DynamoDB Adaptive Capacity Split Threshold and Opt-Out Configuration Gap

Question

DynamoDB Adaptive Capacity automatically splits partitions when sustained traffic concentrates on a hot key, yet the service exposes no configuration parameter, API, or console setting to define the throughput threshold that triggers a split, manually initiate one, or opt out of the behavior entirely. For integration tests that must validate partition-boundary assumptions without production credentials, this creates a gap: DynamoDB Local and containerized mocks run with adaptive capacity disabled, so they reflect only the static partition model, while exercising the real service in a development account risks unexpected splits and billing.
The core uncertainty is whether AWS intends to surface split-threshold controls or an opt-out flag for workloads that require deterministic partition behavior during pre-production validation. Without such controls, teams cannot reliably reproduce or freeze the partition topology that their application logic depends on.
Is there a documented configuration boundary or service limit that indirectly constrains when Adaptive Capacity splits partitions? Can Contributor Insights or detailed CloudWatch monitoring provide enough visibility to infer split events without enabling costly high-resolution metrics? Does AWS maintain a public roadmap or feature request channel for adding explicit split-threshold management to DynamoDB tables?

Tasadduq Burney · Answer

Answer to the Core Questions
1. Is there a documented configuration boundary that limits when Adaptive Capacity splits?

No public setting exists. Internally, AWS triggers a split when a single key’s read or write traffic exceeds roughly 10,000 RCU/WRU per second, but the exact threshold is not published.
Adaptive Capacity is enabled by default and cannot be disabled through any console, API, or CLI flag.

2. Can Contributor Insights or CloudWatch metrics reveal split events without high‑resolution metrics?

Contributor Insights can identify hot keys but does not emit split events.
Standard CloudWatch metrics such as ThrottledRequests and a sudden drop in ConsumedReadCapacityUnits/ConsumedWriteCapacityUnits per item are the best indirect indicators.
Enabling high‑resolution metrics (1‑second granularity) improves detection but increases cost; consider using alarm thresholds on burst metrics instead.

3. Is there a public roadmap or feature‑request channel for split‑threshold controls?

AWS does not publish a roadmap for adding split‑threshold configuration. Customer feedback can only be submitted via AWS Support or the DynamoDB GitHub repository, with no guarantee of feature inclusion.

Practical Steps for Your Use Case

Confirm Adaptive Capacity is active. In the console, open the table’s Capacity tab; the Adaptive Capacity toggle should show Enabled. No opt‑out is available.
Set up CloudWatch alarms. Create alarms on ThrottledRequests and on a sudden drop in ConsumedReadCapacityUnits per item. A pattern of sustained throttling followed by a capacity‑unit dip often indicates a split.
Use DynamoDB Local for deterministic tests. Local and containerized mocks run with Adaptive Capacity disabled, so they preserve a static partition topology. Run your integration suite against Local to validate partition‑boundary assumptions.
For production‑like validation, provision a very low throughput table. Setting provisioned throughput to 1 RCU/WCU guarantees a key will never hit the split threshold, giving you a stable topology while still exercising the live service.
Monitor hot keys with Contributor Insights. Even though it doesn’t show splits, it helps you identify which keys to watch for potential throttling.
Keep costs in mind. If you enable high‑resolution metrics, set a retention period that balances visibility with expense.

Missing Diagnostic Detail
To refine the recommendation for your specific scenario, could you confirm whether your table uses Provisioned or On‑Demand capacity? This affects how CloudWatch metrics behave and what thresholds are meaningful.

DynamoDB Adaptive Capacity Split Threshold and Opt-Out Configuration Gap

1 answer

Answer to the Core Questions

Practical Steps for Your Use Case

Missing Diagnostic Detail

Your perspective belongs here.

0 question comments