Cloud cost
Cluster idle
Node capacity that is schedulable but that no pod requested. It is a bin-packing and autoscaling problem, not a request-sizing one.
Last updated
Definition
Cluster idle is the gap between what a cluster's nodes can schedule, the allocatable capacity, and what pods requested. The capacity is provisioned and billed, and it sits empty. No workload asked for it. That makes it a different failure from workload idle, covered below. The fix is autoscaler tuning, tighter bin packing, and a node shape that matches the workload. Resizing one pod's request does not help.
Datadog's State of Cloud Costs research puts roughly 54 of the 83 idle percentage points it measures in container spend into this bucket. It is the larger of the two idle categories on the clusters it sampled.
Where it shows up
Cluster idle shows up as the gap between a node's allocatable capacity and the sum of the requests of every pod on it. You see it in a cluster resource-usage dashboard. You also see it in the Allocated resources section of kubectl describe node. Sum it across the whole node pool.
What makes it expensive
Do not treat cluster idle as a request-sizing problem and tune individual pods. The real lever is autoscaler configuration and bin packing across the whole node pool. One node with a wide allocated-versus-allocatable gap is noise. The same gap across most of the pool is a scheduling decision nobody made on purpose.
Related
More in Cloud cost
- Committed use discount (CUD)
- Reserved instance (RI)
- Savings plan (SP)
- On-demand vs. spot instances
- Data egress
- Amortized, blended, and unblended cost
- Billed vs. effective vs. list vs. contracted cost
- Rightsizing
- Idle and orphaned resources
- Storage class
- Lifecycle policy
- Incremental snapshot
- Allocatable capacity
- Requests vs. limits
- Workload idle
- Bin packing
- Cost allocation tags
- Tag coverage
- Unallocated spend
- Showback vs. chargeback
- Unit economics
- RI/SP coverage and utilization
- Observability tax
- Metric cardinality
- High-water-mark billing
- Log ingest vs. index
- Snowflake credit
- Auto-suspend / resume minimum
- BigQuery slot
- Bytes scanned & partition pruning
- Time Travel & Fail-safe
The meter you don't watch still grows.
Waste rarely hides in the service you check every week. CostMon watches every meter on your stack, including the ones you forgot were running.