Comment 0 for bug 1271568

cinder-scheduler send a rpc message to cinder-volume which was killed just now.

Then when cinder-volume restart, it may receive the rpc message and begin to handle it before driver initialized.

Most APIs need to judge wether the driver has been initialized, so the rpc request would be reject. But those requests shoud be handle in fact.

The reason is, in service.py, we create consumer before manager call "init_host()", and we call "self.driver.set_initialized()" in init_host.