DynamoDB Warm Throughput: Pre-Warming Tables for Traffic Spikes 🔥
You've planned for the big event—a flash sale, a game launch, a marketing campaign that's about to drive a surge of traffic. Your DynamoDB table is on on-demand capacity mode, so it should scale automatically, right?
You've planned for the big event—a flash sale, a game launch, a marketing campaign that's about to drive a surge of traffic. Your DynamoDB table is on on-demand capacity mode, so it should scale automatically, right? Mostly yes. But there's a catch: DynamoDB tables have a "warm" throughput level based on recent traffic patterns. If your table has been handling 100 requests per second and suddenly needs to handle 50,000, there's a ramp-up period that can cause throttling during those critical first minutes.
To put this in perspective: during Prime Day 2025 , DynamoDB handled tens of trillions of API calls over 4 days, peaking at 151 million requests per second—all while maintaining single-digit millisecond responses. That kind of performance at that scale doesn't happen by accident. Amazon prepares its infrastructure well in advance, and the same tooling they rely on is available to you.
DynamoDB's warm throughput feature addresses this head-on. It gives you visibility into your table's current warm capacity and lets you proactively pre-warm it before a traffic spike hits. The same "prepare, don't pray" philosophy that powers Prime Day—now in your hands. ✅
🤔 What Is Warm Throughput?
Every DynamoDB table and GSI maintains a "warm" throughput level—the amount of read and write traffic it can handle immediately without throttling. This level grows automatically as your traffic increases, but it takes time to scale up if additional partitions are needed to handle the increased capacity.
Warm throughput provides two capabilities:
- Visibility — See the current warm read and write capacity of your table and indexes in the DynamoDB console or via API
- Pre-warming — Proactively set higher warm throughput values before an anticipated traffic spike
The pre-warming blog post from the AWS Database Blog walks through the mechanics in detail.
📊 How It Differs from Capacity Modes
Warm throughput isn't a replacement for on-demand or provisioned capacity—it's a complement:
| Feature | On-Demand | Provisioned | Warm Throughput |
|---|---|---|---|
| 🎯 Purpose | Handle any traffic volume, up to the tables warm capacity | Fixed capacity with auto-scaling | Pre-warm for anticipated spikes |
| 📈 Scaling | Automatic increases (managed via partition splits) | Manual or auto-scaling policies | Proactive, instant readiness |
| 💰 Cost | Pay per request | Pay per provisioned unit | No additional charge for visibility; pre-warming may incur costs |
| ⏱️ Best for | Unpredictable traffic | Steady, predictable traffic | Known upcoming traffic events |
Think of it this way: on-demand mode lets you ignore day to day capacity management. Warm throughput is your "we know Black Friday is coming" preparation tool.
🎯 When to Use Pre-Warming
Pre-warming is valuable when you can predict a traffic spike. Amazon uses this exact strategy to prepare DynamoDB for Prime Day—they know the start time, they know the scale, and they pre-warm accordingly. Your events may be bigger or smaller, but the principle is identical:
- 🛒 Flash sales and promotions — E-commerce events with a known start time
- 🎮 Game launches — Thousands of players hitting your backend simultaneously
- 📺 Live events — Streaming events that drive concurrent API calls
- 🏟️ Prime Day-scale events — Any scenario where you know traffic will 10x–1000x within minutes
- 🔄 Migration cutover — Switching traffic from an old system to a new DynamoDB-backed one
🛠️ How to Pre-Warm
- Check current warm throughput — In the DynamoDB console, navigate to your table's "Additional settings" to see current warm read/write values
- Set target values — Specify the warm throughput you expect to need. DynamoDB will prepare the underlying infrastructure
- Time it right — Pre-warm at least 30 minutes before your expected traffic spike to give the system time to prepare (For every doubling of warm throughput, DynamoDB needs to split each partition in your table, i.e 1000 WCU to 8000 WCU requires 3 split generations)
- Monitor with CloudWatch — Use the enhanced throttling observability metrics to verify your table is ready
For a detailed guide on choosing the right throughput strategy overall, see the throughput strategy blog post .
⚠️ Things to Keep in Mind
- Pre-warming doesn't replace good data modeling. If you have hot partitions, pre-warming the table won't fix them. Distribute your writes evenly across partition keys.
- GSIs need pre-warming too. Each GSI has its own warm throughput. If your spike involves GSI queries, pre-warm those indexes as well.
🏆 How Amazon Prepares DynamoDB for Prime Day
Here's what Prime Day 2025 looked like for DynamoDB at Amazon's scale:
| Metric | Prime Day 2025 |
|---|---|
| 📊 Total API calls | Tens of trillions over 4 days |
| ⚡ Peak throughput | 151 million requests per second |
| ⏱️ Response time | Single-digit milliseconds maintained throughout |
| 🏗️ Services powered | Alexa, Amazon.com, all fulfillment centers |
Amazon doesn't flip a switch on Prime Day morning and hope DynamoDB scales fast enough. They use the same warm throughput tooling to pre-provision capacity across tables and GSIs weeks in advance. Load testing, capacity reviews, and pre-warming are all part of the AWS Countdown playbook—and that same preparation strategy scales down to your workload.
💡 Quick Checklist for Traffic Events
- Identify the expected peak read/write throughput
- Check current warm throughput values for your table and GSIs
- Pre-warm at least 30 minutes before the event
- Verify with CloudWatch metrics that warm capacity meets your target
- Monitor throttling metrics during the event
- Review post-event to calibrate for next time
👉 Preparing for a big launch? Join us in the AWS NoSQL Launchpad Space on Builder Center where we share DynamoDB performance strategies, production war stories, and learn the NoSQL mindset together—one use case at a time.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article