Skip to content

Aggregator of remote partitions is not passed all worker steps upon controller restart #5158

@brian-mcnamara

Description

@brian-mcnamara

Bug description
In a remote partitioned job, with a controller and workers, if the controller job is stopped (for example when the pod is killed and the job is gracefully shutdown), if the remote workers steps complete, and the controller job is restarted, if the job progression is to pass the result to a aggregator, the aggregator is passed an empty list of step executions, even tho the remote steps completed successfully.

This is due to SimpleStepExecutionSplitter filtering out the executions since they were completed.

Environment
Spring boot 4.0.0, batch 6.0.0, batch-integration 6.0.0, JDK21
However this is also a issue in batch 5

Steps to reproduce

See https://github.com/brian-mcnamara/SpringBatch6/tree/shutdownAggregator for example repo

(can be run with ./gradlew run) Specifically note line 235 which expects the aggregator to be passed a list of two step executions.

Expected behavior

When the aggregator is called, it should contain the steps the worker completed.

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions