---
title: "Get SPA Recommendations with a shard parameter"
method: GET
path: "/api/v2/spa/recommendations/{service}/{shard}"
tags: ["Spa"]
---

# Get SPA Recommendations with a shard parameter

`GET /api/v2/spa/recommendations/{service}/{shard}`

This endpoint is currently experimental and restricted to Datadog internal use only. Retrieve resource recommendations for a Spark job. The caller (Spark Gateway or DJM UI) provides a service name and shard identifier, and SPA returns structured recommendations for driver and executor resources.

## Path parameters

- `shard` string, required
- `service` string, required

## Query parameters

- `bypass_cache` string

## Response `200`

OK

- RecommendationDocument — JSON:API document containing a single Recommendation resource. Returned by SPA when the Spark Gateway requests recommendations.
  - `data` RecommendationData, required — JSON:API resource object for SPA Recommendation. Includes type, optional ID, and resource attributes with structured recommendations.
    - `attributes` RecommendationAttributes, required — Attributes of the SPA Recommendation resource. Contains recommendations for both driver and executor components.
      - `confidence_level` number, double — The confidence level of the recommendation, expressed as a value between 0.0 (low confidence) and 1.0 (high confidence).
      - `driver` ComponentRecommendation, required — Resource recommendation for a single Spark component (driver or executor). Contains estimation data used to patch Spark job specs.
        - `estimation` Estimation, required — Recommended resource values for a Spark driver or executor, derived from recent real usage metrics. Used by SPA to propose more efficient pod sizing.
          - `cpu` Cpu — CPU usage statistics derived from historical Spark job metrics. Provides multiple estimates so users can choose between conservative and cost-saving risk profiles.
            - `max` integer — Maximum CPU usage observed for the job, expressed in millicores. This represents the upper bound of usage.
            - `p75` integer — 75th percentile of CPU usage (millicores). Represents a cost-saving configuration while covering most workloads.
            - `p95` integer — 95th percentile of CPU usage (millicores). Balances performance and cost, providing a safer margin than p75.
          - `ephemeral_storage` integer — Recommended ephemeral storage allocation (in MiB). Derived from job temporary storage patterns.
          - `heap` integer — Recommended JVM heap size (in MiB).
          - `memory` integer — Recommended total memory allocation (in MiB). Includes both heap and overhead.
          - `overhead` integer — Recommended JVM overhead (in MiB). Computed as total memory - heap.
      - `executor` ComponentRecommendation, required — Resource recommendation for a single Spark component (driver or executor). Contains estimation data used to patch Spark job specs.
        - `estimation` Estimation, required — Recommended resource values for a Spark driver or executor, derived from recent real usage metrics. Used by SPA to propose more efficient pod sizing.
          - `cpu` Cpu — CPU usage statistics derived from historical Spark job metrics. Provides multiple estimates so users can choose between conservative and cost-saving risk profiles.
            - `max` integer — Maximum CPU usage observed for the job, expressed in millicores. This represents the upper bound of usage.
            - `p75` integer — 75th percentile of CPU usage (millicores). Represents a cost-saving configuration while covering most workloads.
            - `p95` integer — 95th percentile of CPU usage (millicores). Balances performance and cost, providing a safer margin than p75.
          - `ephemeral_storage` integer — Recommended ephemeral storage allocation (in MiB). Derived from job temporary storage patterns.
          - `heap` integer — Recommended JVM heap size (in MiB).
          - `memory` integer — Recommended total memory allocation (in MiB). Includes both heap and overhead.
          - `overhead` integer — Recommended JVM overhead (in MiB). Computed as total memory - heap.
    - `id` string — Resource identifier for the recommendation. Optional in responses.
    - `type` 'recommendation', required — JSON:API resource type for Spark Pod Autosizing recommendations. Identifies the Recommendation resource returned by SPA.

## Other responses

- `400` — Bad Request
- `403` — Not Authorized
- `429` — Too many requests

---

[API](https://skmtc.net/datadog/apis/api-v2.md) · [All operations](https://skmtc.net/datadog/apis/api-v2/llms.txt) · [OpenAPI document](https://skmtc-service-staging.skmtc.workers.dev/v1/apis/datadog/api-v2/revisions/da68bf029e4c/schema)