Notes from operating platform infrastructure: Kubernetes, observability, incidents.
When --check deletes for real
An Ansible check-mode rehearsal deleted a Kubernetes Job. The API server ignored the dry-run flag because the request also had a body, and the rehearsal ran a different collection version than production.
Three silent uninstalls: what k3s does to your HelmCharts on every patch upgrade
Four incidents in four months had one cause: every k3s patch upgrade re-runs every HelmChart, and the default failure policy turns a failed upgrade into an uninstall within seconds.