issues

What is your issue?

In attempting to concatenate many datasets along a large dimension (total size ~100,000,000) I'm finding very slow performance, e.g., tens of seconds just to concatenate two datasets.

With some profiling, I find all the time is being spend in this list comprehension:

https://github.com/pydata/xarray/blob/51554f2638bc9e4a527492136fe6f54584ffa75d/xarray/core/concat.py#L584

I don't know exactly what's going on here, but it doesn't look right - e.g., if the size of the dimension to be concatenated is large, this list comprehension can run millions of loops, which doesn't seem related to the intended behaviour.

Sorry I don't have an MRE for this yet but please let me know if I can help further.

What is your issue?

If a user is working in an environment where zarr is not initially installed (e.g., colab), then imports xarray, the list of available backends (engines) will not include zarr, as expected.

If the user subsequently installs zarr, then tries to open a zarr dataset via open_zarr() without restarting their kernel, they will get an error like:

ValueError: unrecognized engine zarr must be one of: ['netcdf4', 'scipy', 'store']

This can also be triggered in a more subtle way, because importing other packages like pandas will also trigger xarray to inspect and then fix this list of available engines.

Obviously this can be resolved by instructing users to restart their kernel after any install commands, or by telling them to always run install commands before any imports.

But I was wondering if there is a way to tell xarray to refresh its information about available engines?

I have been able to hack it by running:

xarray.backends.zarr.ZarrBackendEntrypoint.available = module_available("zarr") xarray.backends.list_engines.cache_clear()

...but maybe there is a better way?

Thanks :heart:

I have a dataset comprising several variables. All variables are dask arrays (e.g., backed by zarr). I would like to use one of these variables, which is a 1d boolean array, to index the other variables along a large single dimension. The boolean indexing array is about ~40 million items long, with ~20 million true values.

If I do this all via dask (i.e., not using xarray) then I can index one dask array with another dask array via fancy indexing. The indexing array is not loaded into memory or computed. If I need to know the shape and chunks of the resulting arrays I can call compute_chunk_sizes(), but still very little memory is required.

If I do this via xarray.Dataset.isel() then a substantial amount of memory (several GB) is allocated during isel() and retained. This is problematic as in a real-world use case there are many arrays to be indexed and memory runs out on standard systems.

There is a follow-on issue which is if I then want to run a computation over one of the indexed arrays, if the indexing was done via xarray then that leads to a further blow-up of multiple GB of memory usage, if using dask distributed cluster.

I think the underlying issue here is that the indexing array is loaded into memory, and then gets copied multiple times when the dask graph is constructed. If using a distributed scheduler, further copies get made during scheduling of any subsequent computation.

I made a notebook which illustrates the increased memory usage during Dataset.isel() here:

https://colab.research.google.com/drive/1bn7Sj0An7TehwltWizU8j_l2OvPeoJyo?usp=sharing

This is possibly the same underlying issue (and use case) as raised by @eric-czech in #4663, so feel free to close this if you think it's a duplicate.

This PR adds a Dataset.query() method which enables making a selection from a dataset based on values in one or more data variables, where the selection is given as a query expression to be evaluated against the data variables in the dataset. See also discussion.

• [x] Tests added
• [x] Passes pre-commit run --all-files
• [x] User visible changes (including notable bug fixes) are documented in whats-new.rst
• [x] New functions/methods are listed in api.rst

Apologies for the very short notice, but I'm giving a webinar on zarr tomorrow and would like to include an xarray demo, especially if it included computing on some pangeo data stored in the cloud. Wouldn't have to be complex at all, just proof of concept. Does anyone have a code example I could borrow? cc @rabernat @jhamman @shoyer @mrocklin

Just to start getting familiar with xray, I've had a go at implementing support for opening a dataset from a specific group within a NetCDF file. I haven't tested on real data but there are a couple of unit tests covering simple cases. Let me know if you'd like to take this forward, happy to work on it further.