[FLINK-19630] Sink data in [ORC] format into Hive By using Legacy Table API caused unexpected Exception

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.11.2
• Fix Version/s: 1.12.0
• Component/s: Documentation, Table SQL / Ecosystem
• Labels: pull-request-available
• Assignee: Rui Li
• Reporter: Lsw_aka_laplace
• Created: 14/Oct/20 03:20
• Updated: 27/Apr/21 18:05
• Resolved: 24/Nov/20 02:36

Description

ENV:

Flink version 1.11.2

Hive exec version: 2.0.1

Hive file storing type ：ORC

SQL or Datastream: SQL API

Kafka Connector :  custom Kafka connector which is based on Legacy API (TableSource/`org.apache.flink.types.Row`)

Hive Connector : totally follows the Flink-Hive-connector (we only made some encapsulation upon it)

Using StreamingFileCommitter：YES

 

 

Description:

   try to execute the following SQL:

    """

      insert into hive_table (select * from kafka_table)

    """

   HIVE Table SQL seems like:

    """

CREATE TABLE `hive_table`(
 // some fields
PARTITIONED BY (
`dt` string,
`hour` string)
STORED AS orc
TBLPROPERTIES (
'orc.compress'='SNAPPY',
'type'='HIVE',
'sink.partition-commit.trigger'='process-time',
'sink.partition-commit.delay' = '1 h',
'sink.partition-commit.policy.kind' = 'metastore,success-file',
)   

   """

When this job starts to process snapshot, here comes the weird exception:

As we can see from the message：Owner thread shall be the [Legacy Source Thread], but actually the streamTaskThread which represents the whole first stage is found. 

So I checked the Thread dump at once.

                                                                     The legacy Source Thread

 

                                                               The StreamTask Thread

 

   According to the thread dump info and the Exception Message, I searched and read certain source code and then DID A TEST

 

   Since the Kafka connector is customed, I tried to make the KafkaSource a serpated stage by changing the TaskChainStrategy to Never. The task topology as follows:

 

Fortunately, it did work! No Exception is throwed and Checkpoint could be snapshot successfully!

 

 

So, from my perspective, there shall be something wrong when HiveWritingTask and  LegacySourceTask chained together. the Legacy source task is a seperated thread, which may be the cause of the exception mentioned above.

 

                                                                

 

Issue Links

• is related to: FLINK-13998 Fix ORC test failure with Hive 2.0.x
• links to: GitHub Pull Request #14131