[MAPREDUCE-5055] Binary compatibility for the method Reporter.incrCounter(String group, String counter, long amount) between hadoop-1.x and hadoop-2.x

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Not A Problem
• Affects Version/s: 2.0.3-alpha
• Component/s: mrv2

Environment

Linux

• Assignee: Zhijie Shen
• Reporter: Deepesh Khandelwal
• Created: 08/Mar/13 18:48
• Updated: 09/Mar/13 00:08
• Resolved: 09/Mar/13 00:08

Description

Try to generate a lot of counters that exceed the default max limit so the job errors.
To simulate this edit org.apache.hadoop.examples.RandomWriter class and add the following in the mapper:

        for (int i = 0; i < 150; i++) {
          reporter.incrCounter("StringCounter", "counter" + i, 1L);
        }

Attached is the modified code of RandomWriter.java from branch-1.

When I run it against branch-1,

hadoop jar build/hadoop-examples-1.2.0-SNAPSHOT.jar -Dtest.randomwrite.total_bytes=8 rw2

it fails with the following error:

java.lang.Exception: org.apache.hadoop.mapred.Counters$CountersExceededException: Error: Exceeded limits on number of counters - Counters=120 Limit=120

Using the jar compiled with branch-1 when running in a hadoop 2 cluster completes without any error and don't see the counter. Interestingly there were two more custom counters in the same job that are incremented and they are visible. The difference is that they use the method
Reporter.incrCounter(Enum<?> key, long amount).

Compiled this against 2.x/trunk and ran it. The job did fail with the expected counter exceeded exception. So it seems to be a case of binary incompatibility between 1.x and 2.x.

Issue Links

• is related to: MAPREDUCE-3697 Hadoop Counters API limits Oozie's working across different hadoop versions