[CLI] [Recovery] Source/sink usernames/passwords used incorrectly

What's the issue?
The 'cbrecovery' command does not correctly use the source/sink usernames/passwords resulting in it using the wrong credentials to communicate with the clusters.

host, port, user, pwd, path = \

    pump.parse_spec(opts, sink, 8091)

#retrieve a list of missing vbucket

cmd = "startRecovery"

url = "/pools/default/buckets/%s/controller/%s" % (self.sink_bucket, cmd)

err, conn, response = \

    pump.rest_request(host, int(port), user, pwd,

                      url, method='POST', reason='start_recovery')

The important thing in this code snippet is:

1. We're communicating to the sink cluster
2. We're using 'opts' to get the username

def get_username(username):

    return username or os.environ.get('CB_REST_USERNAME', '')

...

username = get_username(opts.username)

Looking further down, we see the problem; we're using the same username option ('opts.username') when communicating to the source/sink cluster; the same applies to the password.

Why haven't we noticed until now?
I suspect during development both clusters had the same username/password.

What's the fix?
We should use the correct credentials when communicating with the source/sink clusters.

Is there a workaround?
Yes, use the same user on the source/sink cluster. This could be a temporary user if required.

Steps to reproduce

1. Create two, two node clusters a and b (with different usernames/passwords)
2. Create bucket a on cluster a (disable replicas)
3. Create bucket b on cluster b (disable replicas)
4. Set XDCR from cluster a to cluster b (bucket a -> b)
5. Load data into bucket a on cluster a (wait for XDCR to catch up)
6. Hard failover one of the nodes on cluster a (causing half vBuckets to be lost)
7. Use cbrecovery to recover vBuckets from b, to a