Best practices to avoid throttling when Azure Functions integrate with Dataverse:

Best practices to avoid throttling when Azure Functions integrate with Dataverse:

August 20, 2025

🔹 1. Use Custom API / Plugin (preferred approach)

Encapsulate multiple operations inside one Dataverse Custom API.
Azure Function only makes one call per message.
Benefits: fewer API calls, transactional consistency, central business logic.

🔹 2. Use Batch (ExecuteMultiple / $batch)

If you must call Dataverse directly, use $batch requests.
Example: instead of 50 separate POST /contacts, send them in 1 batch request.
This reduces entitlement usage and network overhead.

🔹 3. Distribute load across Application Users

Create multiple App registrations in Azure AD (with Dataverse application users).
Spread API calls across them to avoid hitting the per-user limits.
Example: Function A uses AppUser1, Function B uses AppUser2.

🔹 4. Implement Retry with Exponential Backoff

Always catch 429 (Too Many Requests) and 5xx errors.
Retry after the Retry-After header or use exponential backoff (2s → 4s → 8s…).
Don’t retry immediately — that makes throttling worse.

🔹 5. Control Concurrency in Azure Function

By default, Functions can scale out aggressively, hammering Dataverse.

Use host.json settings to limit parallelism:

json

{
  "extensions": {
    "serviceBus": {
      "maxConcurrentCalls": 10
    }
  }
}

This ensures only 10 messages are processed at a time.

🔹 6. Prefer Service Bus Sessions / Queues for pacing

Queue messages instead of firing all at once.
This lets you control throughput and align with Dataverse API limits.

🔹 7. Use Change Tracking / Delta Loads

For data sync, only move delta records (changed rows).
Reduces unnecessary Dataverse API calls.

🔹 8. Monitor & Tune

Enable Dataverse API metrics in Power Platform Admin Center.
Track: API calls/minute, throttled calls, failed requests.
Tune Function concurrency based on observed limits.

✅ Summary:

Best option → Use Custom API + Plugin to minimize calls.
If direct calls → use batching, retries with backoff, concurrency limits, and multiple app users.
Always monitor entitlements and adjust load.

Comments