Keep heap lean during remote polling

[hpoettker]

Resolves #4598.

The polling in MessageChannelPartitionHandler repeatedly fetches the running job execution with all associated step executions from the job repository. At the moment, it also retains references to step executions that have completed since the last iteration. This can accrue a lot of heap memory as each step execution holds references to a job execution, and that job execution holds references to all step executions.

This PR proposes not to retain any references to step or job executions while there are still running step executions. Instead, the step executions from the last iteration are used to build the result. This works as completed step executions are not expected to change after completion. And it reduces the memory foot print as the objects created in intermediate iterations can be handled by the garbage collector.

No test is added or changed as the changed code is already covered by appropriate tests.

[hpoettker]

@fmbenhassine Do you think this can be included in the 5.0/5.1 releases tomorrow?

It's backward-compatible and seems to solve the problem considering the reproducing example from #4598. I think it even fixes the problem in a way that requires no further refactoring of the polling. But effective feedback on that will probably only come when it's released in a minor release.

[fmbenhassine]

Thank you for the follow up on #4598 and for opening this PR!

Yes, the change LGTM and will be included in 5.0.6 and 5.1.2.

[fmbenhassine]

Rebased and merged as 03a2b4d. Thank you!

[hpoettker]

@adamsaghy I hope it's okay to tag you here.

You commented on #4530, and I assume you were interested due to your involvement in Apache Fineract. You recently upgraded Spring Batch there to 5.1.2 (apache/fineract#3945), mainly for the change from this PR, if I understand correctly.

I would be very interested in feedback. Did this resolve the issue you were seeing? Was there no improvement at all? Or only some degree of improvement?

[adamsaghy]

@adamsaghy I hope it's okay to tag you here.

You commented on #4530, and I assume you were interested due to your involvement in Apache Fineract. You recently upgraded Spring Batch there to 5.1.2 (apache/fineract#3945), mainly for the change from this PR, if I understand correctly.

I would be very interested in feedback. Did this resolve the issue you were seeing? Was there no improvement at all? Or only some degree of improvement?

Hi

We havent had the chance to battle test it yet but its on the todo list. i will let you know once it's done :)