How to open a in-memory tmp rasterio object?

[SongshGeo]

The smallest case:

First, I created an in-memory rasterio file object by:

import pyproj.crs
import affine

data = np.random.random((1, 4, 4))
crs = pyproj.crs.CRS.from_user_input("epsg:4326")
transform = affine.Affine(*np.random.random(6))

with rio.MemoryFile() as mem_file:
    with mem_file.open(
        driver="GTiff",
        height=data.shape[1],
        width=data.shape[2],
        count=data.shape[0],  # number of bands
        dtype=str(data.dtype),
        crs=crs,
        transform=transform,
    ) as dataset:
        dataset.write(data)
    # Open the dataset again for reading and return
    raster = mem_file.open()

type(raster)
>>> rasterio.io.DatasetReader
raster
>>> <open DatasetReader name='/vsimem/9a559d40-652a-4c8c-b9ba-28110febdd37/9a559d40-652a-4c8c-b9ba-28110febdd37.tif' mode='r'>

According to the document, I think DatasetReader object is a legal input, so I tried to open it:

rioxarray.open_rasterio(raster, driver="GTiff")

>>> File rasterio/_base.pyx:310, in rasterio._base.DatasetBase.__init__()
>>> RasterioIOError: No such file or directory

What I did:

1. Searched existing issues and discussions by keyword in-memory. However, I need help finding a similar case.
2. Creating the rasterio object without context manager.

Versions:
xarray = "^2023.8.0"
xarray = "^2023.1.0"
rioxarray = "^0.13.3"

[snowman2]

The main issue in your code is that you are opening the raster outside the with block. Once you do that, the memory file ceases to exist. If you open it inside the with block, it works:

import affine
import numpy
import pyproj.crs
import rasterio
import rioxarray

data = numpy.random.random((1, 4, 4))
crs = pyproj.crs.CRS.from_user_input("epsg:4326")
transform = affine.Affine(*numpy.random.random(6))

with rasterio.MemoryFile() as mem_file:
    with mem_file.open(
        driver="GTiff",
        height=data.shape[1],
        width=data.shape[2],
        count=data.shape[0],  # number of bands
        dtype=str(data.dtype),
        crs=crs,
        transform=transform,
    ) as dataset:
        dataset.write(data)
    # Open the dataset again for reading and return
    with mem_file.open() as raster:
        rds = rioxarray.open_rasterio(raster, driver="GTiff")