Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
0.7.0
-
spark2.4.5, hadoop3.1.1, hive 3.1.1
Description
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
2021-03-25 20:23:14,014 | INFO | AsyncDispatcher event handler | Diagnostics report from attempt_1615883368881_0038_m_000000_0: Error: java.lang.NullPointerException2021-03-25 20:23:14,014 | INFO | AsyncDispatcher event handler | Diagnostics report from attempt_1615883368881_0038_m_000000_0: Error: java.lang.NullPointerException at org.apache.hudi.hadoop.realtime.RealtimeCompactedRecordReader.next(RealtimeCompactedRecordReader.java:101) 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.hudi.hadoop.realtime.RealtimeCompactedRecordReader.next(RealtimeCompactedRecordReader.java:92) 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.HoodieCombineRealtimeRecordReader.next(HoodieCombineRealtimeRecordReader.java:68) at org.apache.hudi.hadoop.realtime.HoodieCombineRealtimeRecordReader.next(HoodieCombineRealtimeRecordReader.java:77) at org.apache.hudi.hadoop.realtime.HoodieCombineRealtimeRecordReader.next(HoodieCombineRealtimeRecordReader.java:42) at org.apache.hadoop.mapred.MapTask$TrackedRecordReader.moveToNext(MapTask.java:205) at org.apache.hadoop.mapred.MapTask$TrackedRecordReader.next(MapTask.java:191) at org.apache.hadoop.mapred.MapRunner.run(MapRunner.java:52) at org.apache.hadoop.hive.ql.exec.mr.ExecMapRunner.run(ExecMapRunner.java:37) at org.apache.hadoop.mapred.MapTask.runOldMapper(MapTask.java:465) at org.apache.hadoop.mapred.MapTask.run(MapTask.java:349) at org.apache.hadoop.mapred.YarnChild$1.run(YarnChild.java:183) at java.security.AccessController.doPrivileged(Native Method) at javax.security.auth.Subject.doAs(Subject.java:422) at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1761) at org.apache.hadoop.mapred.YarnChild.main(YarnChild.java:177)
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
sparksql can avoid this problem by set spark.hadoop.cloneConf=true which is not recommended in spark， however hive has no way to avoid this problem。
test step：
step1:
val df = spark.range(0, 100000).toDF("keyid")
.withColumn("col3", expr("keyid"))
.withColumn("p", lit(0))
.withColumn("p1", lit(0))
.withColumn("p2", lit(7))
.withColumn("a1", lit(Array[String] ("sb1", "rz")))
.withColumn("a2", lit(Array[String] ("sb1", "rz")))
// create hoodie table hive_14b
merge(df, 4, "default", "hive_14b", DataSourceWriteOptions.MOR_TABLE_TYPE_OPT_VAL, op = "bulk_insert")
notice: bulk_insert will produce 4 files in hoodie table
step2:
val df = spark.range(99999, 100002).toDF("keyid")
.withColumn("col3", expr("keyid"))
.withColumn("p", lit(0))
.withColumn("p1", lit(0))
.withColumn("p2", lit(7))
.withColumn("a1", lit(Array[String] ("sb1", "rz")))
.withColumn("a2", lit(Array[String] ("sb1", "rz")))
// upsert table
merge(df, 4, "default", "hive_14b", DataSourceWriteOptions.MOR_TABLE_TYPE_OPT_VAL, op = "upsert")
now : we have four base files and one log file in hoodie table
step3:
spark-sql/beeline:
select count(col3) from hive_14b_rt;
then the query failed.
Attachments
Issue Links
- links to