[
https://issues.apache.org/jira/browse/SPARK-25094?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16578127#comment-16578127
]
Izek Greenfield edited comment on SPARK-25094 at 8/13/18 1:03 PM:
------------------------------------------------------------------
the code that creates this plan is very complex.
I will try to reproduce it in simple code in the meanwhile I can attach the
generated code so you can see the problem is that the code does not create
functions and inline all the Plan into the processNext method.
[^generated_code.txt]
it contains 2 DataFrames on with 80 columns 10 of them built from `case when`
expressions:
like that:
CASE WHEN (`predefined_hc` IS NOT NULL) THEN '/Predefined_hc/' WHEN
(`zero_volatility_adj_ind` = 'Y') THEN '/Zero_Haircuts_cases/' WHEN
(`collateral_allocation_method` = 'FCSM') THEN '/FCSM_Collaterals/' WHEN
((((`underlying_type` = 'DEBT') AND (`issuer_type` = 'CGVT')) AND
((`instrument_cqs_st` <= 4) OR ((`instrument_cqs_st` = 7) AND
(`instrument_cqs_lt` <= 4)))) AND (`residual_maturity_instrument` <= 12.0D))
THEN '/Debt/Central_Government_Issuer/Eligible/res_mat_1Y/' WHEN
((((`underlying_type` = 'DEBT') AND (`issuer_type` = 'CGVT')) AND
((`instrument_cqs_st` <= 4) OR ((`instrument_cqs_st` = 7) AND
(`instrument_cqs_lt` <= 4)))) AND (`residual_maturity_instrument` <= 60.0D))
THEN '/Debt/Central_Government_Issuer/Eligible/res_mat_5Y/' WHEN
(((`underlying_type` = 'DEBT') AND (`issuer_type` = 'CGVT')) AND
((`instrument_cqs_st` <= 4) OR ((`instrument_cqs_st` = 7) AND
(`instrument_cqs_lt` <= 4)))) THEN
'/Debt/Central_Government_Issuer/Eligible/res_mat_G5/' WHEN ((`underlying_type`
= 'DEBT') AND (`issuer_type` = 'CGVT')) THEN
'/Debt/Central_Government_Issuer/Non_Eligible/' WHEN ((((`underlying_type` =
'DEBT') AND (`issuer_type` IN ('INST', 'CORP', 'PSE', 'RGLA', 'IO_LISTED')))
AND ((`instrument_cqs_st` <= 3) OR ((`instrument_cqs_st` = 7) AND
(`instrument_cqs_lt` <= 3)))) AND (`residual_maturity_instrument` <= 12.0D))
THEN '/Debt/Other_Issuers/Eligible/res_mat_1Y/' WHEN ((((`underlying_type` =
'DEBT') AND (`issuer_type` IN ('INST', 'CORP', 'PSE', 'RGLA', 'IO_LISTED')))
AND ((`instrument_cqs_st` <= 3) OR ((`instrument_cqs_st` = 7) AND
(`instrument_cqs_lt` <= 3)))) AND (`residual_maturity_instrument` <= 60.0D))
THEN '/Debt/Other_Issuers/Eligible/res_mat_5Y/' WHEN (((`underlying_type` =
'DEBT') AND (`issuer_type` IN ('INST', 'CORP', 'PSE', 'RGLA', 'IO_LISTED')))
AND ((`instrument_cqs_st` <= 3) OR ((`instrument_cqs_st` = 7) AND
(`instrument_cqs_lt` <= 3)))) THEN '/Debt/Other_Issuers/Eligible/res_mat_G5/'
WHEN ((`underlying_type` = 'DEBT') AND (`issuer_type` IN ('INST', 'CORP',
'PSE', 'RGLA', 'IO_LISTED'))) THEN '/Debt/Other_Issuers/Non_Eligible/' WHEN
(((`underlying_type` = 'SECURITISATION') AND ((`instrument_cqs_st` <= 3) OR
((`instrument_cqs_st` = 7) AND (`instrument_cqs_lt` <= 3)))) AND
(`residual_maturity_instrument` <= 12.0D)) THEN
'/Securitisation/Eligible/res_mat_1Y/' WHEN (((`underlying_type` =
'SECURITISATION') AND ((`instrument_cqs_st` <= 3) OR ((`instrument_cqs_st` = 7)
AND (`instrument_cqs_lt` <= 3)))) AND (`residual_maturity_instrument` <=
60.0D)) THEN '/Securitisation/Eligible/res_mat_5Y/' WHEN ((`underlying_type` =
'SECURITISATION') AND ((`instrument_cqs_st` <= 3) OR ((`instrument_cqs_st` = 7)
AND (`instrument_cqs_lt` <= 3)))) THEN '/Securitisation/Eligible/res_mat_G5/'
