Stop batch job execution from a different node

In  batch

Overview

When running batch job executions in a clustered or multiple-node environment configured with the same jdbc batch job repository, users may need to start job execution from one node and later stop it from a different node. The current implementation assumes the job execution will always be stopped from the same node, and thus stopping from a different node has no effect. The implementation should be enhanced to support stopping a job execution from a different node.

Issue Metadata

Issue

Dev Contacts

QE Contacts

Testing By

  • [X] Engineering

  • QE

Affected Projects or Components

WildFly JBeret

Other Interested Projects

N/A

Requirements

Hard Requirements

  • With jdbc job repository configured in WildFly instances, users should be able to start a batch job execution from node A and later stop it from node B, and then restart it from either node A or B.

  • The above batch job operations should work for jobs that consist of both chunk-type steps and batchlet-type steps.

  • The above batch job operations should work for jobs that consist of both non-partitioned and partitioned steps.

  • There should be no change to WildFly batch subsystem model or schema.

Nice-to-Have Requirements

N/A

Non-Requirements

  • the ability to stop job execution from a different node when non-jdbc job repository is used

Proposed Solution

This feature requires changes to both project JBeret and WildFly

  • Changes to JBeret jberet-core

    • Immediately save the STOPPING batch status to jdbc job repository after receiving the stop request. JOB_EXECUTION, STEP_EXECUTION, and PARTITION_EXECUTION tables should be updated to reflect the STOPPING status.

    • For chunk-type step execution: when persisting the step or partition execution at the end of a chunk, update the step or partition tables only if the persistent batch status is not STOPPING. If the update is not successful, it means the current execution has been requested to stop.

    • For batchlet step execution: when persisting the step or partition execution at the end of a batchlet processing, update the step or partition execution only if the persistent batch status is not STOPPING. If the update is not successful, it means the current execution has been requested to stop.

    • Prepare to stop the current step or partition execution after receiving the stop request, via either database update status, or direct stop call from the current node.

  • Changes to WildFly batch subsystem

    • upgrade to jberet-core 1.3.8.Final

    • update batch-jberet/src/main/java/org/wildfly/extension/batch/jberet/job/repository/JobRepositoryService.java to implement the additional methods introduced to jberet-core class org/jberet/repository/JdbcRepository

Test Plan

The following tests will be performed:

  • standalone batch TCK will be executed

  • component project (jberet) test suites will be executed

  • manual testing of the batch job operations listed in hard requirements will be executed

  • existing WildFly test suites will be executed

  • new testcase will be added to WildFly testsuite/multinode

Community Documentation

Community documentation _admin-guide/subsystem-configuration/Jakarta_Batch.adoc will be updated to document this new feature.