Add opt-in for deleting intermediate runpaths for es_mda

[eilskra]

Added the option to delete the intermediate runpaths for es_mda. Often not not used for prediction runs, QC should be able to be done with the data from storage. Will free up space on scratch.

• PR title captures the intent of the changes, and is fitting for release notes.
• Added appropriate release note label
• Commit history is consistent and clean, in line with the contribution guidelines.
• Make sure unit tests pass locally after every commit (git rebase -i main --exec 'just rapid-tests')

When applicable

• When screenshots are changed: Review screenshot-PR in ert-testdata,
  merge screenshot-PR in ert-testdata before merging this PR.
• When there are user facing changes: Updated documentation
• New behavior or changes to existing untested code: Ensured that unit tests are added (See Ground Rules).
• Large PR: Prepare changes in small commits for more convenient review
• Bug fix: Add regression test for the bug
• Bug fix: Add backport label to latest release (format: 'backport release-branch-name')

[Copilot]

🟡 Changes recommended

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Adds optional ES-MDA intermediate runpath cleanup to reduce scratch-space usage.

Changes:

• Adds GUI and CLI controls for cleanup.
• Deletes intermediate ES-MDA runpaths after evaluation.
• Persists and tests the new configuration field.

File summaries

File	Description
src/ert/__main__.py	Adds the CLI cleanup flag.
src/ert/gui/experiments/multiple_data_assimilation_panel.py	Adds the default-enabled GUI checkbox.
src/ert/run_models/model_factory.py	Passes cleanup configuration to the model.
src/ert/run_models/multiple_data_assimilation.py	Implements intermediate runpath deletion.
src/ert/run_models/run_model_configs.py	Adds and serializes the cleanup setting.
src/ert/storage/local_experiment.py	Extends stored experiment metadata.
tests/ert/unit_tests/run_models/test_multiple_data_assimilation.py	Tests runpath deletion behavior.
tests/ert/unit_tests/run_models/test_model_factory.py	Updates ES-MDA factory fixtures.
tests/ert/unit_tests/run_models/test_experiment_serialization.py	Updates serialization input.
tests/.../snake_oil.json	Updates the ES-MDA snapshot.
tests/.../poly.json	Updates the ES-MDA snapshot.
tests/.../config.json	Updates the ES-MDA snapshot.

Review details

Suppressed comments (2)

src/ert/run_models/multiple_data_assimilation.py:231

• ignore_errors=True silently suppresses permission and filesystem failures, so the run can claim it is deleting paths while freeing no scratch space. Catch OSError and emit a warning (without failing the expensive experiment) so operators can detect cleanup failures.

            logger.info(f"Deleting runpath of intermediate iteration: {path}")
            shutil.rmtree(path, ignore_errors=True)

src/ert/gui/experiments/multiple_data_assimilation_panel.py:233

• This is a new user-facing, default-enabled destructive behavior, but no .rst documentation is updated. Document what is removed, that storage results remain, how GUI users opt out, and that the CLI remains opt-in; the existing ES-MDA how-to at docs/ert/getting_started/howto/esmda_select_prior.rst is a natural location.

        layout.addRow(
            ("Only keep prior and\nposterior on disk:"),
            self._only_keep_first_and_last_iteration_box,
        )

• Files reviewed: 12/12 changed files
• Comments generated: 6
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 93.54839% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 90.74%. Comparing base (67dc075) to head (fbaf347).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/ert/run_models/multiple_data_assimilation.py	89.47%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #14325      +/-   ##
==========================================
- Coverage   90.81%   90.74%   -0.07%     
==========================================
  Files         497      497              
  Lines       36062    36093      +31     
==========================================
+ Hits        32748    32751       +3     
- Misses       3314     3342      +28     

Flag	Coverage Δ
cli-tests	35.67% <41.93%> (+<0.01%)	⬆️
fuzz	43.84% <35.48%> (-0.01%)	⬇️
gui-tests	58.14% <64.51%> (+<0.01%)	⬆️
performance-and-unit-tests	80.92% <93.54%> (-0.13%)	⬇️
test	45.25% <29.03%> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/ert/__main__.py	88.47% <100.00%> (+0.08%)	⬆️
...ui/experiments/multiple_data_assimilation_panel.py	91.59% <100.00%> (+0.25%)	⬆️
src/ert/run_models/model_factory.py	97.69% <ø> (ø)
src/ert/run_models/run_model_configs.py	100.00% <100.00%> (ø)
src/ert/storage/local_experiment.py	96.35% <100.00%> (+<0.01%)	⬆️
src/ert/utils.py	96.77% <100.00%> (+0.05%)	⬆️
src/ert/run_models/multiple_data_assimilation.py	95.61% <89.47%> (-1.23%)	⬇️

