Troubleshooting

Troubleshooting

Every failure listed here was hit while building this lab, not copied from a manual.

The VM will not create

RequestDisallowedByPolicy: Resource 'devstack' was disallowed by policy

An Azure Policy restricts what your subscription may create. The message lists the allowed sizes. Pick a Dv3 or later from that list, and switch the OS disk to StandardSSD_LRS when premium disks are refused too.

stack.sh dies on nova-cell

ValueError: invalid literal for int() with base 10: 'GzDjuJJ'

The password contains a character that breaks a URL. DevStack injects it into rabbit://user:PASSWORD@host, and a / makes the parser read part of the password as a port number.

Generate a URL safe one:

openssl rand -hex 24

make preflight now refuses anything that is not alphanumeric.

stack.sh fails, and fails again on retry

DevStack has no partial rollback. Always wipe before retrying:

make reset
make install

After three failures, destroy the VM and start over. It is faster than debugging a half built stack.

The playbook cannot become the stack user

chmod: invalid mode: 'A+user:stack:rx:allow'

The acl package is missing on the target. Ansible falls back to a chmod syntax that does not exist on Linux. The playbook installs acl for this reason.

The openstack client hangs with no error

The catalog points at the private address of the host:

openstack catalog list    # http://10.0.0.4/identity

Authentication works because it goes straight to auth_url, but every call after it resolves an unreachable address and waits forever. Run OpenStack commands on the host, which is what the scripts do.

A prompt aborts immediately

xx aborted

make swallows stdin, so a plain read gets an empty answer. The scripts read from /dev/tty instead. In a runner with no terminal at all, pass YES=1.

ssh to the instance is refused

Terraform returns as soon as Nova reports ACTIVE, well before CirrOS has started sshd. Wait a minute, or let make demo retry for you.

To see the boot as if you were in front of the screen:

openstack console log show web-01 | tail -20

That log also prints the CirrOS default password, which is only needed for the VNC console since SSH uses your key.

Ping fails but SSH works

Expected. The security group allows TCP 22 and 80, not ICMP. It is a good demonstration that the firewall does its job.

stack.sh gets killed halfway

Memory. 8 GB is DevStack's floor, and the OOM reaper takes the process during the pip and MySQL phases. The playbook adds a 4 GB swap file and disables Cinder for this reason.

free -h    # on the host, during the run