hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Gopal Vijayaraghavan (Jira)" <j...@apache.org>
Subject [jira] [Commented] (HIVE-22269) Stats miss with "hive.optimize.sort.dynamic.partition" (SortedDynPartitionOptimizer) leads to wrong reducer count
Date Mon, 30 Sep 2019 22:18:00 GMT

    [ https://issues.apache.org/jira/browse/HIVE-22269?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16941375#comment-16941375
] 

Gopal Vijayaraghavan commented on HIVE-22269:
---------------------------------------------

Actually, we'd have caught it faster if the missing statistics was printed as "Statistics:
Missing" or something like that.

> Stats miss with "hive.optimize.sort.dynamic.partition" (SortedDynPartitionOptimizer)
leads to wrong reducer count
> -----------------------------------------------------------------------------------------------------------------
>
>                 Key: HIVE-22269
>                 URL: https://issues.apache.org/jira/browse/HIVE-22269
>             Project: Hive
>          Issue Type: Bug
>          Components: Statistics
>            Reporter: Rajesh Balamohan
>            Assignee: Rajesh Balamohan
>            Priority: Minor
>         Attachments: HIVE-22269.1.patch, HIVE-22269.2.patch
>
>
> {{hive.optimize.sort.dynamic.partition=true}} introduces new stage to reduce number of
writes in dynamic partitioning usecase. Earlier {{SortedDynPartitionOptimizer}} added this
new operator via {{Optimizer.java}} and the stats for the newly added operator was populated
via {{StatsRulesProcFactory$ReduceSinkStatsRule}}.
> However, with "HIVE-20703" this got changed. This is moved to {{TezCompiler}} for cost
based decision. Though the operator gets added correctly, the stats for this does not get
added (as it runs after runStatsAnnotation()).
> This causes reducer count to be mis-estimated in the query.
> {noformat}
> e.g For the following query, reducer_2 would be estimated as "2" instead of "1009". This
causes huge delay in the runtime.
> explain 
> from tpcds_xtext_1000.store_sales ss
> insert overwrite table store_sales partition (ss_sold_date_sk)
> select
>         ss.ss_sold_time_sk,
>         ss.ss_item_sk,
>         ss.ss_customer_sk,
>         ss.ss_cdemo_sk,
>         ss.ss_hdemo_sk,
>         ss.ss_addr_sk,
>         ss.ss_store_sk,
>         ss.ss_promo_sk,
>         ss.ss_ticket_number,
>         ss.ss_quantity,
>         ss.ss_wholesale_cost,
>         ss.ss_list_price,
>         ss.ss_sales_price,
>         ss.ss_ext_discount_amt,
>         ss.ss_ext_sales_price,
>         ss.ss_ext_wholesale_cost,
>         ss.ss_ext_list_price,
>         ss.ss_ext_tax,
>         ss.ss_coupon_amt,
>         ss.ss_net_paid,
>         ss.ss_net_paid_inc_tax,
>         ss.ss_net_profit,
>         ss.ss_sold_date_sk
>         where ss.ss_sold_date_sk is not null
> insert overwrite table store_sales partition (ss_sold_date_sk)
> select
>         ss.ss_sold_time_sk,
>         ss.ss_item_sk,
>         ss.ss_customer_sk,
>         ss.ss_cdemo_sk,
>         ss.ss_hdemo_sk,
>         ss.ss_addr_sk,
>         ss.ss_store_sk,
>         ss.ss_promo_sk,
>         ss.ss_ticket_number,
>         ss.ss_quantity,
>         ss.ss_wholesale_cost,
>         ss.ss_list_price,
>         ss.ss_sales_price,
>         ss.ss_ext_discount_amt,
>         ss.ss_ext_sales_price,
>         ss.ss_ext_wholesale_cost,
>         ss.ss_ext_list_price,
>         ss.ss_ext_tax,
>         ss.ss_coupon_amt,
>         ss.ss_net_paid,
>         ss.ss_net_paid_inc_tax,
>         ss.ss_net_profit,
>         ss.ss_sold_date_sk
>         where ss.ss_sold_date_sk is null
>         distribute by ss.ss_item_sk
> ;
> {noformat}



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Mime
View raw message