Explanation
Why the waste happens and who it affects.
When those nodes are billed at on-demand rates, the organization pays the most expensive rate for its most predictable data warehouse capacity. The Redshift documentation advises to use on-demand pricing during evaluation and proofs of concept, then reserve nodes once production needs are established.
Coverage gaps appear because Redshift commitments are separate from every other AWS commitment product: Compute Savings Plans do not apply to Redshift, and Redshift is not among the services covered by Database Savings Plans. Reservations are also tied to a Region, node type and node count, so teams delay purchases while a resize or node-type migration is being considered, and reservations that retire at the end of their term silently return nodes to on-demand billing.
Billing model
The pricing dimensions that drive this cost.
Provisioned cluster nodes are billed at on-demand rates unless a matching active reservation exists.
- On-demand node-hours
- Each compute node is billed hourly at the on-demand rate for its node type and Region, and billing stops only when the cluster is paused or deleted
- Reserved nodes
- 1- or 3-year reservations for a specific Region, node type and number of nodes, with No Upfront, Partial Upfront or All Upfront payment
- Charged regardless of use
- The reservation's hourly rate applies for every hour of the term whether or not a matching node is running
- Unmatched nodes
- Nodes beyond the reserved count, or of a different node type after a resize, are billed at the on-demand rate
How to detect
5 checks to find it in your estate.
- Review the Trusted Advisor check Amazon Redshift reserved node purchase recommendations, sourced from Cost Optimization Hub, and Cost Explorer reservation recommendations for Amazon Redshift with a 30- or 60-day lookback
- Compare the node type and node count of each long-running cluster with active reservations in the same Region (describe-reserved-nodes, or the Reserved nodes page in the Redshift console) to find nodes billed on-demand
- Group Redshift spend by Purchase Option in Cost Explorer to see how much node usage is on-demand
- List reservations with their end dates and flag those retiring soon, since retired reservations stop discounting nodes that keep running
- Exclude clusters that are paused on a schedule, are being migrated to Serverless, or are planned for a node-type change from the baseline you intend to reserve
How to fix
5 ways to remove the waste.
- Purchase reserved nodes for the node type, count and Region of clusters that will keep running for the term; AWS documents No Upfront at about a 20 percent discount and Partial or All Upfront at larger discounts depending on duration
- Complete planned resizes or node-type migrations before buying, because a resize to a different node type moves the nodes to on-demand rates while the original reservation keeps billing until its term ends
- Reserve only the steady node count, leaving temporary resizes and concurrency scaling at on-demand rates
- Remember that reserved node benefits are shared across accounts under consolidated billing, and account for linked accounts when deciding how many nodes to reserve
- For steady Redshift Serverless usage, consider serverless reservations, which commit to an RPU level billed hourly with usage above it charged at on-demand rates
Documentation
Vendor references for pricing and configuration.