Bugzilla – Bug 5498
Add "mark FAILED" to BatchStatus state transitions
Last modified: 2015-09-03 20:57:26 UTC
Should state transition include mention that, for something like a JVM that dies... SOMEONE must set a job status to FAILED in order to restart.
See Mailing List dicusssion:
Add, for example,
E.g. in Sec. 10.8:
"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."
At the very end of Section 8.7, (after both numbered lists), add the following completely new paragraphs:
In addition to these conditions and events which are well-defined by this specification, it is also recognized that the runtime may be forced to make another transition of job and step batch status.
For example, a JVM hang may cause a job to appear in STARTED state even though it is no longer running. The specification forbids running multiple executions of a given job instance at the same time. In order to recover and allow restart it is expect that a batch runtime implementation might provide a mechanism to automatically or through user intervention mark the appropriate job and step execution(s) as FAILED (i.e. set the batch status as FAILED).
The details are left entirely to the implementation, we are just recognizing here that this is a valid state transition.