mirror of
https://github.com/shankar0123/certctl.git
synced 2026-06-07 12:21:31 +00:00
shankar0123
9155ec9174
Sprint 6 ACQ DEPL-004 closure follow-up. CI run on commit 58a15e0
caught two issues:
1. The fail-closed guard in templates/servicemonitor.yaml used
`{{ required "msg" nil }}`, which is wrong Helm syntax — the
bareword `nil` isn't valid in Go templates and Helm interprets
it as no value, hitting "wrong number of args for required:
want 2 got 0". The B3-helm-chart-coherence ci-guard's
production-hardening render
(`--set monitoring.serviceMonitor.enabled=true` without
explicit tlsConfig) failed with this error AND with the
downstream "missing kind: ServiceMonitor / PodDisruptionBudget /
NetworkPolicy" cascades (the entire render aborted before
producing the matrix).
2. The original DEPL-004 framing — "operators MUST explicitly
choose tlsConfig or you get a chart-render error" — was the
right intent but the wrong default. The chart's existingSecret
integration mounts the CA bundle at a canonical path
(/etc/prometheus/secrets/certctl-ca/ca.crt); defaulting to that
path closes the implicit-skipVerify gap without forcing every
operator to repeat the same boilerplate.
Fixes
=====
deploy/helm/certctl/values.yaml — flips
monitoring.serviceMonitor.tlsConfig from commented-out (which fell
through to implicit insecureSkipVerify: true) to a real verify
default:
tlsConfig:
caFile: /etc/prometheus/secrets/certctl-ca/ca.crt
serverName: certctl-server
Operators with a different CA mount path override caFile;
operators who genuinely want skipVerify back must set
`{ insecureSkipVerify: true }` explicitly. Operators who blank
tlsConfig entirely (`tlsConfig: null` or `tlsConfig: {}`) still
trip the fail-closed guard.
deploy/helm/certctl/templates/servicemonitor.yaml — replaces
`required "msg" nil` with `fail "msg"`. The `fail` builtin is
the correct Helm pattern for an unconditional render-time error;
`required` is for "this value MUST be non-empty" which is the
wrong semantic here (we want to fail when the operator went OUT OF
THEIR WAY to blank the default). Failure message updated to
reflect the new default + the operator-action recipes.
docs/operator/helm-deployment.md — rewrites the
"2026-05-16 — ServiceMonitor TLS default flipped" subsection to
match the new default-on-real-verify semantics. The three operator
recipes (default install / different CA mount / explicit
skipVerify) are documented; the explicit "there is no way to
inherit pre-2026-05-16 implicit-skipVerify behavior silently"
guarantee is preserved.
Verified locally: python3 YAML parse on values.yaml clean; the
helm-templates-lint and B3-helm-chart-coherence ci-guards require
helm itself which isn't in the sandbox — both should pass on the
CI re-run.