Repository navigation
Fix: Chunksize when opening sharded zarr - #11464
Conversation
d5c6b78 to
4e1d823
Compare
|
is anything missing for this to be merged? |
94b5c9b to
22df04a
Compare
headtr1ck
left a comment
There was a problem hiding this comment.
Feel free to add a whats-new entry.
Couple of smaller fixes.
For me this looks good, maybe someone else should still have a look.
22df04a to
9ed69b6
Compare
9ed69b6 to
bad799c
Compare
bad799c to
22cfdc6
Compare
|
@headtr1ck i made the suggested changes. I noticed this issue #11674 and related PR #11680 dealing with the same section. I changed xarray_chunks to outer_chunks, as that seems to be the terminology used. Maybe @TomNicholas could take a look? |
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Thanks @taus, this looks good! A few small remaining things:
I have these fixes ready locally. Could you enable "Allow edits by maintainers" on this PR so I can push them directly? FYI, this conflicts with #11680 in |
|
hey @headtr1ck. I have enable "Allow edits by maintainers" |

Description
Use Zarr shards as the preferred chunk size in when opening Zarr datasets. Previously it was necessary to
know the shard size when opening to avoid error saving back to disk.
I think the term chunk is overloaded. In this case, chunk can refer to both Dask chunks and Zarr chunks, which are not necessarily the same. In the case of a shared Zarr archive, i would argue that the sensible choice of Dask chunks is the shard size. When writing to disk it could create race conditions with multiple Dask chunks in the same shard. This is also what caused the original error
Checklist