Removing large zarr stores recursively blows up memory

I am trying to remove a rather large zarr store from the pangeo scratch bucket:

I am doing this on the pangeo google deployment (production)

import fsspec
import os

fs = fsspec.filesystem('gcs')
print(fs.ls('pangeo-scratch/jbusecke/detrending_test/'))

for path in fs.ls('pangeo-scratch/jbusecke/detrending_test/'):
    fs.rm(path, recursive=True)
    print(f'Finished with {path}')

This will take some time, over which memory usage on the notebook process increases steadily and eventually the kernel crashes.

Is this a misuse of fsspec (I am very new to all of this).

Or is this a bug that might have a workaround? Any help to debug this would be greatly appreciated.

Start by reproducing the notebook snippet with fsspec's gcs filesystem and the recursive fs.rm(path, recursive=True) call. Trace the recursive removal path while watching notebook-process memory; done means large zarr stores can be removed without steady memory growth or a kernel crash.