CloudWise — AWS Waste Teardowns

CloudWise
  • 0.0 (0)
  • Technology
  • Updated Weekly

Short, technical teardowns of where AWS bills quietly leak. One waste pattern per episode: the billing mechanics behind it, how to spot it yourself, and the safe fix. The two hosts are AI-generated voices, and every episode is reviewed for accuracy before it is published. Find these patterns in your own account, read-only, on your own machine: pip install cloudcostwise

Episodes

  1. 2h ago

    Unattached Elastic IPs: Why AWS Bills Them and How to Release Them Safely

    Since February 2024, AWS charges $0.005 per hour for every public IPv4 address, including Elastic IPs left behind after an instance is terminated, which is about $3.65 a month per address. This episode explains why orphaned Elastic IPs pile up unnoticed, how to find them with the DescribeAddresses call by looking for addresses with no association ID, and why not to mass release them: releasing is usually permanent, and a DNS record, firewall allow list or partner configuration may still point at that exact IP. Find this waste on your own machine, read-only: pip install cloudcostwise

  2. 2h ago

    Idle SageMaker Real-Time Endpoints Bill Every Hour, Even With Zero Traffic

    SageMaker real-time endpoints run on dedicated instances that bill every hour they are in service, whether or not a single request arrives. This episode covers why the billing works that way, the math for an ml.t2.medium endpoint at $0.056 per hour in us-east-1 (about $40 a month while idle) and how larger GPU instances multiply it, how to spot an idle endpoint with the CloudWatch Invocations metric at zero over at least seven days, what deleting the endpoint does (it stops the instance charge and keeps the model and endpoint configuration so it can be recreated), what to check first such as scheduled jobs and hard-coded references, and when serverless or asynchronous inference fits spiky traffic better. Find this waste on your own machine, read-only: pip install cloudcostwise

  3. 2h ago

    Unused VPC Interface Endpoints: About $22 a Month Each, and Why Deleting One Is Risky

    Gateway endpoints for S3 and DynamoDB are free, but interface endpoints (AWS PrivateLink) are not: each one costs about $0.01 per hour in every Availability Zone it is deployed in, plus about $0.01 per GB processed in us-east-1, so one idle endpoint across three zones costs about $22 a month. Endpoints created for a proof of concept are often forgotten, and no alarm flags them. This episode shows how to spot them with the BytesProcessed metric in the AWS/PrivateLinkEndpoints CloudWatch namespace over about 14 days, and why not to delete one on that signal alone: with private DNS enabled, removing the endpoint changes how the service name resolves, and workloads in private subnets, including rare disaster recovery paths, can lose access to that service. Find this waste on your own machine, read-only: pip install cloudcostwise

  4. 2h ago

    Idle RDS Instances: Why Stopping Isn't Enough and How to Delete Safely

    An RDS instance bills by instance class for every hour it runs, whether or not anything connects, and Multi-AZ roughly doubles that instance charge. This episode covers why idle databases slip past monitoring that only alarms on spikes, how to detect one with the CloudWatch DatabaseConnections metric at zero for at least seven days, why stopping is only temporary (storage and backups keep billing, and AWS starts the instance again after seven days), why deletion with a final snapshot is the real fix, and why to confirm with the application owners first, since hard-coded references or recovery scripts may still depend on the database. Find this waste on your own machine, read-only: pip install cloudcostwise

  5. 3h ago

    Idle NAT Gateways: $32 a Month for Nothing, and How to Remove One Safely

    An idle NAT gateway bills about $32 a month in us-east-1 at four and a half cents per hour even if no traffic flows through it, plus four and a half cents per gigabyte processed and half a cent per hour for its public IPv4 address. This episode covers why these gateways outlive the projects that created them, how to spot one with read-only CloudWatch signals (BytesOutToDestination and ActiveConnectionCount at zero over at least seven days), why you must check which route tables still point at it before deleting, why the Elastic IP keeps billing until you release it, and how free gateway endpoints for S3 and DynamoDB avoid NAT data processing charges. Find this waste on your own machine, read-only: pip install cloudcostwise

