> I didn't realize the serialization prevented cloudinit from restoring a ds
> from the obj.pkl.
IIRC the serialization problem prevented cloudinit from *saving* the ds. The "write the ds" code tried to serialize and deserialize the ds before adding it to the larger object to be pickled; since the serialize raised an exception, the function just dropped the ds on the floor and it never wound up in the obj.pkl.
> Looking at the bug you reference, I can see why; it would have been good to
> have a bug for the Azure one as well so we could track that.
I'm writing up the bug now (for the _unset network_config problem), and I'll attach collect-logs output. Your request for "bug first, then fix" is totally legit, and I was just being lazy, and you called me on it.
> I didn't realize the serialization prevented cloudinit from restoring a ds
> from the obj.pkl.
IIRC the serialization problem prevented cloudinit from *saving* the ds. The "write the ds" code tried to serialize and deserialize the ds before adding it to the larger object to be pickled; since the serialize raised an exception, the function just dropped the ds on the floor and it never wound up in the obj.pkl.
> Looking at the bug you reference, I can see why; it would have been good to
> have a bug for the Azure one as well so we could track that.
I'm writing up the bug now (for the _unset network_config problem), and I'll attach collect-logs output. Your request for "bug first, then fix" is totally legit, and I was just being lazy, and you called me on it.