Ask Your Question

Revision history [back]

The compute node seems to be unable to access Keystone at http://controller:35357/v3/auth/tokens. ECONNREFUSED indicates that the controller does not block port 35357, but no process is listening at that port on the controller. Most likely Keystone is not running.

Thus, the next step is checking if Keystone is running on the controller, and if not, why. The Keystone log is the first place to check.

If it is running, use curl to access that URL from the compute node and try to figure out where the Connection Refused error comes from.

Another possibility is that http://controller is not what you think. Perhaps the compute node configures an incorrect IP address for the controller.

Since nova-compute can't authenticate, it can't access the Placement API.

"TypeError: 'NoneType' object has no attribute '__getitem__'", looks like a bug or a mismatch between the versions of two Nova components. Perhaps it's a consequence of the other errors, perhaps it's a bug in the error handling code. I would first fix the Keystone access problem; the other errors might go away then.

The compute node seems to be unable to access Keystone at http://controller:35357/v3/auth/tokens. ECONNREFUSED indicates that the controller does not block port 35357, but no process is listening at that port on the controller. Most likely Keystone is not running.

Thus, the next step is checking if Keystone is running on the controller, and if not, why. The Keystone log is the first place to check.

If it is running, use curl to access that URL from the compute node and try to figure out where the Connection Refused error comes from.

Another possibility is that http://controller is not what you think. Perhaps the compute node configures an incorrect IP address for the controller.

Since nova-compute can't authenticate, it can't access the Placement API.

"TypeError: 'NoneType' object has no attribute '__getitem__'", looks like a bug or a mismatch between the versions of two Nova components. Perhaps it's a consequence of the other errors, perhaps it's a bug in the error handling code. I would first fix the Keystone access problem; the other errors might go away then.

Etcd, MySQL, MemCache, RabbitMQ are normally installed on controllers, not on compute nodes.