Fix table sorting on all jobs page.

It now sorts based on the time that the scheduler
learns about the job rather than the time that the
first stage begins executing.  This is necessary
because the earliest stage that a job depends on
may not necessarily be executed (e.g. if it lies
behind a shuffle boundary for a shuffle that’s
already been computed, or if some RDD has been
cached, etc.).

core/src/main/scala/org/apache/spark/ui/jobs/AllJobsPage.scala

@@ -30,14 +30,6 @@ private[ui] class AllJobsPage(parent: JobsTab) extends WebUIPage("") {
   private val startTime: Option[Long] = parent.sc.map(_.startTime)
   private val listener = parent.listener
 
-  private def getSubmissionTime(job: JobUIData): Option[Long] = {
-    for (
-      firstStageId <- job.stageIds.headOption;
-      firstStageInfo <- listener.stageIdToInfo.get(firstStageId);
-      submitTime <- firstStageInfo.submissionTime
-    ) yield submitTime
-  }
-
   private def jobsTable(jobs: Seq[JobUIData]): Seq[Node] = {
     val someJobHasJobGroup = jobs.exists(_.jobGroup.isDefined)
 
@@ -107,11 +99,11 @@ private[ui] class AllJobsPage(parent: JobsTab) extends WebUIPage("") {
       val now = System.currentTimeMillis
 
       val activeJobsTable =
-        jobsTable(activeJobs.sortBy(getSubmissionTime(_).getOrElse(-1L)).reverse)
+        jobsTable(activeJobs.sortBy(_.startTime.getOrElse(-1L)).reverse)
       val completedJobsTable =
-        jobsTable(completedJobs.sortBy(getSubmissionTime(_).getOrElse(-1L)).reverse)
+        jobsTable(completedJobs.sortBy(_.endTime.getOrElse(-1L)).reverse)
       val failedJobsTable =
-        jobsTable(failedJobs.sortBy(getSubmissionTime(_).getOrElse(-1L)).reverse)
+        jobsTable(failedJobs.sortBy(_.endTime.getOrElse(-1L)).reverse)
 
       val summary: NodeSeq =
         <div>