You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hive.apache.org by "Harish Butani (JIRA)" <ji...@apache.org> on 2014/10/28 02:06:34 UTC
[jira] [Commented] (HIVE-8526) Hive : CBO incorrect join order in
TPC-DS Q45 as self join selectivity has incorrect CE
[ https://issues.apache.org/jira/browse/HIVE-8526?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14186160#comment-14186160 ]
Harish Butani commented on HIVE-8526:
-------------------------------------
The patch fixes the issue with the Selectivity estimate for SemiJoins.
But as far as i can tell, Optiq MultiJoinRel doesn't handle SemiJoins. So for the query above, you are not going to get the optimal order.
[~julianhyde] can you confirm that MultiJoinRel doesn't support SemiJoins.
> Hive : CBO incorrect join order in TPC-DS Q45 as self join selectivity has incorrect CE
> ---------------------------------------------------------------------------------------
>
> Key: HIVE-8526
> URL: https://issues.apache.org/jira/browse/HIVE-8526
> Project: Hive
> Issue Type: Bug
> Components: CBO
> Affects Versions: 0.14.0
> Reporter: Mostafa Mokhtar
> Assignee: Harish Butani
> Priority: Critical
> Fix For: 0.14.0
>
> Attachments: HIVE-8526.1.patch
>
>
> The join order has Item joined last where it should be joined first
> Query
> {code}
> select ca_zip, ca_county, sum(ws_sales_price)
> from
> web_sales
> JOIN customer ON web_sales.ws_bill_customer_sk = customer.c_customer_sk
> JOIN customer_address ON customer.c_current_addr_sk = customer_address.ca_address_sk
> JOIN date_dim ON web_sales.ws_sold_date_sk = date_dim.d_date_sk
> JOIN item ON web_sales.ws_item_sk = item.i_item_sk
> where
> ( item.i_item_id in (select i_item_id
> from item i2
> where i2.i_item_sk in (2, 3, 5, 7, 11, 13, 17, 19, 23, 29)
> )
> )
> and d_qoy = 2 and d_year = 2000
> group by ca_zip, ca_county
> order by ca_zip, ca_county
> limit 100
> {code}
> Plan
> {code}
> 2014-10-20 18:43:16,521 DEBUG [main]: parse.SemanticAnalyzer (SemanticAnalyzer.java:apply(12330)) - HiveSortRel(fetch=[100]): rowcount = 1.710158597922807E7, cumulative cost = {7.169080587598123E10 rows, 3.420317295845614E7 cpu, 0.0 io}, id = 579
> HiveSortRel(sort0=[$0], sort1=[$1], dir0=[ASC], dir1=[ASC]): rowcount = 1.710158597922807E7, cumulative cost = {6.827294821015483E10 rows, 1.710158697922807E7 cpu, 0.0 io}, id = 577
> HiveProjectRel(ca_zip=[$0], ca_county=[$1], _o__c2=[$2]): rowcount = 1.710158597922807E7, cumulative cost = {6.485509054432843E10 rows, 1.0 cpu, 0.0 io}, id = 575
> HiveAggregateRel(group=[{0, 1}], agg#0=[sum($2)]): rowcount = 1.710158597922807E7, cumulative cost = {6.485509054432843E10 rows, 1.0 cpu, 0.0 io}, id = 573
> HiveProjectRel($f0=[$2], $f1=[$1], $f2=[$0]): rowcount = 6.0197670310147226E7, cumulative cost = {6.485509054432843E10 rows, 1.0 cpu, 0.0 io}, id = 571
> HiveProjectRel(ws_sales_price=[$2], ca_county=[$7], ca_zip=[$8]): rowcount = 6.0197670310147226E7, cumulative cost = {6.485509054432843E10 rows, 1.0 cpu, 0.0 io}, id = 569
> HiveFilterRel(condition=[AND(=($11, 2), =($10, 2000))]): rowcount = 6.0197670310147226E7, cumulative cost = {6.485509054432843E10 rows, 1.0 cpu, 0.0 io}, id = 567
> SemiJoinRel(condition=[=($13, $14)], joinType=[inner]): rowcount = 3.371069537368245E10, cumulative cost = {6.485509054432843E10 rows, 1.0 cpu, 0.0 io}, id = 565
> HiveProjectRel(ws_item_sk=[$0], ws_bill_customer_sk=[$1], ws_sales_price=[$2], ws_sold_date_sk=[$3], c_customer_sk=[$9], c_current_addr_sk=[$10], ca_address_sk=[$11], ca_county=[$12], ca_zip=[$13], d_date_sk=[$6], d_year=[$7], d_qoy=[$8], i_item_sk=[$4], i_item_id=[$5]): rowcount = 3.371069537368245E10, cumulative cost = {6.485509054332843E10 rows, 0.0 cpu, 0.0 io}, id = 669
