RFC: report compute providers and release stage per Namespace

openapi/openapiv2.json

@@ -15074,6 +15074,35 @@
       },
       "description": "ComputeProvider stores information used by a worker control plane controller\nto respond to worker lifecycle events. For example, when a Task is received\non a TaskQueue that has no active pollers, a serverless worker lifecycle\ncontroller might need to invoke an AWS Lambda Function that itself ends up\ncalling the SDK's worker.New() function."
     },
+    "v1ComputeProviderReleaseStage": {
+      "type": "string",
+      "enum": [
+        "COMPUTE_PROVIDER_RELEASE_STAGE_UNSPECIFIED",
+        "COMPUTE_PROVIDER_RELEASE_STAGE_PRE_RELEASE",
+        "COMPUTE_PROVIDER_RELEASE_STAGE_PUBLIC_PREVIEW",
+        "COMPUTE_PROVIDER_RELEASE_STAGE_GENERALLY_AVAILABLE"
+      ],
+      "default": "COMPUTE_PROVIDER_RELEASE_STAGE_UNSPECIFIED",
+      "description": "How far along its rollout a compute provider is. Clients use this to label a\nprovider in a picker, and to decide whether to offer it prominently, so the\nlabel follows the server rather than being hardcoded per client.\n\n - COMPUTE_PROVIDER_RELEASE_STAGE_UNSPECIFIED: Release stage is not specified. Clients should treat the provider as\nusable but unlabeled.\n - COMPUTE_PROVIDER_RELEASE_STAGE_PRE_RELEASE: The provider is implemented but not yet ready for general use. Expect\nbreaking changes. Clients should label it accordingly and may choose not\nto offer it by default.\n - COMPUTE_PROVIDER_RELEASE_STAGE_PUBLIC_PREVIEW: The provider is available to all users but still gathering feedback, and\nmay change in backwards-incompatible ways before becoming generally\navailable.\n - COMPUTE_PROVIDER_RELEASE_STAGE_GENERALLY_AVAILABLE: The provider is generally available and covered by compatibility\nguarantees."
+    },
+    "v1ComputeProviderStatus": {
+      "type": "object",
+      "properties": {
+        "type": {
+          "type": "string",
+          "description": "The provider type, matching ComputeProvider.type. This is the value a\nclient sends back when configuring a scaling group, so it is the\nprovider's identity, not a display name."
+        },
+        "enabled": {
+          "type": "boolean",
+          "description": "Whether this Service will accept a ComputeConfig naming this provider.\nDepends both on the build and on Service configuration, the same way the\nper-capability booleans in GetSystemInfoResponse.Capabilities do."
+        },
+        "releaseStage": {
+          "$ref": "#/definitions/v1ComputeProviderReleaseStage",
+          "description": "How far along its rollout the provider is. Lets a client label a\nprovider without shipping its own release-stage table, and lets the\nlabel change without a client release."
+        }
+      },
+      "description": "ComputeProviderStatus describes one compute provider this Service knows\nabout, so a client can discover which providers exist and how to present\nthem without hardcoding a list of its own.\n\nThis is typed where ComputeProvider.details is an opaque Payload, and the\ndifference is deliberate. Provider *config* is open-ended and genuinely\nprovider-specific: a Lambda ARN, a Cloud Run worker pool, and an AgentCore\nendpoint ARN share no shape, which is why an earlier typed\nProviderDetailAWSLambda gave way to a Payload. Provider *status* is uniform.\nEvery provider has an identity, an enabled bit, and a release stage, and none\nneeds a field the others do not.\n\nA Service reports a provider it recognizes even when that provider is\ndisabled, so a client can tell \"this Service has never heard of AgentCore\"\napart from \"this Service knows AgentCore but it is turned off here\", and say\nso rather than silently omitting the option."
+    },
     "v1ComputeScaler": {
       "type": "object",
       "properties": {
@@ -16326,7 +16355,7 @@
         },
         "serverScaledProviderCloudRun": {
           "type": "boolean",
-          "description": "True if the server supports the Cloud Run compute provider for\nserver-scaled deployments. Dependent on server version and the\nprovider being enabled via server configuration."
+          "description": "True if the server supports the Cloud Run compute provider for\nserver-scaled deployments. Dependent on server version and the\nprovider being enabled via server configuration.\n\nSuperseded by NamespaceInfo.compute_providers, which reports every\nprovider and its release stage per Namespace rather than one boolean\nper provider for the whole Service. Retained for wire compatibility\nwith clients that already read it. No further per-provider booleans\nshould be added here: this response is served as a fixed payload in\nsome deployments, so it cannot express per-Namespace availability."
         }
       },
       "description": "System capability details."
@@ -17183,6 +17212,14 @@
           "$ref": "#/definitions/NamespaceInfoLimits",
           "title": "Namespace configured limits"
         },
+        "computeProviders": {
+          "type": "array",
+          "items": {
+            "type": "object",
+            "$ref": "#/definitions/v1ComputeProviderStatus"
+          },
+          "description": "The compute providers available for server-scaled Worker Deployments in\nthis Namespace: which ones this Namespace can use, whether each is\ncurrently enabled, and how far along its rollout each one is.\n\nNamespace-scoped rather than Service-scoped for two reasons. Availability\ngenuinely varies per Namespace: a provider is tied to the cloud a\nNamespace runs in, so an AWS Namespace and a GCP Namespace do not offer\nthe same set, and managed deployments gate providers per account.\nGetSystemInfo cannot express either, and is served as a fixed payload\nahead of the Service in some deployments, so a per-provider boolean there\nis not reachable as a real signal."
+        },
         "supportsSchedules": {
           "type": "boolean",
           "description": "Whether scheduled workflows are supported on this namespace. This is only needed\ntemporarily while the feature is experimental, so we can give it a high tag."

