Error when updating numpy array after ray.get, "output array is read-only".

[robertnishihara]

The following fails.

import numpy as np
import ray
ray.init()

@ray.remote
def f():
  return np.zeros(100)

x = ray.get(f.remote())

x[0] = 1

The last line fails with the following error.

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-2-1a754504edbf> in <module>()
----> 1 x[0] = 1

ValueError: assignment destination is read-only

If instead of x[0] = 1 I do x += np.ones(100) it fails with the following error.

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-3-acaf936369e9> in <module>()
----> 1 x += np.ones(100)

ValueError: output array is read-only

[robertnishihara]

The problem is the following. When f finishes executing, it returns a numpy array of zeros. That array is stored in the object store.

When you call ray.get to get the array, the array is still stored in the object store (in particular, it is not copied into the Python processes heap). Objects in the object store are immutable, and so the array values cannot be changed.

If you wish to change the array values, then you'll need to create a copy of the array locally and mutate that one. For example, the following should work.

y = np.copy(x)
y[0] = 1
y += np.ones(100)