Skip to content

Serving correctness: alert routing, readiness, and proxy wiring #392

Open
6 of 6 issues completed
Open

Description

@amuraru

Outcome

Existing supported deployments route traffic to usable endpoints and preserve a single effective mutation domain. Consolidates #373/#374/#384/#385 under #368.

Scope and acceptance

  • Preserve current leader-election identity. Document exact namespace-input/election-namespace requirements and unsupported overlaps; do not canonicalize hashes or introduce maintenance-migration/ownership machinery.
  • Alert serving reports bind errors, shuts down with the manager, and receives Service traffic through leader failover. Use leader-owned Pod routing labels without changing webhook selectors. Preserve custom/non-Helm listeners when both new identity fields are absent; reject partial identity. Coordinate with open fix: restore alert-driven broker downscale and safe monitoring defaults #365; do not duplicate its payload/downscale work.
  • Readiness reflects usable enabled webhook/alert serving as appropriate. Preserve webhook-serving standbys and Kafka-independent liveness; do not make all nonleaders globally unready.
  • Correct PodMonitor port/protocol, proxy upstream/raw-loopback wiring, and proxy review-role bindings to the actual Pod ServiceAccount. Keep kube-rbac-proxy and existing modes. Document installation-owned protected-scrape prerequisites; managed credential/CA/server-name integration is explicitly deferred.

Limits

No native-metrics migration, new certificate-renewal mode, warmup, higher concurrency, or new tuning switches. Generated serving certificates still expire after 365 days; document existing renewal procedures and the limitation rather than claiming automatic renewal.

No dependency on a broad metrics/benchmark project. Historical children are superseded; their closure is not implementation completion. Routing labels are not fencing or an exactly-once guarantee.

No new certificate/credential fields or issuance machinery. Do not disable TLS verification or claim turnkey protected scraping.

Validation

Focused failing regressions, existing chart tests and make check, plus real multi-replica Kind traffic/failover, raw-port isolation and proxy review-permission checks. End-to-end managed scrape authentication/TLS is deferred, not established by rendered ports. Keep the existing lock identity unchanged. Record exact results and limitations in linked PRs.

Activity

  1. added theissue type on Nov 16, 2026
  2. changed the title [-]Workstream: HA serving, certificates, and metrics security[/-] [+]Serving correctness: leader identity, alerts, readiness, and metrics[/+] on Nov 17, 2026
  3. changed the title [-]Serving correctness: leader identity, alerts, readiness, and metrics[/-] [+]Serving correctness: alert routing, readiness, and protected metrics[/+] on Nov 17, 2026
  4. changed the title [-]Serving correctness: alert routing, readiness, and protected metrics[/-] [+]Serving correctness: alert routing, readiness, and proxy wiring[/+] on Nov 17, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions