W'xOps Portal v0.5.1 — Observability Completion

v0.5.0 gave the portal eyes on the cluster: live ArgoCD and Crossplane status, one-click Grafana links, active alerts. What it couldn't tell you was how a service was actually performing — those Grafana links pointed at cAdvisor CPU and memory, not application data, because nothing in the platform emitted a real metric. v0.5.1 closes that gap.

This is also the first release that isn't portal-only. It spans wxops-templates, wxops-core, wxops-gitops-infrastructure, and this repo, in a hard dependency order — the portal's Monitoring toggle only does something once the other three are deployed too.

Real application metrics

Every golden-path template — Go, Node.js, Python — now emits a consistent metric contract instead of a hand-rolled /metrics with two runtime gauges:

Metric	Type	Labels
http_requests_total	counter	method, path, status
http_request_duration_seconds	histogram	method, path
http_requests_in_flight	gauge	—

path is always the route pattern, never the raw URL — this platform's Prometheus runs with no cluster-wide cardinality limits, so that's the one rule that matters more than any other here.

The Monitoring toggle was silently broken — now it isn't

The scaffold wizard and Edit Config have had a Monitoring checkbox for a while. Enabling it wrote prometheus.io/scrape, prometheus.io/port, and prometheus.io/path pod annotations — a convention Prometheus only honors through a kubernetes_sd_config scrape job with matching relabel rules. This platform's kube-prometheus-stack has an empty additionalScrapeConfigs. The toggle had done nothing since it shipped.

It now writes spec.parameters.monitoring directly on the XTenantApp, and the Crossplane composition turns that into a real ServiceMonitor (or PodMonitor) carrying the release: kube-prometheus-stack label Prometheus Operator actually selects on — without that label, Prometheus silently ignores the monitor too. Port and scrape interval stay composition-owned; the portal writes only enabled and, optionally, a custom path.

One dashboard, every service

A single shared Grafana dashboard — P95/P99, error rate, request rate, in-flight saturation — parameterized by $namespace/$app/$env, works for any golden-path service with zero per-team configuration. The portal deep-links to it with variables pre-filled.

Known gap, not hidden: $env is a naming-suffix convention derived from $app, not a real Prometheus label — the ServiceMonitor sets no targetLabels. A proper environment label is a tracked fast-follow, not a "someday."

Catalog completeness score

Every entity now carries a computed quality badge — description, owner, tags, links, lifecycle, and, for API entities, whether a spec is actually resolvable. It's computed fresh on every read; nothing is stored or cached separately from the entity itself, so the score can never drift from what's actually committed in gitops-infra.

It matters most for manually registered or older entities. Anything scaffolded through the golden path already fills in every one of these fields by default, so a low score on a scaffolded service is a signal something was edited out afterward — not that scaffolding missed it.

The Pipeline tab, rebuilt for real scale

The CI/CD, Releases, and Dependencies cards were designed around small projects and broke down for exactly the case this platform is meant to serve: daily CI activity, dozens of releases, hundreds of transitive dependencies.

• CI runs were hard-capped at 5, with a "Show more" toggle that only ever revealed already-fetched runs — for a team shipping several times a day, that window was exhausted before lunch.
• Releases were capped at 10, same fake-pagination pattern, no way to search.
• Package manifests were silently truncated at 200 entries with no indicator — and because one parser built its list from Go map iteration, which is randomized per process, a package.json with more than 200 dependencies kept a different random subset on every single page load.

Each card is now a compact summary with a View all → action that opens a search and filter view. CI gets status and branch filters plus real backend pagination past the old 5-run ceiling. Releases gets an instant tag search — the same picker now reused by the dependency-compare control, replacing what used to be a bare, unsearchable <select> capped at 10 tags. Dependencies gets instant search across name, ecosystem, and version, plus direct/dev/indirect filtering.

Package manifests now report truncated/total instead of dropping entries silently, the cap is raised 200 → 500, and packages are sorted before truncating — so which packages survive the cap is now deterministic instead of re-randomized on every reload.

The dependency-compare diff is also grouped per manifest now, closing a real correctness gap: a monorepo with more than one manifest in the same ecosystem — this portal's own cli/go.mod alongside backend/go.mod — could previously have one manifest's diff silently overwritten by the other's, since both were keyed into one shared map by name alone.

Upgrading

No new required portal environment variables.

This release only does something if the other three repos are deployed too. The Monitoring toggle produces a correctly-shaped field on the XTenantApp regardless of whether wxops-core's composition and wxops-gitops-infrastructure's dashboard are live — it just won't emit a working ServiceMonitor or have anywhere to point until they are.

Existing scaffolded services don't retroactively get any of this. A service scaffolded before v0.5.1 keeps its old near-empty /metrics and its old, non-functional monitoring annotations until it's rescaffolded or manually updated.

Also in this release

• New page/limit query parameters on the CI and releases endpoints — additive, so omitting them reproduces the exact previous behavior.
• The catalog's entityResponse gains completenessScore: { score, max, checks }.
• Dependency manifests gain truncated/total; the compare endpoint's diff shape is now grouped by manifest instead of one flat list.

Full detail, including the known limitations this surfaced, is in Runtime Observability and Service Catalog.