[DRILL-3826] Concurrent Query Submission leads to Channel Closed Exception

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.1.0, 1.2.0
• Fix Version/s: 1.7.0
• Component/s: Client - JDBC, Execution - RPC
• Labels: filesystem hadoop hbase jdbc rpc

Environment

• CentOS release 6.6 (Final)
• hadoop-2.7.1
• hbase-1.0.1.1
• drill-1.1.0
• jdk-1.8.0_45

• External issue URL: https://issues.apache.org/jira/browse/DRILL-3926
• External issue ID: DRILL-3926
• Assignee: Abdel Hakim Deneche
• Reporter: Yiyi Hu
• Reviewer: Rahul Kumar Challapalli
• Created: 23/Sep/15 11:12
• Updated: 23/Aug/16 23:17
• Resolved: 10/May/16 20:32

Description

Frequently seen CHANNEL CLOSED EXCEPTION while running concurrent quries with relatively large LIMIT.

Here are the details,

SET UP:

• Single drillbit running on a single zookeeper node
• 4G heap size, 8G direct memory
• Storage plugins: local filesystem, hdfs, hbase

TEST DATA:

• A 50,000,000 records json file test.json, with two fields id<Int>, title<String> (approximately 3G).

SHELL TEST:

• Running 4 drill shells concurrently with query:
  SELECT id, title from dfs.`test.json` LIMIT 5000000.

• Queries got canceled. Channel closing between client and server were seen randomly, as an example shown below:

java.lang.RuntimeException: java.sql.SQLException: SYSTEM ERROR: ChannelClosedException: Channel closed /192.168.4.201:31010 <--> /192.168.4.201:48829.

Fragment 0:0

[Error Id: 0bd2b500-155e-46e0-9f26-bd89fea47a25 on TEST-101:31010]
	at sqlline.IncrementalRows.hasNext(IncrementalRows.java:73)
	at sqlline.TableOutputFormat$ResizingRowsProvider.next(TableOutputFormat.java:87)
	at sqlline.TableOutputFormat.print(TableOutputFormat.java:118)
	at sqlline.SqlLine.print(SqlLine.java:1583)
	at sqlline.Commands.execute(Commands.java:852)
	at sqlline.Commands.sql(Commands.java:751)
	at sqlline.SqlLine.dispatch(SqlLine.java:738)
	at sqlline.SqlLine.begin(SqlLine.java:612)
	at sqlline.SqlLine.start(SqlLine.java:366)
	at sqlline.SqlLine.main(SqlLine.java:259)

JDBC TEST:

• 6 separate threads running the same query: SELECT id, title from dfs.`test.json` LIMIT 10000000, each maintains its own connection. ResultSet, statement and connection are closed finally.

• Throws the same channel closed exception randomly. Log file were enclosed for review.

• Memory usage was monitored, all good.

CROSS STORAGE PLUGINS:

• The same issue can be found not only in JSON on a file system (local/hdfs), but also in HBASE.

• The issue was not found in a single thread application.

Issue Links

• relates to: DRILL-3763 Cancel (Ctrl-C) one of concurrent queries results in ChannelClosedException