Unit Test Results

See test report for an extended history of previous test failures. This is useful for diagnosing flaky tests.

     15 files  ± 0       15 suites  ±0   4h 29m 11s ⏱️ -2s
 17 163 tests + 1   15 968 ✅ + 1   1 195 💤 ±0  0 ❌ ±0 
211 476 runs  +13  194 296 ✅ +10  17 180 💤 +3  0 ❌ ±0 

Results for commit f4c0011. ± Comparison against base commit 7393a77.

dask.array.tests.test_array_core ‑ test_normalize_chunks

dask.array.tests.test_array_core ‑ test_normalize_chunks[normalize_chunks]
dask.array.tests.test_array_core ‑ test_normalize_chunks[normalize_chunks_cached]

Yes decent impact.

# BEFORE: dask  : 2024.12.1, xarray: 2025.1.0
# Wall time: 1min 12s
#
# AFTER: xarray: 2024.11.1.dev50+gd9365109.d20250108, dask  : 2024.12.1+966.gf4c001150
# Wall time: 49.1 s
%time xr.open_zarr("gs://gcp-public-data-arco-era5/ar/full_37-1h-0p25deg-chunk-1.zarr-v3")

from dask.array.core import normalize_chunks_cached

normalize_chunks_cached.cache_info()
# CacheInfo(hits=271, misses=2, maxsize=128, currsize=2)

Now the big issue is tokenize.

This is the issue:

from dask.layers import ArraySliceDep

chunks = ((1,)*1_000_000, (721,), (1441,))
%timeit dask.base.tokenize(ArraySliceDep(chunks)) # 55ms

for 300 variables, that alone takes 16s. somehow using a cache with id(chunks) in that tokenization would fix it.

EDIT: The tuples within chunks are the cached values, not chunks itself

for 300 variables, that alone takes 16s. somehow using a cache with id(chunks) in that tokenization would fix it.

EDIT: The tuples within chunks are the cached values, not chunks itself

Yeah, this is tricky, I'll see if we can do something here

@dcherian

can you import from

dask.array.api when you add this to xarray?