Replace N/A Info metrics with actionable V3 health signals

[Vonng]

Summary

Metrics → Info currently displays these two rows permanently as n/a:

• Time since last Heal Activity
• Time since last Scan Activity

This is not only a missing-series problem. BasicDashboard.tsx currently hardcodes both values to "n/a", so fixing the advanced widget queries alone cannot populate the Info page.

The two legacy time-since values should be replaced with higher-signal V3 summaries and the Info page should consume the same widget data as the advanced dashboard.

Proposed replacement

1. Erasure Health

Use:

min(minio_cluster_erasure_set_overall_health{$__query})

Display:

• 1 → Healthy
• 0 → Unhealthy
• missing → Unknown

Using min is conservative when every cluster node is scraped: any unhealthy view should make the summary unhealthy.

2. Usage Data Age

Use:

max(minio_cluster_usage_objects_since_last_update_seconds{$__query})

Display the result as a duration in seconds, for example 1 minute or 3 hours. Missing data before the first usage update should render as Pending, not zero.

This is more relevant than node-local scanner activity because it directly explains how fresh the adjacent Reported Usage value is.

The resulting summary becomes:

Erasure Health    Healthy
Usage Data Age    1 minute
Uptime            4 hours

Why not preserve Time Since Last Heal?

minio_heal_time_last_activity_nano_seconds is not available in V3 and is a weak top-level health signal:

• a healthy cluster may legitimately have no recent healing activity;
• recent activity does not prove that healing succeeded;
• the V2 series is conditional and can be absent;
• erasure-set health directly reports whether storage remains healthy.

minio_cluster_erasure_set_healing_drives_count is still useful, but belongs in the Drives card or Resources/Advanced view. It can be presented as Online / Healing / Offline rather than occupying the Reported Usage summary.

Implementation notes

• Replace widget IDs 80/81 with the new metric definitions and titles.
• Change the age formatter from nanoseconds to seconds.
• Replace the Heal/Scan-specific renderers with a generic status/summary renderer, or make them honor the supplied widget title.
• Make BasicDashboard fetch/reuse the same widget results instead of keeping its own hardcoded values or PromQL.
• Preserve explicit fallback states when Prometheus is unavailable.
• Honor the existing selector assembled from job and CONSOLE_PROMETHEUS_EXTRA_LABELS.

Acceptance criteria

• Metrics → Info no longer shows unconditional n/a values when V3 metrics are available.
• Metrics → Usage and Metrics → Info use the same semantic definitions.
• A healthy single-node deployment renders Healthy and a finite usage age.
• A multi-node deployment produces one conservative cluster summary without duplicate target series.
• Missing first-scan data renders Pending; unreachable metrics render Unknown or Unavailable.
• No new compatibility-only MinIO metric is required.

Related: #2, #7

[Vonng]

Implemented in v2.1.0 — query definitions in e964992, Info-page rewiring in 6a03719.

Resolution

The issue's core observation was the important one: this was never only a missing-series problem. BasicDashboard.tsx destructured lastScan, lastHeal, and upTime from an empty object literal, so no amount of query fixing could have populated the Info page. Both halves are addressed.

Widgets 80 and 81 were replaced, not repaired, along the lines proposed:

	Query	Rendering
Erasure Health (80)	min(minio_cluster_erasure_set_overall_health{$__query}) or (max(minio_cluster_erasure_set_overall_write_quorum{$__query}) * 0)	1 → Healthy · 0 → Unhealthy · empty → Unknown
Usage Data Age (81)	max(minio_cluster_usage_objects_since_last_update_seconds{$__query})	duration in seconds; < 1 h ok, 1–24 h warning, > 24 h danger; empty → Unknown

min() for erasure health is the conservative reading the issue asked for: with every node scraped, any unhealthy view makes the cluster summary unhealthy.

One correction to the proposal is worth recording, because it is exactly the trap this migration kept hitting: min(...overall_health) alone cannot ever report Unhealthy. V3 skips exporting any sample with value ≤ 0, so the unhealthy 0 is absent, and the panel would go blank precisely when the cluster is degraded. The or (max(...overall_write_quorum) * 0) clause supplies the 0 — write quorum is present whenever the erasure-set group is exported at all, so it distinguishes "unhealthy" from "not scraped". Without it the card would have been silently useless in the only case it exists for.

BasicDashboard now consumes the same widget results instead of holding its own values: it looks up panelsConfiguration entries 80 and 81 and renders them through the shared panel renderer, so Info and Usage cannot diverge. Uptime is derived from the real server list (min() across online nodes) rather than a hardcoded "n/a" — see #2. When Prometheus is not configured the two rows are hidden; when it is configured but unreachable they render Unknown.

The age formatter was moved from nanoseconds to seconds, and the heal/scan-specific renderers were replaced with a generic status/summary renderer that honors the supplied widget title.

Acceptance criteria

• Info no longer shows unconditional n/a when V3 metrics are available
• Usage and Info use the same semantic definitions — literally the same widget results
• A healthy single-node deployment renders Healthy and a finite usage age
• Multi-node produces one conservative cluster summary without duplicate target series (min()/max() dedup)
• Missing first-scan data and unreachable metrics render explicit states, not zero
• No compatibility-only MinIO metric was required

minio_cluster_erasure_set_healing_drives_count was not promoted into the summary, per the issue's own reasoning; it remains available for the Drives card / Resources view.

Documented in docs/metrics-v3.md.

Closing as delivered in v2.1.0.