WHEN (`underlying_type` = 'SECURITISATION') THEN
'/Securitisation/Non_Eligible/' WHEN ((`underlying_type` IN ('EQUITY',
'MAIN_INDEX_EQUITY', 'COMMODITY', 'NON_ELIGIBLE_SECURITY')) AND
(`underlying_type` = 'MAIN_INDEX_EQUITY')) THEN '/Other_Securities/Main_index/'
WHEN (`underlying_type` IN ('EQUITY', 'MAIN_INDEX_EQUITY', 'COMMODITY',
'NON_ELIGIBLE_SECURITY')) THEN '/Other_Securities/Others/' WHEN
(`underlying_type` = 'CASH') THEN '/Cash/' WHEN (`underlying_type` = 'GOLD')
THEN '/Gold/' WHEN (`underlying_type` = 'CIU') THEN '/CIU/' WHEN true THEN
'/Others/' END AS
`108_0___Portfolio_CRD4_Art_224_Volatility_Adjustments_Codes____path_CRD4_Art_224_Volatil`
was (Author: igreenfi):
the code that creates this plan is very complex.
I will try to reproduce it in simple code in the meanwhile I can attach the
generated code so you can see the problem is that the code does not create
functions and inline all the Plan into the processNext method.
[^generated_code.txt]
> proccesNext() failed to compile size is over 64kb
> -------------------------------------------------
>
> Key: SPARK-25094
> URL: https://issues.apache.org/jira/browse/SPARK-25094
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 2.4.0
> Reporter: Izek Greenfield
> Priority: Major
> Attachments: generated_code.txt
>
>
> I have this tree:
> 2018-08-12T07:14:31,289 WARN []
> org.apache.spark.sql.execution.WholeStageCodegenExec - Whole-stage codegen
> disabled for plan (id=1):
> *(1) Project [, ... 10 more fields]
> +- *(1) Filter NOT exposure_calc_method#10141 IN
> (UNSETTLED_TRANSACTIONS,FREE_DELIVERIES)
> +- InMemoryTableScan [, ... 11 more fields], [NOT
> exposure_calc_method#10141 IN (UNSETTLED_TRANSACTIONS,FREE_DELIVERIES)]
> +- InMemoryRelation [, ... 80 more fields], StorageLevel(memory,
> deserialized, 1 replicas)
> +- *(5) SortMergeJoin [unique_id#8506], [unique_id#8722], Inner
> :- *(2) Sort [unique_id#8506 ASC NULLS FIRST], false, 0
> : +- Exchange(coordinator id: 1456511137)
> UnknownPartitioning(9), coordinator[target post-shuffle partition size:
> 67108864]
> : +- *(1) Project [, ... 6 more fields]
> : +- *(1) Filter (((((isnotnull(v#49) &&
> isnotnull(run_id#52)) && (asof_date#48 <=> 17531)) && (run_id#52 = DATA_REG))
> && (v#49 = DATA_REG)) && isnotnull(unique_id#39))
> : +- InMemoryTableScan [, ... 6 more fields], [,
> ... 6 more fields]
> : +- InMemoryRelation [, ... 6 more
> fields], StorageLevel(memory, deserialized, 1 replicas)
> : +- *(1) FileScan csv [,... 6 more
> fields] , ... 6 more fields
> +- *(4) Sort [unique_id#8722 ASC NULLS FIRST], false, 0
> +- Exchange(coordinator id: 1456511137)
> UnknownPartitioning(9), coordinator[target post-shuffle partition size:
> 67108864]
> +- *(3) Project [, ... 74 more fields]
> +- *(3) Filter (((isnotnull(v#51) && (asof_date#42
> <=> 17531)) && (v#51 = DATA_REG)) && isnotnull(unique_id#54))
> +- InMemoryTableScan [, ... 74 more fields], [,
> ... 4 more fields]
> +- InMemoryRelation [, ... 74 more
> fields], StorageLevel(memory, deserialized, 1 replicas)
> +- *(1) FileScan csv [,... 74 more
> fields] , ... 6 more fields
> Compiling "GeneratedClass": Code of method "processNext()V" of class
> "org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage1"
> grows beyond 64 KB
> and the generated code failed to compile.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
