pflex_exporter

README

pflex_exporter

A Go exporter for Dell PowerFlex Gen1 clusters. It authenticates to the PowerFlex gateway REST API, collects the full Dell statistic set across all object types, and exposes the metrics via both a Prometheus /metrics endpoint and an OTLP metric push. It replaces Dell's Python + Telegraf + InfluxDB collection layer, following the architecture of nbu_exporter.

Features

• Dual export: Prometheus pull (/metrics) and OTLP metric push, fed from one shared snapshot.
• Full parity: all 7 object types — System (cluster), SDS, SDC, Device, Volume, StoragePool, ProtectionDomain.
• Multi-cluster: one process monitors many clusters; every metric carries a cluster label.
• Gen1-only: detects Gen1 (FineGranularity/MediumGranularity) vs Gen2 (ErasureCoding); Gen2 clusters are flagged and skipped without killing the process.
• Operational: per-cluster OAuth token lifecycle, graceful per-cluster degradation, hot config reload (SIGHUP + file watch), snapshot-based health endpoint, optional OTLP tracing.

Architecture

A single background collection loop polls every cluster on collection.interval, derives metrics, and publishes an immutable snapshot (RWMutex pointer-swap). Both the Prometheus collector (Collect() reads the snapshot) and the OTLP exporter (observable gauges read the snapshot) serve from that snapshot, so API load is independent of the number of scrapers and push cadence.

PowerFlex gateways ──► ClusterClient (auth + REST) ──► collection loop ──► SnapshotStore
                                                                              │      │
                                                                  Prometheus ◄┘      └► OTLP push

Configuration

See config.yaml. Cluster passwords support ${ENV_VAR} interpolation or a passwordFile.

server:        { host: "0.0.0.0", port: "2112", uri: "/metrics", logName: "" }
collection:    { interval: "10s", timeout: "8s" }
opentelemetry:
  metrics:     { enabled: false, endpoint: "localhost:4317", insecure: true, interval: "10s" }
  tracing:     { enabled: false, endpoint: "localhost:4317", insecure: true, samplingRate: 0.1 }
clusters:
  - { name: flex-cluster1, gateway: <ip-or-host>, username: <monitor-user>, password: "${FLEX1_PASSWORD}", insecureSkipVerify: true }

A monitor-type PowerFlex user is sufficient. For PowerFlex 4.0+, gateway is the primary ingress IP (PowerFlex Manager UI).

Running

make cli
FLEX1_PASSWORD=... ./bin/pflex_exporter --config config.yaml
# metrics at http://localhost:2112/metrics, health at /health

# one-shot collection (no server):
./bin/pflex_exporter --config config.yaml --once

Docker / compose (exporter + Prometheus + OTLP collector):

FLEX1_PASSWORD=... docker compose up --build

Metric naming

• Names are pflex_<object>_<metric>: pflex_cluster_*, pflex_sds_*, pflex_sdc_*, pflex_volume_*, pflex_storagepool_*, pflex_device_*, pflex_protectiondomain_*.
• Bandwidth/IO accumulators (*Bwc) become three gauges with op and direction labels: _iops, _bandwidth_kb_per_second, _io_size_kb. Latency accumulators become _latency.
• Other statistics become scalar gauges named from the API field, e.g. pflex_cluster_max_capacity_in_kb, pflex_storagepool_spare_capacity_in_kb.
• Health/meta: pflex_up{cluster}, pflex_last_scrape_timestamp_seconds{cluster}, pflex_cluster_generation{cluster,generation}.

IOPS and bandwidth are already per-second gauges (derived from the API's counters). In PromQL, aggregate them with sum/avg — never rate().

Dashboards

grafana/ contains importable PromQL dashboards: pflex-cluster-overview.json and pflex-storage-pools.json. They demonstrate the metric/label conventions; additional per-object dashboards (devices, volumes, SDC, SDS, protection domains, capacity planning) can be built mechanically from the same naming scheme.

Development

make tools         # install golangci-lint, cyclonedx-gomod, govulncheck (pinned)
make sure          # fmt + vet + test + build + golangci-lint
make test-coverage # race tests + HTML coverage report

CI/CD & SBOM

Everything CI runs is a Makefile target, so it reproduces locally.

• CI (.github/workflows/ci.yml, on push/PR) runs make ci — gofmt check, go vet, golangci-lint, go test -race with coverage, and govulncheck — plus a Semgrep scan and an SBOM artifact.
• Release (.github/workflows/release.yml, on v* tags):
  • make release cross-compiles linux/{amd64,arm64} and darwin/{amd64,arm64} binaries, generates the SBOM, and writes checksums.txt; all are attached to the GitHub Release.
  • A multi-arch container image is pushed to ghcr.io/<owner>/pflex_exporter with build-time SBOM and provenance attestations.
• SBOM: make sbom produces a CycloneDX SBOM (dist/sbom.cdx.json) for the Go module via cyclonedx-gomod; the container image carries its own SBOM attestation.
• Version: injected at build time via -ldflags "-X main.version=$(VERSION)" (VERSION defaults to git describe); check with pflex_exporter --version.

Notes

• Supported baseline is PowerFlex 4.5+ (Gen1). Auth uses the 4.x bearer flow (/rest/auth/login, /rest/auth/update-token); older firmware that only offers Basic auth is out of scope.
• The PowerFlex API spells the occurrence counter numOccured; the exporter also accepts the corrected numOccurred.

License

Apache License 2.0 — see LICENSE. This matches the license of Dell's upstream PowerFlex Gen1 monitoring tooling that this exporter is derived from.

Documentation

Overview

PowerFlex Exporter collects metrics from Dell PowerFlex Gen1 clusters and exposes them via a Prometheus /metrics endpoint and an optional OTLP metric push.

Usage:

pflex_exporter --config config.yaml [--debug] [--once]