engine="h5netcdf"

[brucelaughlin]

I was seeing suspicious "runaway" memory usage again (see old posts of mine about this, where you guys helped resolve it). I again modified reader_ROMS_native.py, adding "engine="h5netcdf"" in the "xr.open_mfdataset()" call (as you suggested before), and this again fixed the problem.

                self.Dataset = xr.open_mfdataset(filename,
                    chunks={'ocean_time': 1}, compat='override', decode_times=False,
                    preprocess=drop_non_essential_vars_pop,
                    data_vars='minimal', coords='minimal', engine="h5netcdf")

Version info:

commit 8022e94 (HEAD -> master, origin/master, origin/HEAD)
Merge: 84786ea 5f20104
Author: Knut-Frode Dagestad knutfrode@gmail.com
Date: Fri Jul 26 17:42:42 2024 +0200

Note: I noticed that in the new version I'm using, engine specification ("engine=....") is no longer an argument to xr.open_mfdataset(), whereas it was an argument in the version I was using when I originally had this problem.