FAQ#
I just saw “UserWarning: The necessary columns optimization failed” what does this mean?
When computing a dask-awkward collection that is reading from disk, the necessary columns optimization tries to discover which parts of a file on disk are necessary for the compute. The optimization then rewrites the data input layer in the graph such that you only read the minimum set of columns from the files on disk. If we are unable to determine the necessary columns, this optimization is simply skipped and the warning (that you’ve seen) is thrown.
Consequences of this optimization being skipped include: an increase in the memory usage of your compute (because more data will be read from disk) and/or an increase in runtime (because it takes time to read the data that you actually do not need).
You can read more about the optimization and how to configure it (by either silencing the warning or raise the exception) at this section of the docs. Please open an issue on the GitHub issue tracker if you think you’ve found a failure in the optimization that should be fixed.
I just saw “UserWarning: metadata could not be determined; a compute on the first partition will occur.” what does that mean?
When dask-awkward stages a new computation it runs the operation on a typetracer array to generate metadata for the new collection. If this is not possible, that is, awkward array itself (independent of Dask) was not able to execute the operation on the existing metadata, then the first partition will be computed to determine the new metadata.
You can bypass the automatic metadata determination using Dask’s configuration manager. The configuration parameter is called
awkward.compute-unknown-meta
. The default setting isTrue
. In code you can do something like this:with dask.config.set({"awkward.compute-unknown-meta": False}): # ... your codeOr you can modify the configuration with environment variables or a YAML file. See Dask’s documentation linked above.