_minskew_seg hostname from gp_segment_configuration where role=&＃39;p&＃39; and content in (select bg1 from gpmg.commontab where seq=v_id and tabname=$1 and bg1 is not null and nm2=v_minskew limit 1);
 
  select into v_maxrows bg2 from gpmg.commontab where seq=v_id and tabname=$1 and bg1 is not null and nm2=v_maxskew limit 1;
  select into v_minrows bg2 from gpmg.commontab where seq=v_id and tabname=$1 and bg1 is not null and nm2=v_minskew limit 1;
 
  v_result =v_result ||&＃39;You Table [&＃39;||$1||&＃39;] skew info: [table_totalrows:&＃39;||v_totalnums||&＃39;, maxskew:seg-&＃39;||v_maxskew_seg||&＃39;, rows-&＃39;||v_maxrows||&＃39; &＃39;||v_maxskew||&＃39;, minskew:seg-&＃39;||v_minskew_seg||&＃39;, rows-&＃39;||v_minrows||&＃39; &＃39;||v_minskew||&＃39;]&＃39;;
  delete from gpmg.commontab where seq=v_id and tabname=$1;
  return v_result;
  v_end_time=clock_timestamp();
end;
$BODY$
  LANGUAGE plpgsql VOLATILE;
ALTER FUNCTION gpmg.data_skew(character varying)
  OWNER TO gpadmin;
GRANT EXECUTE ON FUNCTION gpmg.data_skew(character varying) TO public;
GRANT EXECUTE ON FUNCTION gpmg.data_skew(character varying) TO gpadmin;

bigdatagp=# select gpmg.data_skew(&＃39;gpmg.manager_table&＃39;);
                                                                                                            data_skew                                                  
                                                           
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------
-----------------------------------------------------------
 GP hava 64 instances, Standard skew is 0.01562500000000000000. You Table [gpmg.manager_table] skew info: [table_totalrows:83, maxskew:seg-sdw16, rows-3 0.036144578313
25301205, minskew:seg-sdw2, rows-1 0.01204819277108433735]
(1 row)
 
bigdatagp=# select gpmg.data_skew(&＃39;gpmg.func_log&＃39;);
                                                                                                             data_skew                                                 
                                                             
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------
-------------------------------------------------------------
 GP hava 64 instances, Standard skew is 0.01562500000000000000. You Table [gpmg.func_log] skew info: [table_totalrows:53708, maxskew:seg-sdw10, rows-907 0.016887614508
08073285, minskew:seg-sdw7, rows-773 0.01439264169211290683]
(1 row)
2014-10-14 09:53:00