[OOZIE-1911] SLA calculation in HA mode does wrong bit comparison for 'start' and 'duration'

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: trunk
• Fix Version/s: 4.1.0
• Assignee: Mona Chitnis
• Reporter: Mona Chitnis
• Created: 01/Jul/14 22:14
• Updated: 09/Dec/14 00:45
• Resolved: 15/Jul/14 23:50

Description

In chronological order:

Server 1:
Job's SLA eventProcessed set to 0101 => Start and End sla processed.

Server 2:
Receives above job's status event, processes remaining 'duration' sla. eventProcessed now = 0111, but incremented to 1000 due to

SLACalculatorMemory.addJobStatus() : 762
if (slaCalc.getEventProcessed() == 7) {
      slaInfo.setEventProcessed(8);
     slaMap.remove(jobId);
}

Back to Server 1: (doing periodic SLA checks)

SLACalculatorMemory.updateJobSla() : 483
if ((eventProc & 1) == 0) { // first bit (start-processed) unset
   if (reg.getExpectedStart() != null) {
         if (reg.getExpectedStart().getTime() + jobEventLatency < System.currentTimeMillis()) {
               // goes ahead and enqueues another START_MISS event and DURATION_MET event

Conclusion, need to fix that check for least significant bit (and next to it) for 'start' and 'duration' to avoid duplicate events

Issue Links

• is superceded by: OOZIE-1933 SLACalculatorMemory HA changes assume SLARegistrationBean exists for all jobs
• links to: ReviewBoard