Dashboards

Four ready-made Grafana dashboards ship in the repo under grafana/provisioning/dashboards/json/PowerScale/, and provision into a PowerScale folder:

Dashboard	uid	Focus
OneFS Overview	powerscale-overview	Capacity, performance, protocols — day-to-day health.
OneFS Advanced	powerscale-advanced	Node/drive health, data protection, cache efficiency, quota & CPU detail.
OneFS Capacity & SLA	powerscale-capacity-sla	Availability/latency SLIs and capacity headroom with a days-to-full forecast.
OneFS Workloads	powerscale-workloads	Per-workload operations, throughput and CPU. Requires a configured OneFS performance dataset.

Because the powerscale_ prefix matches dell/csm-metrics-powerscale, existing CSM dashboards also work against this exporter without modification.

Every panel carries a description (hover the ⓘ); provisional panels note that their stat keys are not yet live-validated against OneFS.

Folders

Each directory under json/ becomes a Grafana folder (foldersFromFilesStructure: true in dashboards.yml), so boards arrive filed instead of piled into General:

json/
├── PowerScale/       → folder "PowerScale"     — the four boards below
└── Infrastructure/   → folder "Infrastructure" — third-party boards (Node Exporter Full)

Adding a board is a copy into the right directory; no config change. A file left directly in json/ lands in General, which is the signal that it needs a home.

Navigation

The boards are cross-linked, so you move between them without losing context.

Between boards. Every PowerScale board carries a PowerScale dashboards dropdown in its top-right corner, listing the others. It is driven by the powerscale tag rather than a hardcoded list, so a new tagged board joins every menu automatically. includeVars and keepTime are set, so the selected time range and any template variable the destination board also defines (cluster, and node where present) carry over; a variable the target board does not declare is simply dropped.

Into node detail. The per-node panels on Overview (CPU Idle, Memory Used, Disk IOPS, Used Capacity) and on Capacity & SLA (Per-Node Used Capacity) carry a data link — click a series and pick Inspect node N in OneFS Advanced to land on the node-detail board scoped to that node:

/d/powerscale-advanced?var-cluster=${__field.labels.cluster}&var-node=${__field.labels.node}&${__url_time_range}

The link targets the uid only, with no title slug, so renaming a board never breaks it.

Overview dashboard

One comprehensive board with these rows (in display order):

• SLI Summary — clusters up, per-cluster up/down status, detected OneFS API version, last-scrape age, and NFS export / SMB share / snapshot counts.
• Capacity — Utilization & Saturation — capacity used %, a used/available/free/total timeseries, and a table of the top quotas (usage vs hard limit).
• Compute — Utilization — cluster system / user / idle percent.
• Network & Disk — Utilization — external in/out throughput and cluster disk IOPS.
• Protocol — Rate & Duration (RED) — operations and average latency, broken down by protocol and op.
• Per-Node Detail (collapsed by default) — node CPU idle, memory used, disk IOPS, and used capacity.
• Licensing — per-feature license status (colored) and a days-to-expiry table (soonest first; red under 30 days).

Template variables

Variable	Source
datasource	Any Prometheus datasource — so the board works on import, not just in the bundled stack.
cluster	label_values(powerscale_up, cluster) — multi-select, includes All.
node	label_values(powerscale_node_cpu_idle_percent{cluster=~"$cluster"}, node) — drives the Per-Node Detail row.

Per-second gauges

IOPS / throughput / protocol panels read per-second gauges directly with sum/avg — they intentionally do not use rate(). See the Metrics Reference.

Advanced dashboard

The Advanced board surfaces the health/state and efficiency metrics, with a link back to the Overview board.

Validated rows

• Cluster Health — nodes read-only / smartfailing, drives by state, active events by severity, SyncIQ policies failed.
• Data Protection — SyncIQ policy table (enabled + last-run-failed) and snapshot space used.
• Node CPU Detail — per-node sys / user / idle.
• Quota Detail — logical + physical usage vs advisory / soft / hard thresholds.

Provisional rows (collapsed by default)

The following rows are collapsed by default because their underlying stat keys or schemas were validated against the OneFS 9.14.0 API specification but (except where noted) not yet confirmed against a live cluster.

