Right now what is in the trunk and 4.1 branch, the bundle will go into
"DONEWIHTERROR" stage
________________________________
From: Mona Chitnis <mona.chit...@yahoo.in>
To: Rohini Palaniswamy <rohini.adi...@gmail.com>; "dev@oozie.apache.org"
<dev@oozie.apache.org>
Cc: bowen zhang <bowenzhang...@yahoo.com>
Sent: Thursday, September 18, 2014 5:49 PM
Subject: Re: issue after OOZIE-1807
bq. if a bundle with two actions - one FAILED due to coordinator submission
error, other KILLED), bundle is supposed to KILLED bq. Bundle should be
FAILED and not KILLED. Only when user has KILLED the bundle, should its status
be KILLED.
Thanks for minor correction. I was shooting for bundle will _not_ be
DONEWITHERROR, which Bowen said he's observing
Mona Chitnis
Software Engineer, Hadoop Team
Yahoo!
On Thursday, September 18, 2014 5:09 PM, Rohini Palaniswamy
<rohini.adi...@gmail.com> wrote:
bq. Shouldn't oozie be intelligent enough to do a no-op on a killed coord job?
There are options now to resume a killed coord job. If new end time was
applied on other coord jobs and not applied on that one, user needs to know.
bq. if a bundle with two actions - one FAILED due to coordinator submission
error, other KILLED), bundle is supposed to KILLED Bundle should be FAILED
and not KILLED. Only when user has KILLED the bundle, should its status be
KILLED.
-Rohini
On Thu, Sep 18, 2014 at 4:09 PM, Purshotam Shah
<purus...@yahoo-inc.com.invalid> wrote:
Bowen,
JIRA has explanation. Please update JIRA if you see any issue with
approach.
>Why is it a good idea to throw an exception if one of the coord jobs is
>in "killed" state? In the BundleJobChangeXCommand, the code doesn't even
>attempt to change the coord job. Shouldn't oozie be intelligent >enough
>to do a no-op on a killed coord job?
To let user know the list of coord jobs for which change is not applied.
Puru.
On 9/18/14, 2:11 PM, "bowen zhang" <bowenzhang...@yahoo.com.INVALID> wrote:
>Hi Purshotam,
>Why is it a good idea to throw an exception if one of the coord jobs is
>in "killed" state? In the BundleJobChangeXCommand, the code doesn't even
>attempt to change the coord job. Shouldn't oozie be intelligent enough to
>do a no-op on a killed coord job?
>Bowen
>
>
>
>________________________________
> From: Purshotam Shah <purus...@yahoo-inc.com.INVALID>
>To: "dev@oozie.apache.org" <dev@oozie.apache.org>; Mona Chitnis
><mona.chit...@yahoo.in>; bowen zhang <bowenzhang...@yahoo.com>
>Sent: Wednesday, September 17, 2014 6:17 PM
>Subject: Re: issue after OOZIE-1807
>
>
>Hi Bowen,
> BundleJobChangeXCommand command will get applied to bundle and coord
>jobs. It will aggregate message for all killed coord jobs and throw them
>as exception.
>It is similar to chmod command.
>
>JIRA has more details. Let me know if you need any other information.
>
>Puru.
>
>
>
>
>
>On 9/17/14, 6:05 PM, "Mona Chitnis" <mona.chit...@yahoo.in> wrote:
>
>>
>>Puru,
>>Bowen just gave me a call regarding this issue. Can you answer his
>>question? That'll be faster than me digging through the code.
>> Mona Chitnis
>>Yahoo!
>>
>> On Wednesday, September 17, 2014 5:51 PM, bowen zhang
>><bowenzhang...@yahoo.com.INVALID> wrote:
>>
>>
>> Hi guys,
>>
>>Purshatom, I see you checked oozie-1807 into the trunk. So, I have a
>>question, why does it need to throw an exception when someone wants to
>>change a bundle job where one of its coord job is in KILLED state? Due to
>>the change in BundleJobChangeXCommand, this is throwing exceptions when
>>trying to change a RUNNING bundle job where some of the coord jobs are
>>intentionally killed by the user.
>>Thanks,
>>Bowen
>>
>>
>>
