[
https://issues.apache.org/jira/browse/HADOOP-4189?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12634372#action_12634372
]
Amar Kamat commented on HADOOP-4189:
------------------------------------
bq. For restartability, we really should separate the start event from the
finish event.
I dont think it really matters. Any task-attempt that has started but not
closed is probably useless w.r.t recovery. Hence task-attempt which has a
_START_ and _END_ is really what we want and hence we can keep it as it is.
bq. Why does it double? Hopefully we are using UTF-8 encoding so it should be 1
byte/char.
I checked on my local box and its ASCII. Assuming 3mb as the default, we now
risk 5.2k tasks which should be ok.
> HADOOP-3245 is incomplete
> -------------------------
>
> Key: HADOOP-4189
> URL: https://issues.apache.org/jira/browse/HADOOP-4189
> Project: Hadoop Core
> Issue Type: Bug
> Components: mapred
> Affects Versions: 0.19.0
> Reporter: Amar Kamat
> Assignee: Amar Kamat
> Priority: Blocker
> Fix For: 0.19.0
>
> Attachments: HADOOP-4189-v1.patch, HADOOP-4189-v2.patch,
> HADOOP-4189-v3.1.patch, HADOOP-4189-v3.patch, HADOOP-4189.patch
>
>
> There are 2 issues with HADOOP-3245
> - The default block size for the history files in hadoop-default.conf is set
> to 0. This file will be empty throughout. It should be made null.
> - Same goes for the buffer size.
> - InterTrackerProtocol version needs to be bumped.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