> HiveJoinRel(condition=[=($1, $9)], joinType=[inner]): rowcount = 3.371069537368245E10, cumulative cost = {6.485509054332843E10 rows, 0.0 cpu, 0.0 io}, id = 667
> HiveJoinRel(condition=[=($3, $6)], joinType=[inner]): rowcount = 2.1594638446E10, cumulative cost = {4.3189811941E10 rows, 0.0 cpu, 0.0 io}, id = 664
> HiveJoinRel(condition=[=($0, $4)], joinType=[inner]): rowcount = 2.1594638446E10, cumulative cost = {2.1595100446E10 rows, 0.0 cpu, 0.0 io}, id = 601
> HiveProjectRel(ws_item_sk=[$2], ws_bill_customer_sk=[$3], ws_sales_price=[$20], ws_sold_date_sk=[$33]): rowcount = 2.1594638446E10, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 497
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.web_sales]]): rowcount = 2.1594638446E10, cumulative cost = {0}, id = 341
> HiveProjectRel(i_item_sk=[$0], i_item_id=[$1]): rowcount = 462000.0, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 555
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.item]]): rowcount = 462000.0, cumulative cost = {0}, id = 340
> HiveProjectRel(d_date_sk=[$0], d_year=[$6], d_qoy=[$10]): rowcount = 73049.0, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 551
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.date_dim]]): rowcount = 73049.0, cumulative cost = {0}, id = 342
> HiveJoinRel(condition=[=($1, $2)], joinType=[inner]): rowcount = 7.064015632843196E7, cumulative cost = {1.2E8 rows, 0.0 cpu, 0.0 io}, id = 598
> HiveProjectRel(c_customer_sk=[$0], c_current_addr_sk=[$4]): rowcount = 8.0E7, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 500
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.customer]]): rowcount = 8.0E7, cumulative cost = {0}, id = 343
> HiveProjectRel(ca_address_sk=[$0], ca_county=[$7], ca_zip=[$9]): rowcount = 4.0E7, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 547
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.customer_address]]): rowcount = 4.0E7, cumulative cost = {0}, id = 339
> HiveProjectRel(i_item_id=[$1]): rowcount = 1.05119214745814, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 563
> HiveProjectRel(i_item_sk=[$0], i_item_id=[$1]): rowcount = 1.05119214745814, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 561
> HiveFilterRel(condition=[in($0, 2, 3, 5, 7, 11, 13, 17, 19, 23, 29)]): rowcount = 1.05119214745814, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 559
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.item]]): rowcount = 462000.0, cumulative cost = {0}, id = 340
> {code}
> Then I rewrote the query trying to force CBO to generate the correct join order
> {code}
> with items as (select i_item_sk from
> item where
> ( item.i_item_id in (select i_item_id
> from item i2
> where i2.i_item_sk in (2, 3, 5, 7, 11, 13, 17, 19, 23, 29)
> )
> )
> )
> select ca_zip, ca_county, sum(ws_sales_price)
> from
> web_sales
> JOIN items ON web_sales.ws_item_sk = items.i_item_sk
> JOIN customer ON web_sales.ws_bill_customer_sk = customer.c_customer_sk
> JOIN customer_address ON customer.c_current_addr_sk = customer_address.ca_address_sk
> JOIN date_dim ON web_sales.ws_sold_date_sk = date_dim.d_date_sk
> where
> d_qoy = 2 and d_year = 2000
> group by ca_zip, ca_county
> order by ca_zip, ca_county
> limit 100
> {code}
> But the correct join order wasn't generated because CE for item x item + filter has a selectivity of 1.
