tao meng created HUDI-1722:
------------------------------
Summary: hive beeline/spark-sql query specified field on mor
table occur NPE
Key: HUDI-1722
URL: https://issues.apache.org/jira/browse/HUDI-1722
Project: Apache Hudi
Issue Type: Bug
Components: Hive Integration, Spark Integration
Affects Versions: 0.7.0
Environment: spark2.4.5, hadoop3.1.1, hive 3.1.1
Reporter: tao meng
Fix For: 0.9.0
HUDI-892 introduce this problem。
this pr skip adding projection columns if there are no log files in the
hoodieRealtimeSplit。 but this pr donnot consider that multiple getRecordReaders
share same jobConf。
Consider the following questions：
we have four getRecordReaders:
reader1(its hoodieRealtimeSplit contains no log files)
reader2 (its hoodieRealtimeSplit contains log files)
reader3(its hoodieRealtimeSplit contains log files)
reader4(its hoodieRealtimeSplit contains no log files)
now reader1 run first, HoodieInputFormatUtils.HOODIE_READ_COLUMNS_PROP in
jobConf will be set to be true, and no hoodie additional projection columns
will be added to jobConf （see
HoodieParquetRealtimeInputFormat.addProjectionToJobConf）
reader2 run later, since HoodieInputFormatUtils.HOODIE_READ_COLUMNS_PROP in
jobConf is set to be true, no hoodie additional projection columns will be
added to jobConf. （see HoodieParquetRealtimeInputFormat.addProjectionToJobConf）
which lead to the result that _hoodie_record_key would be missing and merge
step would throw exceptions
Caused by: java.io.IOException: java.lang.NullPointerException
at
org.apache.hadoop.hive.ql.exec.FetchOperator.getNextRow(FetchOperator.java:611)
at org.apache.hadoop.hive.ql.exec.FetchOperator.pushRow(FetchOperator.java:518)
at org.apache.hadoop.hive.ql.exec.FetchTask.fetch(FetchTask.java:150)
at org.apache.hadoop.hive.ql.Driver.getResults(Driver.java:3296)
at
org.apache.hadoop.hive.ql.reexec.ReExecDriver.getResults(ReExecDriver.java:252)
at
org.apache.hive.service.cli.operation.SQLOperation.getNextRowSet(SQLOperation.java:522)
... 24 more
Caused by: java.lang.NullPointerException
at
org.apache.hudi.hadoop.realtime.RealtimeCompactedRecordReader.next(RealtimeCompactedRecordReader.java:93)
at
org.apache.hudi.hadoop.realtime.RealtimeCompactedRecordReader.next(RealtimeCompactedRecordReader.java:43)
at
org.apache.hudi.hadoop.realtime.HoodieRealtimeRecordReader.next(HoodieRealtimeRecordReader.java:79)
at
org.apache.hudi.hadoop.realtime.HoodieRealtimeRecordReader.next(HoodieRealtimeRecordReader.java:36)
at
org.apache.hadoop.hive.ql.exec.FetchOperator.getNextRow(FetchOperator.java:578)
at org.apache.hadoop.hive.ql.exec.FetchOperator.pushRow(FetchOperator.java:518)
at org.apache.hadoop.hive.ql.exec.FetchTask.fetch(FetchTask.java:150)
at org.apache.hadoop.hive.ql.Driver.getResults(Driver.java:3296)
at
org.apache.hadoop.hive.ql.reexec.ReExecDriver.getResults(ReExecDriver.java:252)
at
org.apache.hive.service.cli.operation.SQLOperation.getNextRowSet(SQLOperation.java:522)
Obviously, this is an occasional problem。 if reader2 run first, hoodie
additional projection columns will be added to jobConf and in this case the
query will be ok
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
