[
https://issues.apache.org/jira/browse/SPARK-6534?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14380906#comment-14380906
]
Kay Ousterhout commented on SPARK-6534:
---------------------------------------
[~sowen] is right that the current behavior is correct (albeit understandably
confusing!). The ID is a unique identifier that the scheduler assigns for the
task attempt. The scheduler assigns these IDs when tasks are launched, so they
will always show up as being sequential in the UI. The "Index" is a unique
identifier for the partition that the task is computing. One reason the
Indexes would not be sequential is if the tasks have locality constraints (so
the scheduler will launch them as resources on the appropriate machines become
available). Also, you'll notice that the indexes are always in [0, tasksInJob)
whereas the attempt Id only starts at 0 for the first job. If you run the same
job again, the attempt Ids will begin at <# tasks already launched for previous
jobs>.
> Task ID and Index columns appear to be reversed on AM web UI
> ------------------------------------------------------------
>
> Key: SPARK-6534
> URL: https://issues.apache.org/jira/browse/SPARK-6534
> Project: Spark
> Issue Type: Bug
> Affects Versions: 1.2.0
> Reporter: Alex Shafer
> Priority: Minor
> Attachments: task_id_index.png
>
>
> On the AM web UI at /stages/stage?id=X&attempt=Y it looks like the Index and
> ID columns are reversed for the Tasks table.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
