I'd like to analyze some raster data in Python using NumPy.

This data doesn't necessarily exist as a file on the filesystem, it could be the result of a query against PostGIS, or something transferred over the network.

Is it possible to get GDAL to read data from a buffer in memory, instead of requiring a filename?

I realize one solution could be to open up a named temp file, write the data to that, and pass it as a filename to GDAL, but that seems inefficient, and more likely to incur I/O bottlenecks. Additionally, in the case of PostGIS, it would be nice to read the native format directly, as opposed to converting it to some other file format and reading that.

Edit: to clear up a bit of confusion, just using the PostGIS driver directly won't work either, since the query could be for a raster analysis, rather than just data stored in the DB.