[
https://issues.apache.org/jira/browse/HIVE-27071?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17688037#comment-17688037
]
Sai Hemanth Gantasala commented on HIVE-27071:
----------------------------------------------
[~ayushtkn] - Thanks for providing the patch in a quick time. I have reviewed
the PR. I'll assign this Jira to you since you are the author of this patch.
[~tarak271] - I'm not sure what other files we want to skip other than the
names starting with "_" or ".". Can you suggest any other files? Maybe Ayush
has a better idea if it is good to skip those unwanted files.
> Select query with LIMIT clause can fail if there are marker files like
> "_SUCCESS" and "_MANIFEST"
> -------------------------------------------------------------------------------------------------
>
> Key: HIVE-27071
> URL: https://issues.apache.org/jira/browse/HIVE-27071
> Project: Hive
> Issue Type: Bug
> Components: HiveServer2
> Affects Versions: 4.0.0
> Reporter: Sai Hemanth Gantasala
> Priority: Major
> Labels: pull-request-available
> Time Spent: 20m
> Remaining Estimate: 0h
>
> Spark clients creates marker files like "_SUCCESS" and "_MANIFEST" under the
> table/partition path at the end of a write operation. For example
> 'hdfs://name-node-host/table/partition/_SUCCESS'
> Whenever Hive is trying to read that table with the LIMIT clause, it could to
> the following error:
> {code:java}
> ERROR : Vertex failed, vertexName=Map 1,
> vertexId=vertex_1676095298574_0017_2_00, diagnostics=[Vertex
> vertex_1676095298574_0017_2_00 [Map 1] killed/failed due
> to:ROOT_INPUT_INIT_FAILURE, Vertex Input: trade initializer failed,
> vertex=vertex_1676095298574_0017_2_00 [Map 1],
> org.apache.hadoop.mapred.InvalidInputException: Input path does not exist:
> hdfs://name-node-host/table/partition/_MANIFEST
> Input path does not exist: hdfs://name-node-host/table/partition/_SUCCESS at
> org.apache.hadoop.mapred.FileInputFormat.singleThreadedListStatus(FileInputFormat.java:300)
> at
> org.apache.hadoop.mapred.FileInputFormat.listStatus(FileInputFormat.java:240)
> at
> org.apache.hadoop.mapred.FileInputFormat.getSplits(FileInputFormat.java:328)
> at
> org.apache.hadoop.hive.ql.io.HiveInputFormat.addSplitsForGroup(HiveInputFormat.java:579)
> {code}
> Hive execution engine should ignore these marker files while reading the
> table/partition data.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
