Wish list for xpublish capabilities

[jonmjoyce]

Hello xpublish community!

I'd like to kick off a discussion on high-level capabilities that we'd like to add to xpublish or any of its subcomponents. What needs more work and TLC? What do you wish xpublish could do that it currently does not? I'd like to get some of these ideas out by end of next week, thanks!

[wachsylon]

Just an idea: What about an xarray dataset plugin? Allowing routes like

SPLITOPERATOR/ARGUMENTS/COMBINEOPERAOTR
, e.g.:
groupby/time.month/mean

for all datasets. This would allow clients to make use of server-side resources.

[jonmjoyce]

These are a few milestones that I have suggested:

• Improve caching / reduce data reads
• Add more support for async and parallel operations. Scale out large dataset operations
• Improve aggregations. Support larger time-series (years)
• Improve interoperability with legacy systems (THREDDS/ERDDAP)
• Service endpoints to retrieve STAC metadata spec
• Add metrics to measure which datasets are being accessed and how many bytes are transferred per dataset.
• Investigate and prototype additional functionality for data science / Jupyter notebook access patterns

[jhamman]

Adding:

• Native support for Xarray's Datatree structure
• Improved path handling for nested hierarchies (related to datatree and catalogs)