> {code}
> 2014-10-20 18:46:27,120 DEBUG [main]: parse.SemanticAnalyzer (SemanticAnalyzer.java:apply(12330)) - HiveSortRel(fetch=[100]): rowcount = 1.6595391288544238E7, cumulative cost = {2.8364280421639153E10 rows, 3.3190782577088475E7 cpu, 0.0 io}, id = 1291
> HiveSortRel(sort0=[$0], sort1=[$1], dir0=[ASC], dir1=[ASC]): rowcount = 1.6595391288544238E7, cumulative cost = {2.505357243157397E10 rows, 1.6595391288544238E7 cpu, 0.0 io}, id = 1289
> HiveProjectRel(ca_zip=[$0], ca_county=[$1], _o__c2=[$2]): rowcount = 1.6595391288544238E7, cumulative cost = {2.174286444150879E10 rows, 0.0 cpu, 0.0 io}, id = 1287
> HiveAggregateRel(group=[{0, 1}], agg#0=[sum($2)]): rowcount = 1.6595391288544238E7, cumulative cost = {2.174286444150879E10 rows, 0.0 cpu, 0.0 io}, id = 1285
> HiveProjectRel($f0=[$9], $f1=[$8], $f2=[$2]): rowcount = 6.019767031014723E7, cumulative cost = {2.174286444150879E10 rows, 0.0 cpu, 0.0 io}, id = 1283
> HiveProjectRel(ws_item_sk=[$5], ws_bill_customer_sk=[$6], ws_sales_price=[$7], ws_sold_date_sk=[$8], i_item_sk=[$12], c_customer_sk=[$0], c_current_addr_sk=[$1], ca_address_sk=[$2], ca_county=[$3], ca_zip=[$4], d_date_sk=[$9], d_year=[$10], d_qoy=[$11]): rowcount = 6.019767031014723E7, cumulative cost = {2.174286444150879E10 rows, 0.0 cpu, 0.0 io}, id = 1380
> HiveJoinRel(condition=[=($6, $0)], joinType=[inner]): rowcount = 6.019767031014723E7, cumulative cost = {2.174286444150879E10 rows, 0.0 cpu, 0.0 io}, id = 1378
> HiveJoinRel(condition=[=($1, $2)], joinType=[inner]): rowcount = 7.064015632843196E7, cumulative cost = {1.2E8 rows, 0.0 cpu, 0.0 io}, id = 1309
> HiveProjectRel(c_customer_sk=[$0], c_current_addr_sk=[$4]): rowcount = 8.0E7, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 1269
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.customer]]): rowcount = 8.0E7, cumulative cost = {0}, id = 1035
> HiveProjectRel(ca_address_sk=[$0], ca_county=[$7], ca_zip=[$9]): rowcount = 4.0E7, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 1273
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.customer_address]]): rowcount = 4.0E7, cumulative cost = {0}, id = 1032
> HiveJoinRel(condition=[=($0, $7)], joinType=[inner]): rowcount = 3.856185436785714E7, cumulative cost = {2.16336624308125E10 rows, 0.0 cpu, 0.0 io}, id = 1376
> HiveJoinRel(condition=[=($3, $4)], joinType=[inner]): rowcount = 3.856185436785714E7, cumulative cost = {2.159463857644464E10 rows, 0.0 cpu, 0.0 io}, id = 1316
> HiveProjectRel(ws_item_sk=[$2], ws_bill_customer_sk=[$3], ws_sales_price=[$20], ws_sold_date_sk=[$33]): rowcount = 2.1594638446E10, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 1205
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.web_sales]]): rowcount = 2.1594638446E10, cumulative cost = {0}, id = 1033
> HiveProjectRel(d_date_sk=[$0], d_year=[$6], d_qoy=[$10]): rowcount = 130.44464285714287, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 1279
> HiveFilterRel(condition=[AND(=($10, 2), =($6, 2000))]): rowcount = 130.44464285714287, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 1277
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.date_dim]]): rowcount = 73049.0, cumulative cost = {0}, id = 1034
> HiveProjectRel(i_item_sk=[$0]): rowcount = 462000.0, cumulative cost = {1.0 rows, 1.0 cpu, 0.0 io}, id = 1265