openapi/openapiv3.yaml

@@ -11255,6 +11255,50 @@ components:
          on a TaskQueue that has no active pollers, a serverless worker lifecycle
          controller might need to invoke an AWS Lambda Function that itself ends up
          calling the SDK's worker.New() function.
+    ComputeProviderStatus:
+      type: object
+      properties:
+        type:
+          type: string
+          description: |-
+            The provider type, matching ComputeProvider.type. This is the value a
+             client sends back when configuring a scaling group, so it is the
+             provider's identity, not a display name.
+        enabled:
+          type: boolean
+          description: |-
+            Whether this Service will accept a ComputeConfig naming this provider.
+             Depends both on the build and on Service configuration, the same way the
+             per-capability booleans in GetSystemInfoResponse.Capabilities do.
+        releaseStage:
+          enum:
+            - COMPUTE_PROVIDER_RELEASE_STAGE_UNSPECIFIED
+            - COMPUTE_PROVIDER_RELEASE_STAGE_PRE_RELEASE
+            - COMPUTE_PROVIDER_RELEASE_STAGE_PUBLIC_PREVIEW
+            - COMPUTE_PROVIDER_RELEASE_STAGE_GENERALLY_AVAILABLE
+          type: string
+          description: |-
+            How far along its rollout the provider is. Lets a client label a
+             provider without shipping its own release-stage table, and lets the
+             label change without a client release.
+          format: enum
+      description: |-
+        ComputeProviderStatus describes one compute provider this Service knows
+         about, so a client can discover which providers exist and how to present
+         them without hardcoding a list of its own.
+
+         This is typed where ComputeProvider.details is an opaque Payload, and the
+         difference is deliberate. Provider *config* is open-ended and genuinely
+         provider-specific: a Lambda ARN, a Cloud Run worker pool, and an AgentCore
+         endpoint ARN share no shape, which is why an earlier typed
+         ProviderDetailAWSLambda gave way to a Payload. Provider *status* is uniform.
+         Every provider has an identity, an enabled bit, and a release stage, and none
+         needs a field the others do not.
+
+         A Service reports a provider it recognizes even when that provider is
+         disabled, so a client can tell "this Service has never heard of AgentCore"
+         apart from "this Service knows AgentCore but it is turned off here", and say
+         so rather than silently omitting the option.
     ComputeScaler:
       type: object
       properties:
@@ -12648,6 +12692,13 @@ components:
             True if the server supports the Cloud Run compute provider for
              server-scaled deployments. Dependent on server version and the
              provider being enabled via server configuration.
+
+             Superseded by NamespaceInfo.compute_providers, which reports every
+             provider and its release stage per Namespace rather than one boolean
+             per provider for the whole Service. Retained for wire compatibility
+             with clients that already read it. No further per-provider booleans
+             should be added here: this response is served as a fixed payload in
+             some deployments, so it cannot express per-Namespace availability.
       description: System capability details.
     GetWorkerBuildIdCompatibilityResponse:
       type: object
@@ -13480,6 +13531,22 @@ components:
           allOf:
             - $ref: '#/components/schemas/NamespaceInfo_Limits'
           description: Namespace configured limits
+        computeProviders:
+          type: array
+          items:
+            $ref: '#/components/schemas/ComputeProviderStatus'
+          description: |-
+            The compute providers available for server-scaled Worker Deployments in
+             this Namespace: which ones this Namespace can use, whether each is
+             currently enabled, and how far along its rollout each one is.
+
+             Namespace-scoped rather than Service-scoped for two reasons. Availability
+             genuinely varies per Namespace: a provider is tied to the cloud a
+             Namespace runs in, so an AWS Namespace and a GCP Namespace do not offer
+             the same set, and managed deployments gate providers per account.
+             GetSystemInfo cannot express either, and is served as a fixed payload
+             ahead of the Service in some deployments, so a per-provider boolean there
+             is not reachable as a real signal.
         supportsSchedules:
           type: boolean
           description: |-

