Skip to main content

[jsr352-public] Re: Do we need another BatchStatus transition (mark jobs as FAILED)?

  • From: Scott Kurz < >
  • To:
  • Subject: [jsr352-public] Re: Do we need another BatchStatus transition (mark jobs as FAILED)?
  • Date: Mon, 28 Oct 2013 10:38:43 -0400


Michael,

Thanks for giving this some thought and relaying that discussion.

So note, I'm not suggesting any rules or standards defining how a job such
as this gets marked FAILED by the implementation.
I was simply suggesting that, say in Sec 8.7, we include a sentence or two
acknowledging this transition getting performed by SOMEONE.

Maybe saying the runtime does it was too strong...

How about:

"In the case of a job that ...... e..g hung JVM.... etc....
  It is expected that an implementation provides some
implementation-specific mechanism for transitioning BatchStatus to FAILED
state, so that it can be restarted.  This mechanism can be executed
automatically by the implementation or via an implementation-defined manual
operation."

What's the point of saying that?   I think simply that the spec is clearly
trying to define a complete state transition flowchart.... and so this
obvious gap should be closed, even if the details are implementation
specific.

------------------------------------------------------
Scott Kurz
WebSphere Batch / Compute Grid Development
T/L 295-5649;
External Phone 845-435-5649

--------------------------------------------------------



From:   Michael Minella 
< >
To:     
" "
 
< >,
Date:   10/28/2013 10:26 AM
Subject:        [jsr352-public] Re: Do we need another BatchStatus transition
            (mark jobs as FAILED)?



Spring Batch has taken the approach that this is not something the
framework should support.  We actually had a bit of a spirited debate on
this at this year's SpringOne in one of the batch talks.  Our stance is
that in this scenario, it is almost always a human decision that is
required to determine that this update is required.  In the example you
provide (a JVM dies for some reason), how would the job or other JVM know
what happened?  In the end, it would typically take human investigation and
then action.  The furthest I would suggest we go (if at all) would be
providing a method on the JobOperator to set a Job/Step to a given status
programatically.  This would prevent the need to manually update the
underlying job repository impl (database tables in most cases) by hand.

Michael

Thanks,
Michael Minella

http://www.michaelminella.com


On Fri, Oct 25, 2013 at 9:47 PM, Scott Kurz 
< >
 wrote:
  I don't remember this issue being raised up until now (maybe it was in
  the EG)...

  Say a job is running along and someone kills the JVM or it hits OOM.
  Can't the implementation determine somehow that jobs like this should be
  moved into FAILED state (BatchStatus)?

  I don't think the spec needs to specify how this happens... but it seems
  to try to completely cover all state transitions, and it specifically
  disallows restart of an already-STARTED job (in Sec. 10.8).

  So I think it should say something allowing this "mark FAILED" transition
  to occur in an impl-specific way.

  Thoughts?
  ------------------------------------------------------
  Scott Kurz
  WebSphere Batch / Compute Grid Development
  T/L 295-5649;
  External Phone 845-435-5649
  

  --------------------------------------------------------




GIF image



[jsr352-public] Do we need another BatchStatus transition (mark jobs as FAILED)?

Scott Kurz 10/26/2013

[jsr352-public] Re: Do we need another BatchStatus transition (mark jobs as FAILED)?

Michael Minella 10/28/2013

[jsr352-public] Re: Do we need another BatchStatus transition (mark jobs as FAILED)?

Scott Kurz 10/28/2013

[jsr352-public] Re: Do we need another BatchStatus transition (mark jobs as FAILED)?

Michael Minella 10/28/2013

[jsr352-public] Re: Do we need another BatchStatus transition (mark jobs as FAILED)?

Scott Kurz 10/28/2013

[jsr352-public] Re: Do we need another BatchStatus transition (mark jobs as FAILED)?

Michael Minella 10/28/2013

[jsr352-public] Re: Do we need another BatchStatus transition (mark jobs as FAILED)?

Cheng Fang 10/28/2013
 
 
Close
loading
Please Confirm
Close