Maintenance only. Keeps the deployment healthy, upgraded, backed up and
observable; drains a node before a change and restores a database after one. Has
no reason to read prompts, mint keys or change who may do what — but today the
screens they need are all superadmin-only, so they hold superadmin anyway.
Goal: “I know it is healthy, I can change it without an outage, and I can put
it back if the change goes wrong.”
Dogfood account:dev@rolter.local (superadmin). See
dogfood data.
The observer watches: the gateway’s /metrics and Cluster Config while
D2–D4 run, and SigNoz rolter · overview for anything that got slower.
what the new version will change in the stored configuration, before it runs
works
D1.4
upgrade
new image; migrations run on control-plane boot
the new control plane opens its listener only once migrations finish, so neither probe answers until then; the chart’s startupProbe (5 minutes by default) holds liveness off meanwhile; the migrations check in /readyz only fails for a process serving a database it did not migrate