> HiveFilterRel(condition=[=(1, 1)]): rowcount = 462000.0, cumulative cost = {1.0 rows, 1.0 cpu, 0.0 io}, id = 1263
> SemiJoinRel(condition=[=($1, $2)], joinType=[inner]): rowcount = 462000.0, cumulative cost = {1.0 rows, 1.0 cpu, 0.0 io}, id = 1261
> HiveProjectRel(i_item_sk=[$0], i_item_id=[$1]): rowcount = 462000.0, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 1253
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.item]]): rowcount = 462000.0, cumulative cost = {0}, id = 1024
> HiveProjectRel(i_item_id=[$1]): rowcount = 1.05119214745814, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 1259
> HiveProjectRel(i_item_sk=[$0], i_item_id=[$1]): rowcount = 1.05119214745814, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 1257
> HiveFilterRel(condition=[in($0, 2, 3, 5, 7, 11, 13, 17, 19, 23, 29)]): rowcount = 1.05119214745814, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 1255
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.item]]): rowcount = 462000.0, cumulative cost = {0}, id = 1024
> {code}
> This query generates the correct join order
> {code}
> with items as (select i_item_sk from
> item where
> item.i_item_id in (select i_item_id
> from item i2
> where i2.i_item_sk in (2, 3, 5, 7, 11, 13, 17, 19, 23, 29)
> )
>
> ),
> ws as (
> select ws_bill_customer_sk,ws_sales_price,ws_sold_date_sk
> from web_sales
> JOIN items ON web_sales.ws_item_sk = items.i_item_sk
> )
> select ca_zip, ca_county, sum(ws_sales_price)
> from ws
> JOIN customer ON ws.ws_bill_customer_sk = customer.c_customer_sk
> JOIN customer_address ON customer.c_current_addr_sk = customer_address.ca_address_sk
> JOIN date_dim ON ws.ws_sold_date_sk = date_dim.d_date_sk
> where d_qoy = 2 and d_year = 2000
> group by ca_zip, ca_county
> order by ca_zip, ca_county
> limit 100
> {code}
> Plan
> {code}
> 2014-10-20 19:13:15,989 DEBUG [main]: parse.SemanticAnalyzer (SemanticAnalyzer.java:apply(12330)) - HiveSortRel(fetch=[100]): rowcount = 1.6595391288544238E7, cumulative cost = {4.99203570142713E10 rows, 3.3190783577088475E7 cpu, 0.0 io}, id = 4367
> HiveSortRel(sort0=[$0], sort1=[$1], dir0=[ASC], dir1=[ASC]): rowcount = 1.6595391288544238E7, cumulative cost = {4.6609649024206116E10 rows, 1.6595392288544238E7 cpu, 0.0 io}, id = 4365
> HiveProjectRel(ca_zip=[$0], ca_county=[$1], _o__c2=[$2]): rowcount = 1.6595391288544238E7, cumulative cost = {4.329894103414093E10 rows, 1.0 cpu, 0.0 io}, id = 4363
> HiveAggregateRel(group=[{0, 1}], agg#0=[sum($2)]): rowcount = 1.6595391288544238E7, cumulative cost = {4.329894103414093E10 rows, 1.0 cpu, 0.0 io}, id = 4361
> HiveProjectRel($f0=[$7], $f1=[$6], $f2=[$1]): rowcount = 6.019767031014723E7, cumulative cost = {4.329894103414093E10 rows, 1.0 cpu, 0.0 io}, id = 4359
> HiveProjectRel(ws_bill_customer_sk=[$5], ws_sales_price=[$6], ws_sold_date_sk=[$7], c_customer_sk=[$0], c_current_addr_sk=[$1], ca_address_sk=[$2], ca_county=[$3], ca_zip=[$4], d_date_sk=[$8], d_year=[$9], d_qoy=[$10]): rowcount = 6.019767031014723E7, cumulative cost = {4.329894103414093E10 rows, 1.0 cpu, 0.0 io}, id = 4426
> HiveJoinRel(condition=[=($5, $0)], joinType=[inner]): rowcount = 6.019767031014723E7, cumulative cost = {4.329894103414093E10 rows, 1.0 cpu, 0.0 io}, id = 4424
