All cheat sheets
Databasescheat sheet

Redshift Serverless

Redshift Serverless is the automatic transmission to Pause/Resume's manual lever: same query engine, no cluster sizing, per-second billing, and compute charges that drop to zero when idle. Intermittent and bursty warehouses win big; pegged-24/7 ones may still favor provisioned + RIs.

Last reviewed: July 14, 2026

Prefer step by step? Take the guided Redshift Serverless walkthrough — free, anonymous, interactive.Have an AI assistant? Jump to the copy-paste audit prompt at the bottom of this sheet.

TL;DR: If Pause/Resume is the manual lever, Redshift Serverless (2022) is the automatic transmission — same query engine, no node sizing, no EventBridge scheduling. You set a base capacity in RPUs (a ceiling for the autoscaler, not a floor you pay for), capacity scales in seconds and drops to zero when idle, and you're billed per second for actual RPU-hours. For any workload that isn't pegged at high load 24/7, it's now the simpler default.

The numbers

  • RPU (Redshift Processing Unit) bundles CPU + memory + I/O; base capacity 8–512 RPUs, ~$0.375/RPU-hour (us-east-1, verify current).
  • Idle = $0 compute — the base capacity caps the autoscaler; no queries means no RPU-hours billed. Storage bills separately at ~$0.024/GB-mo (same as provisioned).
  • Cross-over: ~10 hrs/wk at 32 RPUs ≈ 1,385 RPU-hours/mo ≈ ~$520/mo vs ~$360/mo for a dc2.large × 2 running 24/7 — slightly higher absolute, but with auto-scale headroom, zero ops, and no upsize risk. The more intermittent the workload, the more Serverless dominates.
  • Reserved Capacity: commit a predictable RPU-hour floor for 1 or 3 years for up to ~65% off the committed portion; everything above bills on-demand.
  • Field examples: a finance team with month-end bursts went $1,200 → ~$350/mo (~$10,200/yr); a SaaS analytics startup landed at ~$200–300/mo vs $700+ provisioned.

Do this

  1. Start conservative at a 32 RPU base — it handles a lot; scale up only when query queueing actually shows in the Query Monitoring view.
  2. Watch real RPU consumption in CloudWatch for a week before tuning — actual patterns rarely match the week-one gut estimate; most teams over-set and trim back.
  3. Migrate from provisioned via snapshot → restore into a Serverless namespace + workgroup → repoint BI/ETL/dbt connection strings → validate a week → decommission.
  4. Buy Reserved Capacity only after usage stabilizes — don't commit on day one.
  5. Validate query performance after cutover — usually comparable or better (the autoscaler often gives big queries more compute than a fixed cluster).

Gotchas

  • Base capacity is a ceiling, not a paid baseline — the single most common misconception; idle truly costs zero compute.
  • Cold start from idle is usually under 10s — negligible for analytics, relevant only if you've promised a 50ms p99 to a customer-facing dashboard.
  • Soft concurrency limits exist — hit them and AWS scales up, or you raise base capacity.
  • Verify region availability before committing.

Skip this if

  • The warehouse runs pegged at high load most hours of the day and you'll commit to multi-year terms — a provisioned ra3 fleet with Redshift Reserved Nodes can edge out Serverless; price both.
  • Your usage is predictable business-hours-only and you'd rather keep an existing provisioned cluster — Pause/Resume captures most of the savings without a migration. For the OLTP side of the house, see Aurora I/O-Optimized.

Run this audit with your AI assistant

Paste this into Claude, ChatGPT, or any agent that can run the AWS CLI with read-only credentials. It audits your account for exactly the waste this sheet describes — and changes nothing.

You are auditing an AWS account's Redshift usage to decide between
provisioned clusters and Redshift Serverless. Use the AWS CLI with
READ-ONLY credentials. Do not create, modify, or delete anything —
report findings and recommended (unapplied) fixes only.

1. Inventory: aws redshift describe-clusters (provisioned: node type ×
   count, $/hr, RI coverage) and aws redshift-serverless
   list-workgroups / list-namespaces (base RPU capacity).
2. Utilization shape: pull CloudWatch AWS/Redshift metrics hourly over
   30 days — CPUUtilization, query counts, connection counts. Classify:
   intermittent (reports/dashboards/ad-hoc), steady-24/7-heavy, or
   dev/test.
3. Cost compare: provisioned = $/hr × 730 (minus RI). Serverless =
   est. active RPU-hours × ~$0.375. Flag intermittent + dev clusters as
   serverless wins (scale-to-zero); flag pegged-24/7-heavy as
   provisioned + RI likely cheaper.
4. Reserved Capacity note: for predictable serverless floors, 1/3-yr
   commits cut up to ~65% on the committed portion.

Report a table: cluster/workgroup | workload shape | provisioned $/mo |
est. serverless $/mo | recommended base RPU | verdict | notes. Change
nothing.
Works with any assistant that can run shell commands.

Want the guided version?

The Redshift Serverless walkthrough covers this topic interactively — it asks about your setup, branches to what’s relevant, and quizzes you on the tricky parts. Free and anonymous.

Start the walkthrough

More Databases cheat sheets

Amazon RDS Proxy

Connection count, not CPU, is often what forces an RDS upsize. RDS Proxy pools thousands of client connections (especially from Lambda) onto ~10–30 real ones, so a small instance stays small — the proxy fee typically covers 10–25% of the avoided upsize.

Athena Partition Projection

Instead of asking S3 which partitions exist before every query, Partition Projection lets Athena calculate the paths from a pattern you declare — near-instant query startup on tables with thousands of partitions, zero LIST metadata calls.

Athena Query Result Reuse

Athena caches identical query results in S3, so a repeated dashboard refresh or scheduled report returns instantly with zero bytes scanned. A workgroup checkbox that routinely cuts thousands off a scan-heavy bill.

Aurora I/O-Optimized

Aurora bills a sneaky third bucket — I/O operations at ~$0.20/million — that can dwarf compute and storage on I/O-heavy workloads. I/O-Optimized zeroes it out for a ~30–40% higher instance rate. A break-even calc, not a default.