# code not shown queries the CMR for urls, and use constraint expressions to only keep `["/T2M", "/U2M", "/V2M", "/SLP"] ` and their dimensions arrays

len(opendap_urls)
>>> 31

%%time
consolidate_metadata(opendap_urls, concat_dim='time', session=my_session)
>>> datacube has dimensions {'lon[0:1:575]', 'lat[0:1:360]'} , and concat dim: `time`
CPU times: user 530 ms, sys: 163 ms, total: 694 ms
Wall time: 16.5 s

%%time
ds = xr.open_mfdataset(opendap_urls, engine='pydap', session=my_session, combine='nested', concat_dim="time", chunks={'lat':90}, batch=True)
>>> CPU times: user 557 ms, sys: 75 ms, total: 632 ms
Wall time: 704 ms


# lazily compute the daily mean - does not trigger computation 
%%time
nds = ds.isel(lat=slice(20, 90))
nds = nds.resample(time="1D").mean()
>>> CPU times: user 158 ms, sys: 2.01 ms, total: 160 ms
Wall time: 161 ms

######### Store subset data in a local `Test.nc4` file. 
# Before storing, the code below triggers all computations that were "lazy". This is, downloads dap responses of all `T2M`, `U2M`, `V2M` and `SLP` (31 granules, 1 dap url downloads all data for that granule). And before writting into diwnk, xarray computes the time average (daily) per variable .

%%time
nds.to_netcdf("Test.nc4", mode='w')
>>> CPU times: user 3.03 s, sys: 2.26 s, total: 5.29 s
Wall time: 35.3 s


# check that the dap response of a granule is consistent:
my_session.cache.urls()[5].replace("%5B", "[").replace("%5D","]").replace("%3A",":").replace("%3B",";")
>>> 'https://opendap.earthdata.nasa.gov/collections/C1276812863-GES_DISC/granules/M2T1NXSLV.5.12.4:MERRA2_400.tavg1_2d_slv_Nx.20230103.nc4.dap?dap4.ce=SLP[0:1:23][0:1:89][0:1:575];T2M[0:1:23][0:1:89][0:1:575];U2M[0:1:23][0:1:89][0:1:575];V2M[0:1:23][0:1:89][0:1:575]&dap4.checksum=true'