# Idle Azure VM

> Running Azure VMs averaging under 5% CPU over 30 days, with memory and network guards, get a recurring off-hours schedule sized by the measured idle fraction, never a one-shot deallocation. Reservation-covered VMs are instead right-sized via catalog rate deltas, and both paths abstain when pricing or schedule data is absent.

Source: https://zop.dev/integrations/azure/recommendations/idle-azure-vm
Updated: 2026-08-19

---

## A running VM with flat CPU over the lookback

- resource.Status: statusSucceeded
- MetricsAware: azureCPUMetricName over azureCPULookbackDays days

## CPU floor, memory and network guards

```text
azureCPUThreshold=5.0, azureCPULookbackDays=30 (memory-bound guard at 60%, network idle guard at 1 MB bytes/bucket)
```

## Schedule on-demand, rightsize when committed

two branches, no fabricated fraction.
- OnDemand path → schedule (not a one-shot deallocate). A one-shot stop on a running VM is an indefinite outage, not a cost lever; idle or non-prod compute gets a recurring off-hours schedule. The dollar figure is the measured idle fraction from the aggregator's heatmap (SetScheduleData, sd.IdlePercentage): savings = cost × sd.IdlePercentage, category schedule. The previous azureComputeOnlySavings × schedulingFallbackSavings (0.67) static fraction is removed; absent or zero schedule data => abstain (returns nil), never a flat fraction, never $0.
- Committed (RI/SP) path → rightsize. Deallocation saves nothing (commitment bills regardless), so the rule rightsizes via catalog-diff (srcRate − dstRate) × 730 (idleVMRightsizeSavings → DeriveCatalogDiffSavings(..., 0)) with no flat-fraction fallback; it abstains when either rate is absent/≤0 (azureIdleRightsizingSavingsFraction 0.75 lingers in constants.go but is not consumed on this path), category rightsizing. Description notes "managed disk storage continues to bill". Both paths abstain when pricing is absent (cost ≤ 0). EvalContextRule: a VM with a known active schedule or measured-and-near-off uptime (long-off hysteresis) is suppressed.

## Scheduling off hours instead of leaving it stopped

1. Review Azure Monitor CPU, memory, and network metrics for this VM
2. Verify no workloads depend on this VM
3. Apply the recommended recurring off-hours schedule (auto start/stop); do not leave it deallocated indefinitely
4. Note that managed disk storage keeps billing while the VM is stopped

## The single CPU series this rule reads

azureCPUMetricName (azureCPULookbackDays day lookback)