  6. 3h ago

    Converting EBS gp2 to gp3: Why Large Volumes Save Less Than 20 Percent

    AWS prices gp3 EBS storage about 20 percent below gp2, and the change runs in place on an attached volume. If you convert without setting IOPS or throughput, EBS keeps whichever is higher, the gp2 volume's performance or the gp3 baseline of 3,000 IOPS and 125 MB/s, and bills anything above that baseline separately, so large gp2 volumes often save much less than the headline 20 percent. This episode covers sizing gp3 performance from each volume's CloudWatch read and write metrics instead of its old gp2 size, and getting the settings right before you start: a modification cannot be cancelled, can take hours on large volumes, and limits how soon you can modify the volume again. Find this waste on your own machine, read-only: pip install cloudcostwise

  7. 3h ago

    Idle Application Load Balancers: The Hourly Charge for Doing Nothing

    An Application Load Balancer bills by the hour even when it serves no requests and has no healthy targets: $0.0225 per hour in us-east-1 before capacity units, and an internet-facing ALB also holds public IPv4 addresses at $0.005 per hour each. This episode covers why idle load balancers survive (a test environment gets torn down, nobody owns what is left, and nothing crosses an alarm threshold), how to spot one with a CloudWatch request count of zero and DescribeTargetHealth showing no healthy targets over 14 to 30 days, and why deleting one cannot be undone and gives any replacement a new DNS name, so every Route 53 record and external CNAME pointing at it must be found first. Find this waste on your own machine, read-only: pip install cloudcostwise

  8. 3h ago

    EBS Snapshot Sprawl: Why Deleting or Archiving Old Snapshots Often Saves Less

    Deleting an old EBS snapshot often frees far less storage than its size suggests, because standard snapshots are incremental and only blocks no other snapshot needs are released. Moving snapshots to the archive tier at $0.0125 per GB-month is not a cleanup either: archiving converts each snapshot to a full copy, carries a 90-day minimum and charges a fee to restore. This episode looks at where orphaned snapshots usually come from, CI pipelines that create AMIs and never deregister them, the signal to look for (a snapshot whose source volume is gone and that no registered AMI uses), why that still doesn't prove it is unused, and why to set up an AWS Recycle Bin retention rule before deleting anything. Find this waste on your own machine, read-only: pip install cloudcostwise

  9. 3h ago

    CloudWatch Logs Never Expire by Default: The Storage Bill That Keeps Growing

    CloudWatch log groups keep their data forever unless you set a retention policy, so the storage charge grows every month on a pile of logs nobody reads. This episode works through the unit math in us-east-1, covering ingestion per gigabyte and storage per gigabyte each month. It explains why the cost goes unnoticed and how to find affected log groups with describe-log-groups by looking for a missing retentionInDays attribute. It also explains why put-retention-policy is a one-way door: data older than the new period is deleted permanently, so export anything you must keep to S3 before you shorten retention. Find this waste on your own machine, read-only: pip install cloudcostwise

  10. 5h ago

    ECR Images With No Lifecycle Policy: Storage Costs and Safe Expiry Rules

    When a new container image is pushed to Amazon ECR, the image it replaces loses its tag but stays in the repository, and ECR storage is billed at $0.10 per GB-month. This short episode explains why untagged does not mean unused: orchestrators often run images by digest, and the untagged image may be what is running in production or what a rollback needs. It covers checking for a missing policy with GetLifecyclePolicy, then confirming real image counts and push dates with describe-images. It also explains why a sinceImagePushed window counts from the push date and does not protect rollbacks, and how to keep the last N releases with a tagged rule and a tagPatternList, with any catch-all rule at the highest priority number. It notes that the lifecycle preview shows what would expire, not whether expiring it is safe, and recommends checking the digests of running tasks before adding an untagged expiry rule. To find old and untagged ECR images in your own AWS account, the free, source-available CloudWise scanner (pip install cloudcostwise) runs read-only on your machine and sends nothing to CloudWise.

  • 10 Episodes

About

Short, technical teardowns of where AWS bills quietly leak. One waste pattern per episode: the billing mechanics behind it, how to spot it yourself, and the safe fix. The two hosts are AI-generated voices, and every episode is reviewed for accuracy before it is published. Find these patterns in your own account, read-only, on your own machine: pip install cloudcostwise

Information