A municipal Container Apps revision for a permitting assistant keeps restarting. Logs show the model endpoint environment variable is missing. What should ops do?

When Container Apps logs show a permitting assistant restarting because the model endpoint environment variable is missing, the fix is to add the required environment variable or Key Vault-backed secret reference to the revision configuration and redeploy. Crash loops from missing config are configuration defects, not capacity defects. Correct revision settings let the container start, resolve the endpoint, and serve municipal permit assistance.

Increasing CPU and memory tenfold without supplying the missing endpoint fails by concept because extra resources cannot create a missing URL; the process will still exit on required config validation. Silencing restart alerts and leaving the crash loop overnight fails because it ignores an active outage and abandons incident response. Moving the app to a resident's personal Minikube without fixing configuration fails because the same missing endpoint still breaks startup on any host, and personal clusters are not a municipal remediation.

Exam caveat: Prefer secret references for sensitive endpoint keys while non-secret URLs may be plain env vars—either way the revision must define them. Remember new revisions are created when configuration changes; ensure traffic targets the fixed revision. Distinguish missing env (config) from ImagePullBackOff (pull) and OOMKills (resources).

Operational check: Add the model endpoint setting (or secret reference) to the Container Apps revision, redeploy, confirm the permitting assistant replicas reach running state without restart loops, and verify a test permit question reaches the model. Check alerts clear and logs no longer report missing endpoint variables.