Poor performance in a simple join query

Question

We have 4 tables:
-- Customer
CREATE TABLE "Customer" (
  "CustomerId" RAW(16) NOT NULL,
  "MnemonicId" NUMBER(20) );
ALTER TABLE "Customer"
  ADD CONSTRAINT "PK_Customer"
  PRIMARY KEY ("CustomerId");
-- Language
CREATE TABLE "Language" (
  "Description" NVARCHAR2(250),
  "LanguageId" NVARCHAR2(12));
CREATE UNIQUE INDEX "PK_Language" ON "Language"
("LanguageId");
ALTER TABLE "Language" ADD (
  CONSTRAINT "PK_Language"
  PRIMARY KEY ("LanguageId"));
-- CustomerDescription
CREATE TABLE "CustomerDescription" (
  "Description" NVARCHAR2(250),
  "CustomerId" RAW(16) NOT NULL,
  "LanguageId" NVARCHAR2(12) NOT NULL);
ALTER TABLE "CustomerDescription"
  ADD CONSTRAINT "PK_CustomerDescription"
  PRIMARY KEY ("CustomerId", "LanguageId");
ALTER TABLE "CustomerDescription" ADD (
  CONSTRAINT "FK_CustomerDesc1" FOREIGN KEY ("CustomerId") REFERENCES "Customer" ("CustomerId"),
  CONSTRAINT "FK_CustomerDesc2" FOREIGN KEY ("LanguageId") REFERENCES "Language" ("LanguageId"));
-- JOINTABLE
CREATE TABLE JOINTABLE (ID  RAW(16) NOT NULL );We have also built a view vwCustomers on the tables: Customer, CustomerDescription, Language.
CREATE OR REPLACE FORCE VIEW "vwCustomers" (
  "Description",  "CustomerId",  "LanguageId",  "MnemonicId")
AS
   SELECT "CustomerDescription"."Description",
          "Customer"."CustomerId",
          "Language"."LanguageId",
          "Customer"."MnemonicId"
     FROM "Customer" CROSS JOIN "Language"
          LEFT OUTER JOIN "CustomerDescription"
          ON "Customer"."CustomerId" = "CustomerDescription"."CustomerId"
        AND "CustomerDescription"."LanguageId" = "Language"."LanguageId"The JOINTABLE table has only 1 row ("9C740128F750CF4ABC520DAF131D7E96")
The language table has 2 rows ("it-IT") and ("en-EN")
The Customer and CustometDescription tables has 350000 rows each
We have this simple query:
SELECT ROWNUM pos, VW."CustomerId", VW."LanguageId", VW."MnemonicId"
        FROM "vwCustomers" VW
        WHERE (VW."LanguageId" = 'it-IT')
        AND VW."CustomerId" IN ( SELECT Id FROM JOINTABLE)this query of course select only 1 row.
The time execution of this query is about 1 second, that for us is a really bad performance result.
We have also rewritten the query without the VIEW :
SELECT ROWNUM pos, "Customer"."CustomerId", "Language"."LanguageId", "Customer"."MnemonicId"
    FROM "Customer"
    CROSS JOIN "Language"
    LEFT OUTER JOIN "CustomerDescription"
    ON "Customer"."CustomerId" = "CustomerDescription"."CustomerId"
    AND "CustomerDescription"."LanguageId" = "Language"."LanguageId"
    WHERE ("Language"."LanguageId" = 'it-IT' )
    AND "Customer"."CustomerId" IN ( SELECT Id FROM JOINTABLE)Also for this query the execution time is about 1 second.
BUT if we write the query in a similar "way":
SELECT  ROWNUM pos, VW."CustomerId", VW."LanguageId", VW."MnemonicId"
        FROM "vwCustomers" VW
        WHERE (VW."LanguageId" = 'it-IT')
        AND VW."CustomerId" IN (HEXTORAW('9C740128F750CF4ABC520DAF131D7E96'))NOW we obtain as execution time of 0.0005 second!!!
We discover that when it takes 1 second in the first 2 queries this is because the access plan used has done a TABLE ACCESS FULL on the Customer table (that has 350000 rows). In the third query we have good performance because Oracle uses an INDEX UNIQUE SCAN on Customer table.
Please, could help me to obtain good performance in the first query?
thanks in advance
Filip
Edited by: [email protected] on 2-set-2009 12.59

Answer 1

The first execution plan shows at point 3 the TABLE ACCESS FULL SCAN on the table Customer that has Cardinality of 359,793
SELECT STATEMENT  ALL_ROWS Cost: 2,476  Bytes: 130  Cardinality: 1                                
     9 COUNT                           
          8 NESTED LOOPS OUTER  Cost: 2,476  Bytes: 130  Cardinality: 1                      
               6 HASH JOIN RIGHT SEMI  Cost: 2,475  Bytes: 102  Cardinality: 1                 
                    1 TABLE ACCESS FULL TABLE JOINTABLE Cost: 3  Bytes: 17  Cardinality: 1            
                    5 VIEW SYS. Cost: 2,470  Bytes: 30,582,405  Cardinality: 359,793            
                         4 NESTED LOOPS  Cost: 2,470  Bytes: 17,629,857  Cardinality: 359,793       
                              2 INDEX UNIQUE SCAN INDEX (UNIQUE)PK_Language Cost: 0  Bytes: 11  Cardinality: 1 
                              3 TABLE ACCESS FULL TABLE Customer Cost: 2,470  Bytes: 13,672,134  Cardinality: 359,793 
               7 INDEX UNIQUE SCAN INDEX (UNIQUE) PK_CustomerDescription Cost: 1  Bytes: 28  Cardinality: 1       In the second case at point 3 instead of TABLE ACCESS FULL SCAN Oracle execute an efficient TABLE ACCESS BY INDEX ROWID
SELECT STATEMENT  ALL_ROWS Cost: 3  Bytes: 99  Cardinality: 1                                
     8 COUNT                           
          7 MERGE JOIN OUTER  Cost: 3  Bytes: 99  Cardinality: 1                      
               5 VIEW SYS. Cost: 2  Bytes: 71  Cardinality: 1                 
                    4 NESTED LOOPS  Cost: 2  Bytes: 49  Cardinality: 1            
                         1 INDEX UNIQUE SCAN INDEX (UNIQUE) PK_Language Cost: 0  Bytes: 11  Cardinality: 1       
                         3 TABLE ACCESS BY INDEX ROWID TABLE Customer Cost: 2  Bytes: 38  Cardinality: 1       
                              2 INDEX UNIQUE SCAN INDEX (UNIQUE) PK_Customer Cost: 1  Cardinality: 1 
               6 INDEX UNIQUE SCAN INDEX (UNIQUE) PK_ CustomerDescription Cost: 1  Bytes: 28  Cardinality: 1