通过/ distinct查询firebird慢速顺序

Question

我对Firebird中的查询速度有疑问.缓慢是在排序和清晰.

如果我在MySQL中尝试查询,那么他会快一秒.

Firebird - > 1,3s a 1,6s MySQL - > 0,3s a 0,4s

我们在Web服务器/网站上使用Firebird数据库,因此速度很重要.

规格: - Firebird 2.5.1或2.5.2(SuperClassic)64位 - 2,13 Ghz(2个处理器) - RAM 4,00 GB

我能做什么？

我有以下表格:

================================================== ==

CREATE TABLE ARTICLE3_1
(
  IDARTICLE Integer NOT NULL,
  ITEMSTATUS Integer,
  ITEMENTRYDATE Integer,
  ITEMFILTER Integer,
  ARTIKELNUMMER Varchar(250),
  ARTIKELNAAM1 Varchar(250),
  ARTIKELNAAM2 Varchar(250),
  OMSCHRIJVING_DETAIL Blob sub_type 1,
  OMSCHRIJVING1 Varchar(250),
  OMSCHRIJVING2 Varchar(250),
  ARTIKELNR_LEVERANCIER Varchar(250),
  MERK Varchar(250),
  LEVERANCIER Varchar(250),
  EAN Varchar(250),
  LINKAANGROEP Varchar(250),
  LINKAANAANBIEDINGGROEP Varchar(250),
  LINKAANPOPULAIRGROEP Varchar(250),
  LINKAANART Varchar(250),
  ARTGRPNR Varchar(250),
  SUBGROEP Varchar(250),
  PRIJSPER Integer,
  VERKOOPPRIJS Float,
  ADVIESPRIJS Float,
  BTWPERC Float,
  ONLINE Varchar(250),
  TUSGROEPBIJLINK Varchar(250),
  AFBEELDINGKLEIN Varchar(250),
  AFBEELDINGMIDDEL Varchar(250),
  AFBEELDINGGROOT Varchar(250),
  ICECATLINK Varchar(250),
  LINKAANHOMEPAGEGROEP Varchar(250),
  LINKAANMIJNACCOUNTGROEP Varchar(250),
  SORTEER Varchar(250),
  AFBEELDING Varchar(100),
  FLASH Blob sub_type 1,
  EENHEID Varchar(250),
  ALTARTNR1 Varchar(250),
  ALTARTNR2 Varchar(250),
  BESTELLENPER Float,
  INFEED Varchar(250),
  GOOGLE_TAXONOMIE Varchar(250),
  FEED_TITEL Varchar(250),
  FEED_OMSCHRIJVING Blob sub_type 1,
  PRIMARY KEY (IDARTICLE)
);
CREATE INDEX IDX_ARTICLE3_1_2 ON ARTICLE3_1 (MERK);
CREATE INDEX IDX_ARTICLE3_1_3 ON ARTICLE3_1 (ARTIKELNUMMER);
CREATE INDEX IDX_ARTICLE3_1_4 ON ARTICLE3_1 (ARTIKELNR_LEVERANCIER);
CREATE INDEX IDX_ARTICLE3_1_5 ON ARTICLE3_1 (ALTARTNR2);
CREATE INDEX IDX_ARTICLE3_1_6 ON ARTICLE3_1 (ARTIKELNAAM1);
CREATE INDEX IDX_ARTICLE3_1_7 ON ARTICLE3_1 (EAN);

   CREATE TABLE TREE3
(
  IDLINK Integer NOT NULL,
  LINKTYPE Integer,
  IDITEM Integer,
  ITEMTYPE Integer,
  IDTARGETLINK Integer,
  NODEPOSITION Integer,
  NODELEVEL Integer,
  IDLAYOUTDATA Integer,
  IDTEMPLATE Integer,
  ACTIONDATE Integer,
  MARKET1 Integer,
  PRIMARY KEY (IDLINK)
);
CREATE INDEX IDX_TREE3_2 ON TREE3 (IDITEM);
CREATE INDEX IDX_TREE3_3 ON TREE3 (MARKET1);
CREATE INDEX ITREE13 ON TREE3 (IDTARGETLINK,NODEPOSITION);
CREATE INDEX ITREE53 ON TREE3 (IDITEM,ITEMTYPE);

====================================================

FireBird中的查询:

SELECT FIRST 30 SKIP 0 distinct tr.IdLink, tr.IdTargetLink, tr.IdItem, tr.NodePosition
FROM Tree3 tr
inner join article3_1 art on art.idarticle = Tr.iditem
WHERE tr.ItemType = 2 AND tr.Market1 = 1
AND  ((art.IDARTICLE > 0) AND (  (LOWER(art.Artikelnummer) like '%a4 papier%' ) OR ( (LOWER(art.Artikelnummer) like 'a4' )
AND (LOWER(art.Artikelnummer) like 'papier'))  OR  (LOWER(art.Artikelnaam1) like '%a4 papier%' ) OR ( (LOWER(art.Artikelnaam1) like '%a4%' )
AND (LOWER(art.Artikelnaam1) like '%papier%'))  OR  (LOWER(art.Artikelnaam2) like '%a4 papier%' ) OR ( (LOWER(art.Artikelnaam2) like '%a4%' )
AND (LOWER(art.Artikelnaam2) like '%papier%'))  OR  (LOWER(art.Artikelnr_leverancier) like '%a4 papier%' ) OR ( (LOWER(art.Artikelnr_leverancier) like '%a4%' )
AND (LOWER(art.Artikelnr_leverancier) like '%papier%'))  OR  (LOWER(art.Merk) like '%a4 papier%' ) OR ( (LOWER(art.Merk) like '%a4%' )
AND (LOWER(art.Merk) like '%papier%'))  OR  (LOWER(art.EAN) like '%a4 papier%' ) OR ( (LOWER(art.EAN) like '%a4%' )
AND (LOWER(art.EAN) like '%papier%'))  OR  (LOWER(art.AltArtnr1) like '%a4 papier%' ) OR ( (LOWER(art.AltArtnr1) like '%a4%' )
AND (LOWER(art.AltArtnr1) like '%papier%'))  OR  (LOWER(art.AltArtnr2) like '%a4 papier%' ) OR ( (LOWER(art.AltArtnr2) like '%a4%' )
AND (LOWER(art.AltArtnr2) like '%papier%')) ))
AND tr.NODELEVEL =5  and tr.LINKTYPE <> 5
ORDER BY tr.NodePosition

