Truncated netcdf data downloads

[ffang10]

Our user reported incomplete data download cases from this server

https://goldsmr5.gesdisc.eosdis.nasa.gov/opendap/

All cases appear to use the fileout netcdf handler for the .nc4 responses with subsetting, and mostly for the M2I3* datasets under 'MERRA2', which are very large granules themselves. A typical URL is:

https://goldsmr5.gesdisc.eosdis.nasa.gov/opendap/MERRA2/M2I3NPASM.5.12.4/2017/03/MERRA2_400.inst3_3d_asm_Np.20170324.nc4.nc4?T[0:7][0:41][0:360][0:575],H[0:7][0:41][0:360][0:575],O3[0:7][0:41][0:360][0:575],QV[0:7][0:41][0:360][0:575],PHIS[0:7][0:360][0:575],PS[0:7][0:360][0:575],lat[0:360],time[0:7],lon[0:575],lev

While these are large data requests that can overwhelm the server sometimes (presumably when the download incompleteness occur?), a data download would either be a success or failure (It appears that the httpd has status 200 in the access logs).

Caching in the server was turned on. I wonder if cache management in server can interfere with a download, given the pattern of the failures which were occasional and quasi-periodical.