> HiveJoinRel(condition=[=($1, $2)], joinType=[inner]): rowcount = 7.064015632843196E7, cumulative cost = {1.2E8 rows, 0.0 cpu, 0.0 io}, id = 4392
> HiveProjectRel(c_customer_sk=[$0], c_current_addr_sk=[$4]): rowcount = 8.0E7, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 4345
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.customer]]): rowcount = 8.0E7, cumulative cost = {0}, id = 4101
> HiveProjectRel(ca_address_sk=[$0], ca_county=[$7], ca_zip=[$9]): rowcount = 4.0E7, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 4349
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.customer_address]]): rowcount = 4.0E7, cumulative cost = {0}, id = 4099
> HiveJoinRel(condition=[=($2, $3)], joinType=[inner]): rowcount = 3.856185436785714E7, cumulative cost = {4.318973902344464E10 rows, 1.0 cpu, 0.0 io}, id = 4395
> HiveProjectRel(ws_bill_customer_sk=[$1], ws_sales_price=[$2], ws_sold_date_sk=[$3]): rowcount = 2.1594638446E10, cumulative cost = {2.1595100447E10 rows, 1.0 cpu, 0.0 io}, id = 4343
> HiveProjectRel(ws_item_sk=[$0], ws_bill_customer_sk=[$1], ws_sales_price=[$2], ws_sold_date_sk=[$3], i_item_sk=[$4]): rowcount = 2.1594638446E10, cumulative cost = {2.1595100447E10 rows, 1.0 cpu, 0.0 io}, id = 4388
> HiveJoinRel(condition=[=($0, $4)], joinType=[inner]): rowcount = 2.1594638446E10, cumulative cost = {2.1595100447E10 rows, 1.0 cpu, 0.0 io}, id = 4383
> HiveProjectRel(ws_item_sk=[$2], ws_bill_customer_sk=[$3], ws_sales_price=[$20], ws_sold_date_sk=[$33]): rowcount = 2.1594638446E10, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 4277
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.web_sales]]): rowcount = 2.1594638446E10, cumulative cost = {0}, id = 4096
> HiveProjectRel(i_item_sk=[$0]): rowcount = 462000.0, cumulative cost = {1.0 rows, 1.0 cpu, 0.0 io}, id = 4339
> HiveFilterRel(condition=[=(1, 1)]): rowcount = 462000.0, cumulative cost = {1.0 rows, 1.0 cpu, 0.0 io}, id = 4337
> SemiJoinRel(condition=[=($1, $2)], joinType=[inner]): rowcount = 462000.0, cumulative cost = {1.0 rows, 1.0 cpu, 0.0 io}, id = 4335
> HiveProjectRel(i_item_sk=[$0], i_item_id=[$1]): rowcount = 462000.0, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 4327
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.item]]): rowcount = 462000.0, cumulative cost = {0}, id = 4088
> HiveProjectRel(i_item_id=[$1]): rowcount = 1.05119214745814, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 4333
> HiveProjectRel(i_item_sk=[$0], i_item_id=[$1]): rowcount = 1.05119214745814, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 4331
> HiveFilterRel(condition=[in($0, 2, 3, 5, 7, 11, 13, 17, 19, 23, 29)]): rowcount = 1.05119214745814, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 4329
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.item]]): rowcount = 462000.0, cumulative cost = {0}, id = 4088
> HiveProjectRel(d_date_sk=[$0], d_year=[$6], d_qoy=[$10]): rowcount = 130.44464285714287, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 4355
> HiveFilterRel(condition=[AND(=($10, 2), =($6, 2000))]): rowcount = 130.44464285714287, cumulative cost = {0.0 rows, 0.0 cpu, 0.0 io}, id = 4353
> HiveTableScanRel(table=[[tpcds_bin_partitioned_orc_30000.date_dim]]): rowcount = 73049.0, cumulative cost = {0}, id = 4100
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)