[
https://issues.apache.org/jira/browse/SPARK-17403?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16682591#comment-16682591
]
Eyal Farago commented on SPARK-17403:
-------------------------------------
[~paul_lysak], [~hvanhovell],
please notice the exception happens when building the column buffers, hence the
corrupt string comes from upstream - in the repro code this is the join, not
necessarily an in-mem-relation.
[~hvanhovell], I'd suspect that the upstream operator somehow overwrites the
row/buffer returned from the Iterator.next method while the downstream code
(in-mem column building code) doesn't take the necessary precautions (copy). if
this is the case I'd suspect the issue is somehow related to code generation
either in the join or InMemoryTableScanExec operator.
another option, [~paul_lysak] mentioned that the problem happens when the
application is 'big enough', given that DataSet's default persistence level is
_memory and disk_, is it possible that spark starts evacuating these blocks
from memory and then bad things start happening?
> Fatal Error: Scan cached strings
> --------------------------------
>
> Key: SPARK-17403
> URL: https://issues.apache.org/jira/browse/SPARK-17403
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 2.0.0
> Environment: Spark standalone cluster (3 Workers, 47 cores)
> Ubuntu 14
> Java 8
> Reporter: Ruben Hernando
> Priority: Major
>
> The process creates views from JDBC (SQL server) source and combines them to
> create other views.
> Finally it dumps results via JDBC
> Error:
> {quote} # JRE version: Java(TM) SE Runtime Environment (8.0_101-b13) (build
> 1.8.0_101-b13)
> # Java VM: Java HotSpot(TM) 64-Bit Server VM (25.101-b13 mixed mode
> linux-amd64 )
> # Problematic frame:
> # J 4895 C1 org.apache.spark.unsafe.Platform.getLong(Ljava/lang/Object;J)J
> (9 bytes) @ 0x00007fbb355dfd6c [0x00007fbb355dfd60+0xc]
> #{quote}
> SQL Query plan (fields truncated):
> {noformat}
> == Parsed Logical Plan ==
> 'Project [*]
> +- 'UnresolvedRelation `COEQ_63`
> == Analyzed Logical Plan ==
> InstanceId: bigint, price: double, ZoneId: int, priceItemId: int, priceId: int
> Project [InstanceId#20236L, price#20237, ZoneId#20239, priceItemId#20242,
> priceId#20244]
> +- SubqueryAlias coeq_63
> +- Project [_TableSL_SID#143L AS InstanceId#20236L, SL_RD_ColR_N#189 AS
> price#20237, 24 AS ZoneId#20239, 6 AS priceItemId#20242, 63 AS priceId#20244]
> +- SubqueryAlias 6__input
> +-
> Relation[_TableSL_SID#143L,_TableP_DC_SID#144L,_TableSH_SID#145L,ID#146,Name#147,TableP_DCID#148,TableSHID#149,SL_ACT_GI_DTE#150,SL_Xcl_C#151,SL_Xcl_C#152,SL_Css_Cojs#153L,SL_Config#154,SL_CREATEDON#
> .......... 36 more fields] JDBCRelation((select [SLTables].[_TableSL_SID],
> [SLTables]. ... [...] FROM [sch].[SLTables] [SLTables] JOIN sch.TPSLTables
> TPSLTables ON [TPSLTables].[_TableSL_SID] = [SLTables].[_TableSL_SID] where
> _TP = 24) input)
> def pers
> == Optimized Logical Plan ==
> Project [_TableSL_SID#143L AS InstanceId#20236L, SL_RD_ColR_N#189 AS
> price#20237, 24 AS ZoneId#20239, 6 AS priceItemId#20242, 63 AS priceId#20244]
> +- InMemoryRelation [_TableSL_SID#143L, _TableP_DC_SID#144L,
> _TableSH_SID#145L, ID#146, Name#147, ... 36 more fields], true, 10000,
> StorageLevel(disk, memory, deserialized, 1 replicas)
> : +- *Scan JDBCRelation((select [SLTables].[_TableSL_SID],
> [SLTables].[_TableP_DC_SID], [SLTables].[_TableSH_SID], [SLTables].[ID],
> [SLTables].[Name], [SLTables].[TableP_DCID], [SLTables].[TableSHID],
> [TPSLTables].[SL_ACT_GI_DTE], ... [...] FROM [sch].[SLTables] [SLTables]
> JOIN sch.TPSLTables TPSLTables ON [TPSLTables].[_TableSL_SID] =
> [SLTables].[_TableSL_SID] where _TP = 24) input)
> [_TableSL_SID#143L,_TableP_DC_SID#144L,_TableSH_SID#145L,ID#146,Name#147,TableP_DCID#148,TableSHID#149,SL_ACT_GI_DTE#150,SL_Xcl_C#151,...
> 36 more fields]
> == Physical Plan ==
> *Project [_TableSL_SID#143L AS InstanceId#20236L, SL_RD_ColR_N#189 AS
> price#20237, 24 AS ZoneId#20239, 6 AS priceItemId#20242, 63 AS priceId#20244]
> +- InMemoryTableScan [_TableSL_SID#143L, SL_RD_ColR_N#189]
> : +- InMemoryRelation [_TableSL_SID#143L, _TableP_DC_SID#144L,
> _TableSH_SID#145L, ID#146, Name#147, ... 36 more fields], true, 10000,
> StorageLevel(disk, memory, deserialized, 1 replicas)
> : : +- *Scan JDBCRelation((select [SLTables].[_TableSL_SID],
> [SLTables].[_TableP_DC_SID], [SLTables].[_TableSH_SID], [SLTables].[ID],
> [SLTables].[Name], [SLTables].[TableP_DCID], ... [...] FROM [sch].[SLTables]
> [SLTables] JOIN sch.TPSLTables TPSLTables ON [TPSLTables].[_TableSL_SID] =
> [SLTables].[_TableSL_SID] where _TP = 24) input)
> [_TableSL_SID#143L,_TableP_DC_SID#144L,_TableSH_SID#145L,ID#146,Name#147,,...
> 36 more fields]
> {noformat}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
