You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
monitoring + bootstrap: track user-facing availability, expose Garage via ingress
monitoring-rules: replace kube_deployment_status_replicas_unavailable > 0
with (spec_replicas - status_replicas_available) > 0 across all
*Unavailable alerts (mail, identity, gitlab, files, notes, websites,
dns, home, ntp, monitor-the-monitoring). The old metric counts surge
pods from a stuck rolling update against the deployment, firing
major_outage on the statuspage even when desired replicas are all
ready. The new expression compares ready replicas against spec — fires
only when actual user-facing capacity is degraded.
bootstrap: new Ingress garage-bootstrap.mdapi.ch -> garage:3900 (no
ModSec, unlimited body size, 600s timeouts). GitLab container registry
will be repointed to this endpoint; presigned blob URLs sign against a
publicly resolvable hostname so containerd's host-netns DNS can resolve
them. Previously presigned URLs referenced garage.bootstrap.svc.cluster.local
and every fresh layer pull failed with 'no such host' on the node.
0 commit comments