Phase 10 of the deploy-hardening I master bundle. Mirrors the
production-hardening-II Phase 8 OCSP-counter pattern. Per frozen
decision 0.9, the metric naming convention is
`certctl_deploy_<area>_total` with target_type + sub-label.
internal/service/deploy_counters.go:
- DeployCounters struct with sync.Map of per-target-type buckets
(apache, nginx, etc.). Lock-free fast path via sync/atomic
Uint64 counters; LoadOrStore on first tick.
- 8 sub-counters per target-type bucket:
- attemptsSuccess / attemptsFailure
- validateFailures (PreCommit returned error)
- reloadFailures (PostCommit returned error → rollback ran)
- postVerifyFails (post-deploy TLS handshake failed)
- rollbackRestored (rollback succeeded)
- rollbackAlsoFail (operator-actionable escalation)
- idempotentSkips (SHA-256 match → no-op deploy)
- Snapshot returns []DeploySnapshot for the Prometheus exposer.
internal/service/deploy_counters_test.go:
- 5 tests: zero-state, per-target-type tick isolation, race-detector
smoke under concurrent ticks, cross-target bucket isolation,
snapshot-mutation-doesn't-affect-counter.
internal/api/handler/metrics.go:
- New DeployCounterSnapshotter interface (mirrors CounterSnapshotter
for the OCSP counters but uses the per-target-type tuple shape).
- New DeploySnapshotEntry struct copying the service-layer shape;
avoids importing the service package directly so the handler
stays dependency-light.
- New SetDeployCounters setter on MetricsHandler (mirrors
SetOCSPCounters wiring).
- Prometheus exposer extended with 6 new metric blocks per frozen
decision 0.9:
- certctl_deploy_attempts_total{target_type, result}
- certctl_deploy_validate_failures_total{target_type}
- certctl_deploy_reload_failures_total{target_type}
- certctl_deploy_post_verify_failures_total{target_type}
- certctl_deploy_rollback_total{target_type, outcome}
- certctl_deploy_idempotent_skip_total{target_type}
- Output sorted by target_type for stable diffs across requests.
The agent-side wire-up (cmd/agent/main.go ticking counters in the
DeployCertificate dispatch site) is intentionally deferred to a
follow-up commit — Phase 10's load-bearing change is the
infrastructure; per-connector tick wiring is a mechanical follow-on.
Build + go vet clean. go test -count=1 green for service +
handler packages.
Phase 11 next: cross-cutting integration tests at deploy/test/.
shankar0123
135b271197