[
https://issues.apache.org/jira/browse/HIVE-12779?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15307573#comment-15307573
]
Oleksiy Sayankin commented on HIVE-12779:
-----------------------------------------
Alina has found a workaround for this issue.
*ROOT-CAUSE:*
Consider method
{code}
protected int require (int required) throws KryoException
{code}
from class com.esotericsoftware.kryo.io.Input where exception happens.
{code}
int remaining = limit - position;
if (remaining >= required) return remaining;
if (required > capacity) throw new KryoException("Buffer too small: capacity:
" + capacity + ", required: " + required);
int count;
// Try to fill the buffer.
if (remaining > 0) {
count = fill(buffer, limit, capacity - limit);
if (count == -1) throw new KryoException("Buffer underflow.");
{code}
We can see that exception ("Buffer underflow.") occurs when count == -1. So let
us see method fill(byte[] buffer, int offset, int count) in details:
{code}
if (inputStream == null) return -1;
try {
return inputStream.read(buffer, offset, count);
} catch (IOException ex) {
throw new KryoException(ex);
}
{code}
It returns -1 either when inputStream == null or from inputStream.read(buffer,
offset, count). We definitely know that inputStream can not be equal null here
because of constructor:
{code}
public Input (InputStream inputStream) {
this(4096);
if (inputStream == null) throw new IllegalArgumentException("inputStream
cannot be null.");
this.inputStream = inputStream;
}
{code}
>From Java docs we know that if no byte is available because the stream is at
>end of file, the value -1 is returned by the method inputStream.read(buffer,
>offset, count). Hence we suspect here some errors in HDFS here that causes -1
>to be a return value. Skipping usage of file system as query plan storage and
>sending it via RPC directly will fix the issue.
*SOLUTION:*
Use
{code}
<property>
<name>hive.rpc.query.plan</name>
<value>true</value>
</property>
{code}
in hive-site.xml as workaround. This property defines whether to send the query
plan via local resource or RPC.
> Buffer underflow when inserting data to table
> ---------------------------------------------
>
> Key: HIVE-12779
> URL: https://issues.apache.org/jira/browse/HIVE-12779
> Project: Hive
> Issue Type: Bug
> Components: Database/Schema, SQL
> Environment: CDH 5.4.9
> Reporter: Ming Hsuan Tu
> Assignee: Alan Gates
>
> I face a buffer underflow problem when inserting data to table from hive
> 1.1.0.
> the block size is 128 MB and the data size is only 10MB, but it gives me 891
> mappers.
> Task with the most failures(4):
> -----
> Task ID:
> task_1451989578563_0001_m_000008
> URL:
>
> http://0.0.0.0:8088/taskdetails.jsp?jobid=job_1451989578563_0001&tipid=task_1451989578563_0001_m_000008
> -----
> Diagnostic Messages for this Task:
> Error: java.lang.RuntimeException: Failed to load plan:
> hdfs://tpe-nn-3-1:8020/tmp/hive/alec.tu/af798488-dbf5-45da-8adb-e4f2ddde1242/hive_2016-01-05_18-34-26_864_3947114301988950007-1/-mr-10004/bb86c923-0dca-43cd-aa5d-ef575d764e06/map.xml:
> org.apache.hive.com.esotericsoftware.kryo.KryoException: Buffer underflow.
> at
> org.apache.hadoop.hive.ql.exec.Utilities.getBaseWork(Utilities.java:450)
> at
> org.apache.hadoop.hive.ql.exec.Utilities.getMapWork(Utilities.java:296)
> at
> org.apache.hadoop.hive.ql.io.HiveInputFormat.init(HiveInputFormat.java:268)
> at
> org.apache.hadoop.hive.ql.io.HiveInputFormat.getRecordReader(HiveInputFormat.java:234)
> at
> org.apache.hadoop.hive.ql.io.CombineHiveInputFormat.getRecordReader(CombineHiveInputFormat.java:701)
> at
> org.apache.hadoop.mapred.MapTask$TrackedRecordReader.<init>(MapTask.java:169)
> at org.apache.hadoop.mapred.MapTask.runOldMapper(MapTask.java:432)
> at org.apache.hadoop.mapred.MapTask.run(MapTask.java:343)
> at org.apache.hadoop.mapred.YarnChild$2.run(YarnChild.java:163)
> at java.security.AccessController.doPrivileged(Native Method)
> at javax.security.auth.Subject.doAs(Subject.java:415)
> at
> org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1671)
> at org.apache.hadoop.mapred.YarnChild.main(YarnChild.java:158)
> Caused by: org.apache.hive.com.esotericsoftware.kryo.KryoException: Buffer
> underflow.
> at
> org.apache.hive.com.esotericsoftware.kryo.io.Input.require(Input.java:181)
> at
> org.apache.hive.com.esotericsoftware.kryo.io.Input.readBoolean(Input.java:783)
> at
> org.apache.hive.com.esotericsoftware.kryo.serializers.UnsafeCacheFields$UnsafeBooleanField.read(UnsafeCacheFields.java:120)
> at
> org.apache.hive.com.esotericsoftware.kryo.serializers.FieldSerializer.read(FieldSerializer.java:507)
> at
> org.apache.hive.com.esotericsoftware.kryo.Kryo.readObject(Kryo.java:672)
> at
> org.apache.hadoop.hive.ql.exec.Utilities.deserializeObjectByKryo(Utilities.java:1069)
> at
> org.apache.hadoop.hive.ql.exec.Utilities.deserializePlan(Utilities.java:960)
> at
> org.apache.hadoop.hive.ql.exec.Utilities.deserializePlan(Utilities.java:974)
> at
> org.apache.hadoop.hive.ql.exec.Utilities.getBaseWork(Utilities.java:416)
> ... 12 more
> Container killed by the ApplicationMaster.
> Container killed on request. Exit code is 143
> Container exited with a non-zero exit code 143
> Thank you.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