... and 6 files with indirect coverage changes

[codspeed-hq]

Merging this PR will not alter performance

✅ 36 untouched benchmarks

---

_{Comparing eilskra:iteration-deletion (a59a9ed) with main (67dc075)}

[ertomatic]

Screenshots differ from baselines. A baseline update PR has been prepared: equinor/ert-testdata#112

[Copilot]

🟡 Changes recommended

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Review details

• Files reviewed: 16/17 changed files
• Comments generated: 2
• Review effort level: Balanced

[eilskra]

No? It will still create subfolders in the realization iteration, otherwise the realization iteration would overwrite itself for each realization?

[berland]

Interesting comment from Copilot, it is absolutely possible to set
RUNPATH poly_out/realiation-<IENS>/iter-foo
and es-mda will actually run quite fine, reusing the the runpath for each realization over the iterations, and you have effectively implicitly deleted iter-0 to iter-2. In this case the deletion feature is superfluous (but it does represent a safety net as rerunning in an existing realization can have side-effects, just not for iter-1) but I don't see any big problems with it.

[berland]

I am not sure about the usage of "intermediate runpath" in the GUI text here - wondering if users will understand that concept, and if they are unsure of what it means, they will for sure not click it.

It may also be unclear to the user when this deletion is going to happen.

We can calculate what the intermediate iterations are, based on the input further up in the GUI element, and include "iter-1, iter-2" in the text somehow. On-prem users would also like to see the term "scratch" in the message, they might relate more to that than the term "runpath", but then "scratch" is a specific user group tribe-language and not Ert lingo.

[eilskra]

@berland

I am not sure about the usage of "intermediate runpath" in the GUI text here - wondering if users will understand that concept, and if they are unsure of what it means, they will for sure not click it.
It may also be unclear to the user when this deletion is going to happen.

I was thinking of wording the label in such a manner that it highlights that this is scratch disk optimization. However, my concern was that we could ruin runs for the users if we are not very explicit that this is a non-reversal deletion. The tooltip explains when the deletion happens, but I guess we could try to add this to the label too

We can calculate what the intermediate iterations are, based on the input further up in the GUI element, and include "iter-1, iter-2" in the text somehow. On-prem users would also like to see the term "scratch" in the message, they might relate more to that than the term "runpath", but then "scratch" is a specific user group tribe-language and not Ert lingo.

I tried to reduce the amount of tribe-language used (I am still not consistent, e.g restart vs prediction), but in the user documentation I highlight that this frees up scratch disk space (selling point of functionality).

I am not sure if we should accommodate wording or not, feel like we're currently in an awkward middle position:

• If we don't use Ert lingo, it wont get picked up and will not be established
• If we don't use tribe-language, feature might not get used due to users' uncertainties/confusions about the feature

Suggested labels:

• "Delete intermediate realization iterations after completion"
• "Minimize scratch-disk storage used
  (Delete intermediate realization iterations)"
• "Only store initial prior and final posterior on scratch disk"

[berland]

Instead of newlines, can't we have Qt or something wrap the text? The newlines make this text hard to maintain later.

[eilskra]

Text also gets used for the CLI-arg helper, unsure if there are methods to wrap the text there?

[berland]

This last sentence is probably slightly incorrect. "Restart" is a very overloaded term (it is both an es-mda term, and an Eclipse term), and for some definitions (the Eclipse one), this statement is very correct. But since the parameters are in storage, es-mda can probably restart from intermediate iterations.

[eilskra]

If it can be used as prior despite being removed from scratch, should we also add opt-in to remove initial prior (iter-0)?

[ertomatic]

Screenshot tests now pass. The baseline update PR equinor/ert-testdata#112 has been closed.

[berland]

Another label suggestion: "Keep only prior and posteriors runpaths"

[oyvindeide]

Does this belong here? Could see this belonging with the rest of the runpath configuration. While only needed in esmda at the moment, we do aim for this to be available and used for all run models down the line.

[eilskra]

If not added to the MultipleDataAssimilationConfig the experiment serialization test fails with:

AssertionError: Key delete_intermediate_runpaths not found in saved experiment. Expected value: False

Test added in this commit

Edit: Post discussion, will look into adding this to runpath config