[
https://issues.apache.org/jira/browse/NUTCH-2177?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15029037#comment-15029037
]
Julien Nioche commented on NUTCH-2177:
--------------------------------------
I am on
Hadoop version: 2.4.0-amzn-7
not clear which version of Yarn it comes with.
mapred.job.tracker does not appear in the conf at all but
mapreduce.jobtracker.address does and has 'local' as value
I can see
15/11/26 15:58:13 INFO Configuration.deprecation: mapred.job.tracker is
deprecated. Instead, use mapreduce.jobtracker.address
15/11/26 15:58:13 INFO crawl.Generator: Generator: jobtracker is 'local',
generating exactly one partition.
so my guess is that the Configuration maps the old key (mapred.job.tracker) to
the new one (mapreduce.jobtracker.address)
@markus17 what do you get for mapreduce.jobtracker.address?
> Generator produces only one partition even in distributed mode
> --------------------------------------------------------------
>
> Key: NUTCH-2177
> URL: https://issues.apache.org/jira/browse/NUTCH-2177
> Project: Nutch
> Issue Type: Bug
> Components: generator
> Reporter: Julien Nioche
> Priority: Blocker
> Fix For: 1.11
>
>
> See
> [https://github.com/apache/nutch/blob/trunk/src/java/org/apache/nutch/crawl/Generator.java#L542]
> 'mapred.job.tracker' is deprecated and has been replaced by
> 'mapreduce.jobtracker.address', however when running Nutch on EMR
> mapreduce.jobtracker.address has local as a value. As a result we generate a
> single partition i.e. have a single map fetching later on (which defeats the
> object of having a distributed crawler).
> We should probably detect whether we are running on YARN instead, see
> [http://stackoverflow.com/questions/29680155/why-there-is-a-mapreduce-jobtracker-address-configuration-on-yarn]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