• Cache Efficiency — L1/L2/L3 read hit-vs-miss and a computed hit-ratio, from the node-scoped node.ifs.cache.* keys. Both the key names and their unit semantics (per-second bytes/second rates) are now confirmed against a live cluster via --trace (see Metrics Reference); the row stays collapsed by default alongside the other advanced rows.
• Storage Efficiency — deduplication logical saved / deduplicated bytes (provisional).
• Per-Drive — top drive IOPS and busy % (provisional).
• Per-Client — operations by protocol/class and throughput in/out (provisional).
• Hardware — power-supply failures, node temperature and fan speed (provisional).
• OneFS 9.15 — protocol latency by processing stage and top-10 quota data-reduction ratio; panels stay empty on older clusters (provisional).

Capacity & SLA dashboard

A focused planning board built only on live-validated metrics (no provisional keys), in two sections:

• SLA — Availability & Error Budget — a stat strip (cluster availability %, scrape freshness, nodes read-only / smartfailing, SyncIQ failures, active critical events).
• SLA — Latency, Throughput & Errors (RED) — protocol latency (Duration), protocol operations (Rate), and active events by severity (Errors).
• Capacity — Headroom & Forecast — capacity used % gauge, a Days to Full stat and a 7-day forecast line (both from predict_linear/deriv over the trailing 24h trend), per-node capacity balance, snapshot space, and a quota table ranked by closeness to the hard limit.
• Storage Pools — Capacity — per node-pool/tier table (used %, used/total/available, and the SSD vs HDD media split), a node-pool used-% trend, and SSD-vs-HDD available capacity. The list holds both node pools and tiers; filter type="nodepool" for a non-overlapping cluster total.
• Licensing — min days to license expiry and a count of expired licenses.

Forecast is a trend projection

Days-to-full and the +7d line extrapolate the trailing 24h growth rate; a flat or shrinking trend reads as effectively "never". Use a longer dashboard time range (default now-7d) so the projection has history to draw from.

Workloads dashboard

Per-workload performance from OneFS statistics summaries (uid powerscale-workloads):

• Operations/sec, Throughput (in/out), and CPU µs/sec timeseries, one series per workload, plus a Workload Snapshot table (instant, sorted by ops/sec).
• Template variables cluster / zone / protocol / username slice the view — use them to bound the series count on a broad dataset.

Requires a performance dataset

Workload rows are produced only when a OneFS performance dataset is configured (isi performance datasets). Without one, this board is empty — every other dashboard is unaffected. Cardinality is governed by your dataset definition; the exporter exposes a fixed label set (node, zone, protocol, username, system_name, job_type) and omits the unbounded dimensions (path, IPs, SIDs).

Per-second gauges

Operations, throughput and CPU are per-second gauges — aggregate with sum/avg, never rate().

Auto-provisioned (compose stack)

The local stack provisions both the Prometheus datasource and this dashboard. docker-compose.yml mounts ./grafana/provisioning into Grafana, and grafana/provisioning/dashboards/dashboards.yml loads every JSON under json/:

PSCALE1_PASSWORD='your-monitor-password' docker compose up -d --build
# Grafana: http://localhost:3000  (admin / admin)
# Dashboards → PowerScale → OneFS Overview

Manual import (existing Grafana)

1. Dashboards → New → Import.
2. Upload grafana/provisioning/dashboards/json/PowerScale/powerscale-overview.json (or paste its contents).
3. Pick your Prometheus datasource when prompted, then Import.

Import all four to keep the cross-board dropdown populated — it lists whichever powerscale-tagged boards exist in your Grafana, so a partial import yields a partial menu. Choose a folder at import time; the folder mapping above applies to provisioning only.

Customising

The provider sets allowUiUpdates: true, so you can tweak panels in the Grafana UI. To persist changes, export the dashboard JSON (Share → Export → Save to file) and replace the file in the repo. Keep the uid stable so links and provisioning stay consistent.

Node Exporter Full (Grafana 1860)

This repo bundles the community Node Exporter Full dashboard (node-exporter-full.json, auto-provisioned). It visualizes host OS metrics (CPU, memory, disk, network) exposed by prom/node-exporter — not this exporter's own metrics.

node_exporter is not part of this demo stack: it belongs on the hosts you actually want to monitor, not bolted onto the exporter's compose. To use this dashboard, run prom/node-exporter on those hosts and add a node-exporter scrape job to your Prometheus; the dashboard then visualizes them.