This repository contains a Go library to list your unused persistent disks in different cloud providers, and binaries for displaying them at the CLI or exporting Prometheus metrics.
At Grafana Labs we host our workloads in multiple cloud providers. Our workloads orchestration is managed by Kubernetes, and we've found that due to some misconfiguration in the backend storage, we used to have lots of unused resources, specially persistent disks. These leaked resources cost money, and because these are resources that are not in use anymore, it translates into wasted money. This library and its companion tools should help you out to identify these resources and clean them up.
This module exports some interfaces and implementations to easily list all your unused persistent disks in GCP, AWS, and Azure. You can find the API in the Go package documentation.
This repository also provides two binaries ready to use to interactively view unused disks (cmd/unused) or expose unused disk metrics (cmd/unused-exporter) to Prometheus.
Both programs can authenticate against the following providers using the listed CLI flags:
| Provider | Flag | Description |
|---|---|---|
| GCP | -gcp.project |
ID of the GCP project |
| AWS | -aws.profile |
AWS configuration profile name |
| Azure | -azure.sub |
Azure subscription ID |
These flags can be specified more than once, allowing to have different configurations for each provider.
Both binaries are opinionated on how to authenticate against each Cloud Service Provider (CSP).
| Provider | Notes |
|---|---|
| GCP | Depends on default credentials |
| AWS | Uses profile names from your credentials file or AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, and AWS_REGION env variables |
| Azure | Either specify an AZURE_CLIENT_ID, AZURE_CLIENT_SECRET, and AZURE_TENANT_ID, or requires Azure CLI installed on the host and signed in |
TUI tool to query all given providers and list them as a neat table. It also supports an interactive mode which allows to select and delete disks in an easy way.
go install github.com/grafana/unused/cmd/unused@latest
Web server exposing Prometheus metrics about each providers count of unused disks. It exposes the following metrics:
| Metric | Description |
|---|---|
unused_disks_count |
How many unused disks are in this provider |
unused_disks_total_size_bytes |
Total size of unused disks in this provider in bytes |
unused_disk_size_bytes |
Size of each disk in bytes |
unused_disks_last_used_timestamp_seconds |
Last timestamp (unix seconds) when this disk was used. GCP only! |
unused_provider_duration_seconds |
How long in seconds took to fetch this provider information |
unused_provider_info |
CSP information |
unused_provider_success |
Static metric indicating if collecting the metrics succeeded or not |
All metrics have the provider and provider_id labels to identify to which provider instance they belong.
The unused_disks_count, unused_disk_size_bytes, and unused_disks_total_size_bytes metrics have an additional k8s_namespace metric mapped to the kubernetes.io/created-for/pvc/namespace annotation assigned to persistent disks created by Kubernetes.
Information about each unused disk is currently logged to stdout given that it contains more changing information that could lead to cardinality explosion.
go install github.com/grafana/unused/cmd/unused-exporter@latest
In order to make E2E and UI testing easier, we implemented a fake provider that is only available when running go with the -tags=fake flag.
Usage of this flag should produce a deterministic output of fake unused disks for different providers.
For example, the following runs the unused binary using the fake provider:
➜ go run -tags=fake ./cmd/unused
time=2025-10-08T15:06:22.926-03:00 level=WARN msg="Using fake provider"
PROVIDER DISK AGE UNUSED TYPE SIZE_GB
Medium pvc-000-11371241257079532652-14470142590855381128 1y 210d ssd 79
Medium pvc-001-760102831717374652-9221744211007427193 128d 64d ssd 44
Medium pvc-002-3389241988064777392-12210202232702069999 72d 36d ssd 44
Medium pvc-003-2240328155279531677-7311121042813227358 1y 257d ssd 78
Medium pvc-004-9381769212557126946-1350674201389090105 1y 199d ssd 33
Medium pvc-005-11814882063598695543-3824056318896229933 26d 13d ssd 2
Medium pvc-006-13037077871211764336-13617661536098221011 182d 91d ssd 82
Medium pvc-007-8499734460532927466-10917103977888096435 1y 262d ssd 67
Medium pvc-008-137166597355566241-11349393104106307790 274d 137d ssd 94
Medium pvc-009-1873294050595685917-6999251097555031736 1y 191d hdd 90
Medium pvc-010-7129530298895469460-12892292024388140256 1y 230d hdd 5
Medium pvc-011-7832170631747924588-17404438405895559565 345d 172d hdd 55
Medium pvc-012-15199846812963376007-15282558343396517750 1y 243d ssd 89
Medium pvc-013-5425635553360250513-2558341386968588437 356d 178d ssd 58
This flag is also available for running tests: go test -tags=fake ./...