[
https://issues.apache.org/jira/browse/IGNITE-3167?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Taras Ledkov updated IGNITE-3167:
---------------------------------
Comment: was deleted
(was: - *1, 2 ,5* - fixed;
- *3* - because external process doesn't contain ignite node. So, we have to
use JdkMarshaller.
- *4* - {{Process.toString()}} contains real PID of process. It may be useful
with debug cases. I add {{childProcId}} to output.)
> Hadoop: restore external execution.
> -----------------------------------
>
> Key: IGNITE-3167
> URL: https://issues.apache.org/jira/browse/IGNITE-3167
> Project: Ignite
> Issue Type: Task
> Components: hadoop
> Affects Versions: 1.5.0.final
> Reporter: Vladimir Ozerov
> Assignee: Taras Ledkov
> Priority: Critical
> Labels: roadmap
> Fix For: 1.8
>
>
> Some time ago we decided to get rid external execution mode. It appears to be
> a wrong decision.
> Hadoop users rely on its process-per-job nature in lot's places. One of such
> case could be observed in HiBench Bayes benchmark:
> 1) Job creates something in the local file system through Hadoop FileSystem
> API.
> 2) Then it tries to get this data using regular java.io.FileReader and
> relative paths.
> This doesn't work in embedded mode because our LocalFileSystem wrapper
> assigns different work dirs for jobs, but process-wide working directory is
> always the same. As a result, aforementioned benchmark doesn't work in
> Ignite, but work with standard Hadoop job tracker.
> It seems that we must return external execution back.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
