[
https://issues.apache.org/jira/browse/SPARK-15112?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15273837#comment-15273837
]
Cheng Lian edited comment on SPARK-15112 at 5/6/16 10:22 AM:
-------------------------------------------------------------
Actually there's another issue that contributes to this bug. The problem lies
in the {{EmbedSerializerInFilter}} optimization rule. In short, this rule
optimizes plan fragments like
{noformat}
SerializeFromObject <--.
Filter | column order may differ
DeserializeToObject |
<child-plan> <--'
{noformat}
into
{noformat}
Filter
<child-plan>
{noformat}
by embedding the deserializer expression into the {{Filter}} condition
expression. Namely, when filtering an input row, the new {{Filter}} operator
always deserializes the input row into a Scala object, and then uses the object
as argument to invoke the user-provided Scala predicate function.
The problem here is that, output column order of {{SerializeFromObject}} may
differ from column order of the child plan (as explained in my comment above).
Thus the simplified plan fragment may produce wrong result because the column
order isn't adjusted accordingly.
To fix this issue, we should add a {{Project}} on top of the result {{Filter}}
plan when necessary to adjust output column order.
was (Author: lian cheng):
Actually there's another issue that contributes to this bug. The problem lies
in the {{EmbedSerializerInFilter}} optimization rule. In short, this rule
optimizes plan fragments like
{noformat}
SerializeFromObject <--.
Filter | column order may differ
DeserializeToObject |
<child-plan> <--'
{noformat}
into
{noformat}
Filter
<child-plan>
{noformat}
by embedding the deserializer expression into the {{Filter}} condition
expression. Namely, when filtering an input row, always deserialize the input
row into a Scala object, then uses the object as argument to invoke the
user-provided Scala predicate function.
The problem here is that, output column order of {{SerializeFromObject}} may
differ from column order of the child plan (as explained in my comment above).
Thus the simplified plan fragment may produce wrong result because the column
order isn't adjusted accordingly.
To fix this issue, we should add a {{Project}} on top of the result {{Filter}}
plan when necessary to adjust output column order.
> Dataset filter returns garbage
> ------------------------------
>
> Key: SPARK-15112
> URL: https://issues.apache.org/jira/browse/SPARK-15112
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Reporter: Reynold Xin
> Assignee: Cheng Lian
> Priority: Blocker
> Attachments: demo 1 dataset - Databricks.htm
>
>
> See the following notebook:
> https://databricks-prod-cloudfront.cloud.databricks.com/public/4027ec902e239c93eaaa8714f173bcfc/6122906529858466/2727501386611535/5382278320999420/latest.html
> I think it happens only when using JSON. I'm also going to attach it to the
> ticket.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
