Lockfile updates plus adapt to Dask changes to delayed and internals #6438
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
github-actions
bot
added
the
benchmark_this
trexfeathers
commented
May 1, 2025
trexfeathers
changed the title
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #6438 +/- ##
=======================================
Coverage 89.80% 89.80%
=======================================
Files 90 90
Lines 23752 23752
Branches 4418 4418
=======================================
Hits 21331 21331
Misses 1672 1672
Partials 749 749 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Contributor
⏱️ Performance Benchmark Report: 755a7cdPerformance shiftsFull benchmark resultsGenerated by GHA run |
trexfeathers
commented
May 5, 2025
Comment on lines
+115
to
+140
| def is_pp_layer(obj): | ||
| result = False | ||
| if hasattr(obj, "values"): | ||
| if len(obj) == 1: | ||
| (result,) = [hasattr(v, "_lbpack") for v in obj.values()] | ||
| return result | ||
|
|
||
| # Find the single reference to the PPDataProxy within the dask graph. | ||
| (layer,) = [ | ||
| lay for lay in lazy_mask_array.dask.layers.values() if is_pp_layer(lay) | ||
| ] | ||
| ((layer_key, proxy),) = layer.items() | ||
| # Replace the PPDataProxy with a mock. | ||
| # By replacing directly within the dask graph, we can prove whether an | ||
| # operation is relying on the graph in the expected way. | ||
| mock_proxy = mock.MagicMock(spec=pp.PPDataProxy, wraps=proxy) | ||
| layer.mapping[layer_key] = mock_proxy | ||
|
|
||
| self.assertFalse(mock_proxy.__getitem__.called) | ||
| with contextlib.suppress(TypeError): | ||
| # Expect this to crash - data realisation is complex, and it is not | ||
| # worth mocking this fully. | ||
| _ = lazy_soildata_array.compute() | ||
| # Proves that the soil data graph includes the mask graph - computing | ||
| # the soil data has accessed the mask graph object which we mocked. | ||
| self.assertTrue(mock_proxy.__getitem__.called) |
Contributor
Author
There was a problem hiding this comment.
This change in approach is more agnostic to Dask's internals (although still somewhat vulnerable). It successfully runs, only failing in a classic test failure rather than falling over.
But all I have managed to do is prove that Dask >=2025.4.0 breaks this test on a philosophical level - not just because the structures have changed.
Results of updated test
fileformats/pp.py |
Dask version | Pass/Fail | what this proves |
|---|---|---|---|
v2.2.0 |
2025.3.0 |
Fail ❌ | |
| #3255 | 2025.3.0 |
Pass ✔ | The updated test still detects the case it was designed for |
main |
2025.3.0 |
Pass ✔ | |
main |
2025.4.0 |
Fail ❌ | The Dask graph for the data payload no longer incorporates the mask graph |
Understanding if this is cause for concern - and if so, how to fix things - will be complex. The development team are fully booked on other projects for the foreseeable, so I would suggest a patch release with a dask < 2025.4 pin. @SciTools/iris-devs what are your thoughts?
Contributor
Author
There was a problem hiding this comment.
Discussed at length with @pp-mo, who explained that this test was not written to demonstrate the other changes in #3255, but because there was a concern about something else similar happening again.
Our increased experience with Dask in the 6 years since has given us the confidence that:
- We now write code that is much less vulnerable to things like Don't nest compute calls #3237
- Dask itself has matured, and is less vulnerable to unexpected corner cases
It looks like the refactoring that made this test fail (dask/dask#11736) was technical debt work to make Dask safer and more robust, and is NOT evidence that we have become vulnerable to nested compute() calls.
Based on all of this, we have decided to remove the test.
Contributor
⏱️ Performance Benchmark Report: 2df2034Performance shiftsFull benchmark resultsGenerated by GHA run |
Contributor
Author
|
The performance shifts are most pronounced with smaller data sizes, becoming smaller as the data gets larger. This indicates increased overhead in setting up objects - perhaps dask/dask#11736 - but not something any part of the code that scales with data size. The impact will therefore remain in the milliseconds, even for very large data payloads, and is not a concern (agreed with @pp-mo). |
Contributor
Author
|
Note I am now targetting the changes for a patch release (#6451), which we can then merge back that should unblock future lock-file PRs |
Contributor
Author
|
Closed by #6412 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
🚀 Pull Request
Description
array.compute()todask.compute(array)#6437Consult Iris pull request check list
Add any of the below labels to trigger actions on this PR: