[OOZIE-2802] Spark action failure on Spark 2.1.0 due to duplicate sharelibs

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 5.0.0b1, 4.3.1
• Assignee: Gézapeti
• Reporter: Satish Saley
• Created: 17/Feb/17 00:01
• Updated: 25/Jan/18 21:00
• Resolved: 03/Mar/17 00:58

Description

There are some same jars in launcher lib and sharelib. These jars are causing failures in spark action for spark 2.1.0. SPARK-18099 added an exception to be thrown if same file gets added multiple times to distributed cache.

$ hdfs dfs -ls /user/saley/share/lib/launcher_20170216150752/oozie
17/02/16 15:14:22 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Found 3 items
-rwxr-xr-x   1 saley supergroup      11871 2017-02-16 15:07 /user/saley/share/lib/launcher_20170216150752/oozie/oozie-hadoop-utils-hadoop-2-4.4.0-SNAPSHOT.jar
-rwxr-xr-x   1 saley supergroup      17514 2017-02-16 15:07 /user/saley/share/lib/launcher_20170216150752/oozie/oozie-sharelib-hcatalog-4.4.0-SNAPSHOT.jar
-rwxr-xr-x   1 saley supergroup      55092 2017-02-16 15:07 /user/saley/share/lib/launcher_20170216150752/oozie/oozie-sharelib-oozie-4.4.0-SNAPSHOT.jar



$ hdfs dfs -ls /user/saley/share/lib/lib_20170216150706/oozie
17/02/16 15:14:39 WARN util.NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Found 3 items
-rw-r--r--   1 saley supergroup      16046 2017-02-16 15:07 /user/saley/share/lib/lib_20170216150706/oozie/json-simple-1.1.jar
-rw-r--r--   1 saley supergroup      11871 2017-02-16 15:07 /user/saley/share/lib/lib_20170216150706/oozie/oozie-hadoop-utils-hadoop-2-4.4.0-SNAPSHOT.jar
-rw-r--r--   1 saley supergroup      55092 2017-02-16 15:07 /user/saley/share/lib/lib_20170216150706/oozie/oozie-sharelib-oozie-4.4.0-SNAPSHOT.jar

The file with same name can't be in twice in any of --files , --archives, --jars (ie one file in both --files and --jars) and can't be in --files twice and can't be in --archives twice. It is allowed to be in --jars twice.

Possible solutions:
1. While specifying the --files options, filter out the duplicate jars
2. Separate out jars and files. Use --jars options to distribute jars and --files for files. Also remember to remove duplicates from --files, and make sure that there is nothing common in --jars and --files.

Issue Links

• supercedes: OOZIE-2806 Don't overwrite --files and --archives in Spark action