Job restart after stop fails [BATCH-2315]

[spring-projects-issues]

Chris Haas opened BATCH-2315 and commented

In our jobs we use the following construct to conditionally stop a job based on a property passed into the job:

<decision id="decisionId" decider="decider">
    <stop on="NEEDS_FINALIZATION" restart="placeAccounts"/>
    <next on="*" to="placeAccounts"/>
</decision>

In Spring Batch 2.1, "placeAccounts" is executed when the job is restarted. In Spring Batch 3, the decision is re-executed, which of course means the job immediately stops again.

In 3.0.1, the job's flow has stateTransitions for the decision look like

StateTransition: [state=placementImport.placementFinalizationDecision, pattern=*, next=placementImport.placeAccounts],
StateTransition: [state=placementImport.placementFinalizationDecision, pattern=NEEDS_FINALIZATION, next=placementImport.stop60],
StateTransition: [state=placementImport.stop60, pattern=*, next=null],

but in 2.1, they look like:

StateTransition: [state=placementImport.placementFinalizationDecision, pattern=*, next=placementImport.placeAccounts],
StateTransition: [state=placementImport.placementFinalizationDecision, pattern=NEEDS_FINALIZATION, next=placementImport.stop60],
StateTransition: [state=placementImport.stop60, pattern=*, next=placementImport.placeAccounts],

Notice that the next element on stop60 is blank, which is probably the cause of the problem.

---

Affects: 3.0.1

2 votes, 3 watchers

[spring-projects-issues]

Chris Haas commented

Attach an updated DecisionStepTests that illustrates the problem.

[spring-projects-issues]

Chris Haas commented

Note that, in all of the existing test cases, the step/decider keeps track of how many times it was called and, on a even number, returns a string to match the <next> tag, which means the restart attribute on the <stop> is never executed.

[spring-projects-issues]

Chris Haas commented

I've discovered the source of the problem. I had switched to the 3.0 schema, which adds the attribute exit-code to the <stop> tag, with a default value of "STOPPED". This value causes the restart attribute to be ignored. If I add exit-code="", the code runs as expected.

So,the bug fix is to update the documentation for the stop tag to indicate that it uses the exit-code attribute.

[spring-projects-issues]

Chris Haas commented

It would be nice to fix this, so that, during a <stop> you are allowed to change the ExitStatus - currently you cannot.

As a suggestion to see how bad the problem is, try upgrading one of the XML-based unit tests that uses <stop> to change the xsd reference from 2.2 to 3.0 - it will fail on restart.

[fmbenhassine]

Minimal complete example to reproduce the issue here: https://github.com/benas/spring-batch-lab/tree/main/issues/gh1287