FWIW, tests are not, by default, run in the default namespace, so you don't really have to switch namespace here. It's already not the default. Although, I guess it might make it more readable to show that the namespace is different explicitly.

I noticed in the customer's example that they don't explicitly pass namespace. I would think that the namespace would be implied by passing in parent. For grins, I tried removing the namespace arg from this test and I get a namespace mismatch exception in the model constructor. Is there maybe a little more work to do here?

Well, I confess I added this to make the test pass, but the customer's code does run unchanged if using a python script outside of the test machinery. Maybe I'm missing something on the test setup?

It looks like it's just because in the customer's example the context's namespace is still the default namespace when that entity is created, so there won't be a mismatch between entity2 and its parent. Does still seem like specifying parent should imply the namespace, though.

I know you're starting your new job. Would you like me to take this one?

Yes, please take this. I appreciate the offer.

FWIW, the following code caused the mismatch error before the fix was applied, and it works fine with the fix in:

import os
from google.cloud import ndb

class Dummy(ndb.Model):
    foo = ndb.StringProperty(default='')

with ndb.Client().context():
    entity1 = Dummy(foo='bar', namespace='')
    entity1.put()
    key_entity1 = entity1.key

    entity2 = Dummy(foo='child', parent=key_entity1)
    entity2.put()

    entity3 = Dummy(foo='childless')
    entity3.put()

with ndb.Client().context(namespace='other_context'):
    l = Dummy.query(ancestor=key_entity1, namespace='').fetch()
    for e in l:
        print(e)

I couldn't quite duplicate that in the text.