Default write empty chunks false #951
Conversation
… into default-write-empty-chunks-false
The call to `np.any(array)` in zarr.util.all_equal triggers the following ValueError: ``` > return ufunc.reduce(obj, axis, dtype, out, **passkwargs) E ValueError: invalid literal for int() with base 10: 'baz' ``` Extending the catch block allows test_array_with_categorize_filter to pass, but it's unclear if this points to a deeper issue.
|
Also trying a no change PR ( #952 ) to see if this is a CI issue outside of these changes |
|
Fewer are hanging but not nothing. 😵💫 |
|
It would be good to get a clearer idea on why that is happening. Did we manage to get a log that we could look at? Waited for a while to see if GH Actions would include output, but nothing showed up. |
|
@joshmoore @jakirkham this is very reminiscent of my own misadventures with
This commit adds the timeout: |
|
Heh. I think I thought that that was now done everywhere. Pushing a commit with lots of timeouts here now. |
joshmoore
force-pushed
the
default-write-empty-chunks-false
branch
from
b30abbb to
8b12549
Compare
|
The thing is this hang showed up very recently. It makes me wonder if there is something we changed or a dependency update in the environment that changed, which is causing issues. For example |
|
Interesting that not even |
joshmoore
force-pushed
the
default-write-empty-chunks-false
branch
from
666b500 to
080a854
Compare
|
0.16.3 seemes safe, @jakirkham. Trying 0.17.1 here now. |
|
0.17.1 seems to be hanging on 3.9 != 1.21.0 as well. |
|
Pushed the commits up to the 0.16.3 pin to #853. |
|
The fact that timeout doesn't work makes me wonder if the hang is in C. Is it possible this is NumPy related? Is NumPy 1.22 hanging somewhere? What if we pinned to NumPy 1.20 in the jobs currently skipping 1.21? |
This reverts commit 57351b9.
joshmoore
force-pushed
the
default-write-empty-chunks-false
branch
from
a9fd445 to
f222cbe
Compare
Codecov Report
@@ Coverage Diff @@
## master #951 +/- ##
==========================================
- Coverage 99.94% 99.94% -0.01%
==========================================
Files 32 32
Lines 11225 11134 -91
==========================================
- Hits 11219 11128 -91
Misses 6 6
|
joshmoore
force-pushed
the
default-write-empty-chunks-false
branch
from
f222cbe to
a062a85
Compare
|
Yeah sorry I think you may be right. Going with 0.16 seems like the right idea. There were deadlock issues in previous 0.17 releases it seems |
Yeah, I was feeling pretty good about it until 080a854 timed out again. But perhaps that PR is somehow "tainted". |
|
Maybe we should run tests with more verbosity? Perhaps that points out which test hangs so we can investigate further |
|
Power cycling for CI |
|
Looks like this is going to fail without the NumPy upgrade ( #955 ). Or perhaps the new NumPy conflicts with pinnings we tried here? |
|
Closing along with #952 Looks like the scare is over. |
Re-opening #853 from another branch to see if that fixes the github action jobs.