Curtin failure is not detected as an installation script failure

[Lee Trager]

When Curtin signals its status using the status handler. On Curtin completion it includes a value with the Curtin output. MAAS stores this file as a script result. However Curtin doesn't include a return code so the default, 0, is used. When 0 is used MAAS assumes the script result passed. The machine is still put into failed_deployment, the script result needs to be failed as well.

[Andres Rodriguez (andreserl)]

can you provide more logs, information on how this is causing an error ? If the machine was marked failed deployment even though the curtin log was passed successfully, there may be a different reason why it was marked failed deployment.

For example, curtin may work successfully, but cloud-init failed and because of that, marked the machine as failed.

That said, this is a log, not a script result. As such, this should not be treated as one with status codes.

[Lee Trager (ltrager)]

This is triggered whenever Curtin fails to deploy, I discovered it while testing CentOS updates which have since been fixed. MAAS has always stored installation logs as results. Previously they were stored as NodeResults, now ScriptResults. Both models store the return code and when none is given default to 0.

With ScriptResults MAAS also stores the status, if none is set its determined by looking at the return code. So because no return code or status is set the status is set to passed. The UI and API use ScriptSet.status to show the user what the overall status of commissioning, testing, and installation are. The result is while the node is still put into FAILED_DEPLOYMENT and the installation log shows an error, the UI displays a green check instead of a red x. Likewise

maas <profile> node-script-results read <system> type=installation | jq -r '.[0]["status_name"]'

Will always return 'Passed' even though one look at the log file shows that is not true.