Comment 2 for bug 1264029

I was wrong about the first example - if the server is down then a socket.error (which is an IOError) is thrown by the underlying socket module, not a BadStoreUri error.

Regarding a catchall BadStoreUri, though it makes coding easier, I feel would it make debugging and troubleshooting more difficult. For e.g. Glance and the remote store are controlled by two different operators. Then, if the user encounters a failure for a Glance operation and only sees BadStoreUri, he doesn't know what to do and how to escalate the issue with the "store operator" - is it an authentication error, or that the object wasn't found or some permission error ... he doesn't know how to describe the problem to the store operator.

For that matter, he doesn't even know if it's a bug in the glance code logic that's giving rise to the problem. He has to guess what Glance is trying to do with the backend store and try to repeat that operation to see what error it was.

What do you think?