Edit

Indexer execution on Serverless and Standard 3 High Density (S3 HD)

This article describes the indexer execution model that Azure AI Search uses for Serverless and Standard 3 High Density (S3 HD) search services. Both options have a service-level daily runtime quota that governs how much total indexer time you can use per 24-hour UTC window.

Important

The capabilities described in this article are in preview under supplemental terms of use.

Where it applies

The execution model in this article applies to:

  • Serverless search services that run indexers using the 2026-05-01-preview REST API or later.
  • S3 HD search services that run indexers using the 2026-05-01-preview REST API or later.

Existing indexer definitions, data sources, skillsets, and knowledge sources work without modification on both options.

Execution model

Indexers on Serverless and S3 HD have the following execution characteristics:

  • You don't provision or manage indexer infrastructure. The service handles capacity for you.

  • Indexers run only in the public execution environment. The private execution environment, which is provided through shared private link resources, isn't available for indexers on these SKUs.

  • For S3 HD, if you need indexer connections to stay off the public internet, configure a network security perimeter (NSP) on your search service to control inbound and outbound traffic through explicit access rules.

  • For Serverless, there's no support for private connections.

Daily cumulative runtime quota

Indexer execution is governed by a daily runtime quota that resets at 00:00 UTC. The quota is:

  • Service level: It applies to the search service as a whole.
  • Cumulative: Runtime from every indexer in the service counts toward the same budget. The quota isn't applied per indexer.

The following table lists the daily quota by SKU and the minimum API version that supports it:

SKU Daily quota per 24-hour UTC window Minimum API version
S3 HD 6 hours 2025-11-01-preview
Serverless 6 hours 2026-05-01-preview

When the daily quota is exhausted:

  • Indexers that are currently running stop within about five minutes.

  • New indexer runs don't process documents and immediately return a transient failure indicating that the daily quota has been exceeded.

  • Normal indexer execution resumes after the counter resets at 00:00 UTC.

Recover from quota exhaustion

To recover from quota exhaustion and reduce the likelihood of hitting it again:

Monitor cumulative runtime

This section explains how to track runtime usage and remaining budget using the Search Service REST APIs. There's no portal experience for cumulative runtime during the preview.

Service-level runtime

Use Get Service Statistics (REST API) to retrieve cumulative indexer runtime across all indexers in the service for the current 24-hour window:

GET {endpoint}/servicestats?api-version=2026-05-01-preview

The response includes an indexersRuntime section. The following JSON shows a service whose six-hour daily quota hasn't been used:

"indexersRuntime": {
    "usedSeconds": 0,
    "remainingSeconds": 21600,
    "beginningTime": "2026-05-16T00:00:00.000Z",
    "endingTime": "2026-05-17T00:00:00.000Z"
}

Key points:

  • usedSeconds: Total seconds that all indexers in the service have run during the current window.
  • remainingSeconds: Seconds still available before the daily quota is reached. Present when a tier-specific limit applies.
  • beginningTime and endingTime: Start and end of the current 24-hour UTC counting window.

Indexer-level runtime

Use Get Indexer Status (REST API) to retrieve cumulative runtime for an individual indexer:

GET {endpoint}/indexers('{indexerName}')/search.status?api-version=2026-05-01-preview

The response includes a runtime section. The following JSON shows an indexer on a service whose six-hour daily quota hasn't been used:

"runtime": {
    "usedSeconds": 0,
    "remainingSeconds": 21600,
    "beginningTime": "2026-05-16T00:00:00.000Z",
    "endingTime": "2026-05-17T00:00:00.000Z"
}

Key points:

  • usedSeconds: Total seconds the indexer has run during the current window.
  • remainingSeconds: Seconds still available to all indexers in the service, not just this indexer. Present when a tier-specific limit applies.
  • beginningTime and endingTime: Start and end of the current 24-hour UTC counting window.

Best practices

Indexer support on S3 HD and Serverless is in preview. Follow this guidance to size workloads appropriately and to plan for billing for Serverless to be introduced at a later date.

S3 HD

During the preview, S3 HD indexer support is designed for workloads with no skillsets or small skillsets. To stay within the daily quota:

  • Plan for small indexes of about 1 GB in size.

  • Size skillset usage carefully. Skills that call external services, such as the Azure OpenAI Embedding skill, GenAI Prompt skill, and Azure Content Understanding skill, significantly increase runtime and can consume the daily quota quickly, especially in multitenant scenarios.

  • Expect limited parallelism during the preview. Use scheduled, staggered runs for large indexer fleets so that work spreads across the 24-hour window rather than competing for the same budget.

Serverless

During the preview, Serverless indexers are designed to simplify ingestion for retrieval-augmented generation (RAG) and knowledge base scenarios:

  • Indexer execution (excluding skills) is currently free. Writing documents to an index incurs a cost.

  • Skillset execution is billed the same way as on dedicated indexers. Calls to external services, such as the Azure OpenAI Embedding skill, GenAI Prompt skill, and Azure Content Understanding skill, are billed through the attached Foundry or Azure AI services resource.

Limits and quotas

For indexer limits on Serverless and S3 HD, see Indexer limits.

Related content

  • Last updated on