[
https://issues.apache.org/jira/browse/MAPREDUCE-1854?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12877740#action_12877740
]
Balaji Rajagopalan commented on MAPREDUCE-1854:
-----------------------------------------------
changing visibility from 'package private' to 'public' for testing purpose
isn't advisable. Consider injecting a getter with public access
The tasktrackerStatus is a writable object, should'nt the inner class of
writable object be
public for others to use, I understand we don't want to change permission of
jobtracker or
tasktracker, but tasktrackerStatus is a writable object which can be streamed
across the wire,
but if you still insist I will have aspect hooks to make the inner class
methods public.
using StringBuffer to append a couple of tokens and convert the result to a new
String seems excessive. Why not to use String?
I used stringbuffer for appends, since each time a new string object is
created, and temporary
object is discarded a new string object is created, and I thought if I use
string then people
will comment why not use stringbuffer since it avoids all the temporary object
creations :(.
Do agree with you AbstractTestCase methods looks more of a helper, will move it
a static
helper than have it part of parent hierarchy, I was in a dilemma on how to do
this myself.
My real intention of having abstract parent class is have common functionality
that can be
shared among all the test cases, which can be reused easily.
The changes to the abstractdaemoncluster are related to the patch, because the
health script automation
validates that a single task tracker should be stopped/started instead of
stopping and starting the
whole cluster.
I do agree DaemonProtocolAspect.aj is not related to the patch, it was same fix
vinay mentioned,
but I had to have that code to make the test case pass. So I will remove this
from my subsequent patch.
Also I will have to put an dependency on the vinay's jira meaning I can only
check in my patch
after he completes his changes and checks it in.
Also I am planning to break the individual test methods to seperate classes
since the test occasionally timesout.
The clusterprocessmanager, defined the iterface method to get a daemon process
given the hostname,
HadoopDaemonRemoateCluster provides the implemetation, for the interface. The
pushConfig in
Remote Process was not returning the newConf Dir, now I need to return that
since health script
selectively restarts a particual daemon. So it is all related to the health
script patch.
> [herriot] Automate health script system test
> --------------------------------------------
>
> Key: MAPREDUCE-1854
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-1854
> Project: Hadoop Map/Reduce
> Issue Type: New Feature
> Components: test
> Environment: Herriot framework
> Reporter: Balaji Rajagopalan
> Assignee: Balaji Rajagopalan
> Attachments: health_script_5.txt
>
> Original Estimate: 120h
> Remaining Estimate: 120h
>
> 1. There are three scenarios, first is induce a error from health script,
> verify that task tracker is blacklisted.
> 2. Make the health script timeout and verify the task tracker is blacklisted.
> 3. Make an error in the health script path and make sure the task tracker
> stays healthy.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
