Kill the Zombie Cloud Bills: Finding and Removing Orphaned Cloud Resources Before Your Next Invoice

Kill the Zombie Cloud Bills: Finding and Removing Orphaned Cloud Resources Before Your Next Invoice

Your cloud bill arrives, and it is higher than last month. Nobody launched anything new. Nobody ran a big batch job. Yet the number keeps climbing. The usual culprit is not active workloads — it is zombie cloud resources: orphaned load balancers, unattached disks, idle clusters, and snapshots nobody remembers creating.

For small and midsize businesses, these ghosts are expensive. You pay for them every hour, every month, with zero return. The longer it sits, the harder it is to remember why it existed. The worse part is that they accumulate silently, one forgotten experiment or one failed deployment at a time.

How Zombie Resources Pile Up

Zombies are rarely created on purpose. A developer spins up a database for a feature branch and deletes the branch but not the database. A load balancer is created for a migration and never removed. An EKS or GKE cluster is built for a proof of concept and left running. Snapshots are taken before a change and never deleted.

Finance notices first, but by then the money is already gone. The fix is not a one-time cleanup. It is an audit, a triage process, and automation that prevents zombies from returning.

The One-Hour Audit: Find Orphans in AWS, Azure, and GCP

Start with the resources that cost the most while doing nothing. Unattached EBS volumes in AWS:

# Zombie disks: EBS volumes with no attachment
aws ec2 describe-volumes \
  --filters Name=status,Values=available \
  --query 'Volumes[*].{ID:VolumeId,SizeGiB:Size,Age:CreateTime,Type:VolumeType}' \
  --output table

Load balancers with no backends:

aws elbv2 describe-load-balancers \
  --query 'LoadBalancers[*].{Name:LoadBalancerName,DNS:DNSName,Scheme:Scheme,Type:Type}' \
  --output table

Cross-check each load balancer against its target groups. A healthy load balancer with zero healthy targets is a zombie.

Idle Kubernetes clusters are another common zombie. A cluster that is running but has no active workloads still charges for the control plane and every node. In AWS, check for EKS clusters with no running pods:

aws eks list-clusters --query 'clusters' --output table
# For each cluster, list nodes:
aws eks list-nodegroups --cluster-name <cluster-name> --output table

Snapshots are quieter zombies. They cost less per gigabyte, but they accumulate fast. An automated snapshot policy with a 30-day retention limit stops the creep:

aws ec2 describe-snapshots --owner-ids self \
  --query 'Snapshots[*].{ID:SnapshotId,Age:StartTime,Size:VolumeSize}' \
  --output table | sort -k2

Azure orphaned network interfaces:

az network nic list \
  --query "[?virtualMachine==null].{Name:name,PrivateIP:ipConfigurations[0].privateIPAddress,Subnet:ipConfigurations[0].subnet.id}" \
  -o table

GCP persistent disks with no attached instance:

gcloud compute disks list --filter="users:[]" --format="table(name,sizeGb,zone,status)"

Triage: Safe Delete, Archive, or Keep

Not every idle resource is a zombie. A volume attached to a stopped instance may hold critical data. Before deleting anything, apply a triage rule:

  • Safe delete: unattached volumes and idle load balancers older than 30 days with no Owner tag.
  • Archive: snapshots older than 90 days — move to cold storage or delete after team confirmation.
  • Keep: resources with a valid Owner tag, a RetainUntil date in the future, or an active incident ticket.

Tagging is what makes triage possible. Every cloud resource should carry at least Owner, Environment, CostCenter, and RetainUntil:

resource_tags:
  required:
    - Owner
    - Environment
    - CostCenter
    - RetainUntil
  optional:
    - Project
    - DataClassification

Resources without these tags are guilty until proven innocent.

Tagging alone does not enforce anything. Use a policy engine such as AWS Config, Azure Policy, or Open Policy Agent to reject resources that arrive without the required tags. The goal is to make the untagged resource impossible to create, not just discouraged.

Automation: Scheduled Cleanup Without Breaking Production

Manual audits do not scale. Schedule a read-only scan every Monday that reports candidates, and a destroy job that only removes resources older than your grace period. Never automate deletion of resources with a RetainUntil date in the future or an Owner tag pointing to an active employee.

Report the results to a shared channel where the team can see the numbers. A monthly “zombie report” with the total wasted spend creates accountability without blame. When engineers see the cost of their forgotten experiments, the behavior changes faster than any policy document.

# Example: tag-based guardrail check (read-only report)
aws resourcegroupstaggingapi get-resources \
  --tag-filters Key=Owner,Values="" \
  --resource-type-filters ec2:volume elasticloadbalancing:loadbalancer \
  --query 'ResourceTagMappingList[*].ResourceARN' \
  --output table

Guardrails: Stop Zombies Before They Are Born

The cheapest zombie is the one never created. Enforce defaults in infrastructure as code so every resource is born tagged and time-boxed:

provider "aws" {
  default_tags {
    tags = {
      Owner        = "platform-team"
      Environment  = "production"
      CostCenter   = "engineering"
      RetainUntil  = "2027-06-01"
    }
  }
}

resource "aws_db_instance" "analytics" {
  ...
  lifecycle {
    prevent_destroy = true
  }
}

Combine that with cloud budgets and anomaly alerts so finance hears about spikes before the invoice arrives.

For more on keeping cloud costs under control:

Book your free 30-minute DevOps consultation →

Scroll to Top