This should enable the cluster client to properly switch databases and user credentials for all clients in the cluster, not just a random one. It doesn't look pretty and actually I'd like to suggest an alternative implementation of the InfluxDBClusterClient which will actually inherit the InfluxDBClient and instead of maintaining a list of InfluxDBClients, we can just keep the list of (host, port) tuples and alternate those on each request. This way we win a few things:

• Less memory usage, obviously. There's no need to create 3 InfluxDBClient objects when we can get away with a single one;
• We get access to all attributes of the InfluxDBClient (not just the callable ones);
• Docstrings available for all methods. It also means that auto-completion will be available for developers;
• Easier maintenance and avoiding bugs similar to this one.

Since some users of the library might be using the clients attribute of the cluster client current implementation, we can make the new one backwards compatible by having a list with only one element (self) as the clients attribute, i.e. self.clients = [self]

@aviau , @gst, @areski , any thoughts on that?