[
https://issues.apache.org/jira/browse/HIVE-20644?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16637463#comment-16637463
]
Thejas M Nair commented on HIVE-20644:
--------------------------------------
For completeness, maybe make similar update in
ql/src/java/org/apache/hadoop/hive/ql/exec/spark/SparkReduceRecordHandler.java
as well ?
Also, DEBUG level is often used for debugging variety of problems, and people
can set entire HS2 logging at that level. Sometimes production systems can be
put in DEBUG level logging for sometime while troubleshooting an issue. AFAIK,
debug level messages in HS2 would be sent to beeline if HS2 is running in debug
mode and user has set hive.server2.logging.operation.level=verbose. (Since this
error is happening in the tasks, its possible that this doesn't get sent
currently. But again, that can change in future)
I think its unusual to have production systems running with TRACE level logs,
so maybe log the rowString part only in trace level logging ? I would expect
only specific classes to be enabled at trace level logging, so that would be a
safer option IMO.
> Avoid exposing sensitive infomation through a Hive Runtime exception
> --------------------------------------------------------------------
>
> Key: HIVE-20644
> URL: https://issues.apache.org/jira/browse/HIVE-20644
> Project: Hive
> Issue Type: Improvement
> Components: HiveServer2
> Affects Versions: 3.1.0
> Reporter: Ashutosh Bapat
> Assignee: Ashutosh Bapat
> Priority: Minor
> Labels: pull-request-available
> Fix For: 3.1.0
>
> Attachments: HIVE-20644.01, HIVE-20644.02, HIVE-20644.03
>
>
> The HiveException raised from the following methods is exposing the datarow
> the caused the run time exception.
> # ReduceRecordSource::GroupIterator::next() - around line 372
> # MapOperator::process() - around line 567
> # ExecReducer::reduce() - around line 243
> In all the cases, a string representation of the row is constructed on the
> fly and is included in
> the error message.
> VectorMapOperator::process() - around line 973 raises the same exception but
> it's not exposing the row since the row contents are not included in the
> error message.
> While trying to reproduce above error, I also found that the arguments to a
> UDF get exposed in log messages from FunctionRegistry::invoke() around line
> 1114. This too can cause sensitive information to be leaked through error
> message.
> This way some sensitive information is leaked to a user through exception
> message. That information may not be available to the user otherwise. Hence
> it's a kind of security breach or violation of access control.
> The contents of the row or the arguments to a function may be useful for
> debugging and hence it's worth to add those to logs. Hence proposal here to
> log a separate message with log level DEBUG or INFO containing the string
> representation of the row. Users can configure their logging so that
> DEBUG/INFO messages do not go to the client but at the same time are
> available in the hive server logs for debugging. The actual exception message
> will not contain any sensitive data like row data or argument data.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
