[MAPREDUCE-6737] HS: job history recovery fails with NumericFormatException if the job wasn't initted properly

• Type: Bug
• Status: Patch Available
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 2.7.0, 2.5.1
• Component/s: jobhistoryserver
• Assignee: Unassigned
• Reporter: Roman Gavryliuk
• Created: 20/Jul/16 12:55
• Updated: 10/Aug/16 17:50

Description

The problem shows itself while recovering old apps information:
2016-07-18 16:08:35,031 WARN
org.apache.hadoop.mapreduce.v2.jobhistory.FileNameIndexUtils: Unable to parse
start time from job history file
job_1468716177837_21790-1468845880296-username-applicationname-1468845889100-0-0-FAILED-root.queuename--1.jhist
: java.lang.NumberFormatException: For input string:
""

The problem is in JobHistoryEventHandler.java class in the
following part of code:

//initialize the launchTime in the JobIndexInfo of MetaInfo
if(event.getHistoryEvent().getEventType() == EventType.JOB_INITED ){
JobInitedEvent jie = (JobInitedEvent) event.getHistoryEvent();
mi.getJobIndexInfo().setJobStartTime(jie.getLaunchTime());

Because of job was not initialized properly, the 'if' statement takes value
'false' and .setJobStartTime() is not called.

In JobIndexInfo constructor, we have a default value for jobStartTime:
this.jobStartTime = -1;

When history server recovers any application's info, it passes all parameters
to array of strings jobDetails:
String[] jobDetails = fileName.split(DELIMITER);

Please note, DELIMETER is initialized in the following way:
static final String DELIMITER = "-";

So, jobDetails array has 10 elements - job ID, submit time, username, job name,
finish time, number of maps, number of reducers, job status, queue, and start
time).
If jobStartTime = -1, the minus symbol is considered as delimeter and the code
will assign an empty string "" as a value for 9-th element in jobDetails array.

In org.apache.hadoop.mapreduce.v2.jobhistory.FileNameIndexUtils class a
NumberFormatException will appear while trying to parse empty string to long
type.
Long.parseLong(decodeJobHistoryFileName(jobDetails[JOB_START_TIME_INDEX])));

The most simple fix is to change the value this.jobStartTime to 0 in
JobIndexInfo constructor.