temporal/api/compute/v1/status.proto

@@ -0,0 +1,45 @@
+syntax = "proto3";
+
+package temporal.api.compute.v1;
+
+option go_package = "go.temporal.io/api/compute/v1;compute";
+option java_package = "io.temporal.api.compute.v1";
+option java_multiple_files = true;
+option java_outer_classname = "StatusProto";
+option ruby_package = "Temporalio::Api::Compute::V1";
+option csharp_namespace = "Temporalio.Api.Compute.V1";
+
+import "temporal/api/enums/v1/compute.proto";
+
+// ComputeProviderStatus describes one compute provider this Service knows
+// about, so a client can discover which providers exist and how to present
+// them without hardcoding a list of its own.
+//
+// This is typed where ComputeProvider.details is an opaque Payload, and the
+// difference is deliberate. Provider *config* is open-ended and genuinely
+// provider-specific: a Lambda ARN, a Cloud Run worker pool, and an AgentCore
+// endpoint ARN share no shape, which is why an earlier typed
+// ProviderDetailAWSLambda gave way to a Payload. Provider *status* is uniform.
+// Every provider has an identity, an enabled bit, and a release stage, and none
+// needs a field the others do not.
+//
+// A Service reports a provider it recognizes even when that provider is
+// disabled, so a client can tell "this Service has never heard of AgentCore"
+// apart from "this Service knows AgentCore but it is turned off here", and say
+// so rather than silently omitting the option.
+message ComputeProviderStatus {
+    // The provider type, matching ComputeProvider.type. This is the value a
+    // client sends back when configuring a scaling group, so it is the
+    // provider's identity, not a display name.
+    string type = 1;
+
+    // Whether this Service will accept a ComputeConfig naming this provider.
+    // Depends both on the build and on Service configuration, the same way the
+    // per-capability booleans in GetSystemInfoResponse.Capabilities do.
+    bool enabled = 2;
+
+    // How far along its rollout the provider is. Lets a client label a
+    // provider without shipping its own release-stage table, and lets the
+    // label change without a client release.
+    temporal.api.enums.v1.ComputeProviderReleaseStage release_stage = 3;
+}