MySQL中的查询:

SELECT  distinct tr.IdLink, tr.IdTargetLink, tr.IdItem, tr.NodePosition
FROM Tree3 tr
inner join article3_1 art on art.idarticle = Tr.iditem
WHERE tr.ItemType = 2 AND tr.Market1 = 1
AND  ((art.IDARTICLE > 0) AND (  (LCASE(art.Artikelnummer) like '%a4 papier%' ) OR ( (LCASE(art.Artikelnummer) like 'a4' )
AND (LCASE(art.Artikelnummer) like 'papier'))  OR  (LCASE(art.Artikelnaam1) like '%a4 papier%' ) OR ( (LCASE(art.Artikelnaam1) like '%a4%' )
AND (LCASE(art.Artikelnaam1) like '%papier%'))  OR  (LCASE(art.Artikelnaam2) like '%a4 papier%' ) OR ( (LCASE(art.Artikelnaam2) like '%a4%' )
AND (LCASE(art.Artikelnaam2) like '%papier%'))  OR  (LCASE(art.Artikelnr_leverancier) like '%a4 papier%' ) OR ( (LCASE(art.Artikelnr_leverancier) like '%a4%' )
AND (LCASE(art.Artikelnr_leverancier) like '%papier%'))  OR  (LCASE(art.Merk) like '%a4 papier%' ) OR ( (LCASE(art.Merk) like '%a4%' )
AND (LCASE(art.Merk) like '%papier%'))  OR  (LCASE(art.EAN) like '%a4 papier%' ) OR ( (LCASE(art.EAN) like '%a4%' )
AND (LCASE(art.EAN) like '%papier%'))  OR  (LCASE(art.AltArtnr1) like '%a4 papier%' ) OR ( (LCASE(art.AltArtnr1) like '%a4%' )
AND (LCASE(art.AltArtnr1) like '%papier%'))  OR  (LCASE(art.AltArtnr2) like '%a4 papier%' ) OR ( (LCASE(art.AltArtnr2) like '%a4%' )
AND (LCASE(art.AltArtnr2) like '%papier%')) ))
AND tr.NODELEVEL =5  and tr.LINKTYPE <> 5
ORDER BY tr.NodePosition LIMIT 30;

================================================== ==

我有使用FlameRobin执行的查询:

> Prepare time: 0.016s Field #01: TREE3.IDLINK Alias:IDLINK Type:INTEGER
> Field #02: TREE3.IDTARGETLINK Alias:IDTARGETLINK Type:INTEGER Field
> #03: TREE3.IDITEM Alias:IDITEM Type:INTEGER Field #04: TREE3.NODEPOSITION Alias:NODEPOSITION Type:INTEGER PLAN SORT (SORT
> (JOIN (TR INDEX (IDX_TREE3_2, IDX_TREE3_3), ART INDEX
> (RDB$PRIMARY2))))
> 
> 873424 fetches, 0 marks, 12892 reads, 0 writes. 0 inserts, 0 updates,
> 0 deletes, 380580 index, 0 seq. Delta memory: 1784 bytes. Total
> execution time: 1.311s

谢谢!

Answer 1

现在可能有点旧了，但希望仍然有帮助。

一般来说，distinct和order by操作都需要排序。排序由索引辅助。考虑为 order by 子句中指定的列创建索引 - NodePosition，我能看到的唯一其他索引是与另一列组合的，因此 order by 不会查阅该索引。对于不同的，您可以尝试为 tr.IdLink、tr.IdTargetLink、tr.IdItem、tr.NodePosition 列或单独创建复合索引。（我不太确定索引对不同的值有多大帮助，但值得一试）。

其他需要考虑的事情：您的 where 子句使用函数 - 在此上下文中使用函数将导致全表扫描，甚至可能不会查看索引。我不相信 mySql 支持基于函数的索引，不确定 FireBird。但可以通过创建另一个可以保存 LOWER(column) 结果的列来解决此问题，您需要使用触发器（如果可用）来维护该列。

OR 条件和 LIKE '%a4%' 也会导致全表扫描。我意识到您的业务逻辑可能不允许您从“%a4%”字符串的开头删除通配符，因此为了可能改进此类用例，您可以考虑子查询 - 首先尝试缩小结果集的范围可能在子查询中避免任何 LIKE 或 OR，然后用父查询包装该结果，这将进一步过滤结果（将子查询放入 FROM 子句）。因此，在子查询中，您将具有以下条件： tr.ItemType = 2 AND tr.Market1 = 1 和 tr.NODELEVEL =5 和 tr.LINKTYPE <> 5