[
https://issues.apache.org/jira/browse/HADOOP-2801?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12567197#action_12567197
]
Devaraj Das commented on HADOOP-2801:
-------------------------------------
That's right. The startTime is initialized just once inside TaskInProgress and
the speculative check takes that into consideration. So for cases where a
failed task is retried (let's say at least after a 1 minute interval), and
another tasktracker asks for a task at around the same time instant, the check
_(System.currentTimeMillis() - startTime >= SPECULATIVE_LAG)_ is going to
return true, and a speculative instance will be launched almost simultaneously.
Also i noticed there is a related field execStartTime. This is also updated
only once.
> Non-speculative and speculative tasks for a TIP start at the same time.
> -----------------------------------------------------------------------
>
> Key: HADOOP-2801
> URL: https://issues.apache.org/jira/browse/HADOOP-2801
> Project: Hadoop Core
> Issue Type: Bug
> Components: mapred
> Reporter: Amar Kamat
> Fix For: 0.16.1
>
>
> I ran into a situation where the non-speculative task and the speculative
> task were started at the same time. Ideally there should be a 1 min gap for
> speculation. One possible reason would be the usage of {{startTime}} in
> deciding whether there is a speculative lag.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