temporal/api/enums/v1/compute.proto

@@ -0,0 +1,30 @@
+syntax = "proto3";
+
+package temporal.api.enums.v1;
+
+option go_package = "go.temporal.io/api/enums/v1;enums";
+option java_package = "io.temporal.api.enums.v1";
+option java_multiple_files = true;
+option java_outer_classname = "ComputeProto";
+option ruby_package = "Temporalio::Api::Enums::V1";
+option csharp_namespace = "Temporalio.Api.Enums.V1";
+
+// How far along its rollout a compute provider is. Clients use this to label a
+// provider in a picker, and to decide whether to offer it prominently, so the
+// label follows the server rather than being hardcoded per client.
+enum ComputeProviderReleaseStage {
+    // Release stage is not specified. Clients should treat the provider as
+    // usable but unlabeled.
+    COMPUTE_PROVIDER_RELEASE_STAGE_UNSPECIFIED = 0;
+    // The provider is implemented but not yet ready for general use. Expect
+    // breaking changes. Clients should label it accordingly and may choose not
+    // to offer it by default.
+    COMPUTE_PROVIDER_RELEASE_STAGE_PRE_RELEASE = 1;
+    // The provider is available to all users but still gathering feedback, and
+    // may change in backwards-incompatible ways before becoming generally
+    // available.
+    COMPUTE_PROVIDER_RELEASE_STAGE_PUBLIC_PREVIEW = 2;
+    // The provider is generally available and covered by compatibility
+    // guarantees.
+    COMPUTE_PROVIDER_RELEASE_STAGE_GENERALLY_AVAILABLE = 3;
+}

temporal/api/namespace/v1/message.proto

@@ -12,6 +12,7 @@ option csharp_namespace = "Temporalio.Api.Namespace.V1";
 import "google/protobuf/duration.proto";
 import "google/protobuf/timestamp.proto";
 
+import "temporal/api/compute/v1/status.proto";
 import "temporal/api/enums/v1/namespace.proto";
 
 
@@ -83,6 +84,19 @@ message NamespaceInfo {
         int64 workflow_task_completion_size_limit_error = 3;
     }
 
+    // The compute providers available for server-scaled Worker Deployments in
+    // this Namespace: which ones this Namespace can use, whether each is
+    // currently enabled, and how far along its rollout each one is.
+    //
+    // Namespace-scoped rather than Service-scoped for two reasons. Availability
+    // genuinely varies per Namespace: a provider is tied to the cloud a
+    // Namespace runs in, so an AWS Namespace and a GCP Namespace do not offer
+    // the same set, and managed deployments gate providers per account.
+    // GetSystemInfo cannot express either, and is served as a fixed payload
+    // ahead of the Service in some deployments, so a per-provider boolean there
+    // is not reachable as a real signal.
+    repeated temporal.api.compute.v1.ComputeProviderStatus compute_providers = 9;
+
     // Whether scheduled workflows are supported on this namespace. This is only needed
     // temporarily while the feature is experimental, so we can give it a high tag.
     bool supports_schedules = 100;

temporal/api/workflowservice/v1/request_response.proto

@@ -1392,6 +1392,13 @@ message GetSystemInfoResponse {
         // True if the server supports the Cloud Run compute provider for
         // server-scaled deployments. Dependent on server version and the
         // provider being enabled via server configuration.
+        //
+        // Superseded by NamespaceInfo.compute_providers, which reports every
+        // provider and its release stage per Namespace rather than one boolean
+        // per provider for the whole Service. Retained for wire compatibility
+        // with clients that already read it. No further per-provider booleans
+        // should be added here: this response is served as a fixed payload in
+        // some deployments, so it cannot express per-Namespace availability.
         bool server_scaled_provider_cloud_run = 13;
 
     }