Merge pull request #943 from Vafilor/feat/model.deployment

feat: Inference Service API
onepanelio · Aug 12, 2021 · 5705bfa · 5705bfa
2 parents 147c937 + 44a78ef
commit 5705bfa
Show file tree

Hide file tree

Showing 10 changed files with 2,408 additions and 0 deletions.
diff --git a/api/api.swagger.json b/api/api.swagger.json
@@ -763,6 +763,114 @@
  ]
  }
  },
+ "/apis/v1beta1/{namespace}/inferenceservice": {
+ "post": {
+ "operationId": "CreateInferenceService",
+ "responses": {
+ "200": {
+ "description": "A successful response.",
+ "schema": {
+ "$ref": "#/definitions/GetInferenceServiceResponse"
+ }
+ },
+ "default": {
+ "description": "An unexpected error response.",
+ "schema": {
+ "$ref": "#/definitions/google.rpc.Status"
+ }
+ }
+ },
+ "parameters": [
+ {
+ "name": "namespace",
+ "in": "path",
+ "required": true,
+ "type": "string"
+ },
+ {
+ "name": "body",
+ "in": "body",
+ "required": true,
+ "schema": {
+ "$ref": "#/definitions/CreateInferenceServiceRequest"
+ }
+ }
+ ],
+ "tags": [
+ "InferenceService"
+ ]
+ }
+ },
+ "/apis/v1beta1/{namespace}/inferenceservice/{name}": {
+ "get": {
+ "operationId": "GetInferenceService",
+ "responses": {
+ "200": {
+ "description": "A successful response.",
+ "schema": {
+ "$ref": "#/definitions/GetInferenceServiceResponse"
+ }
+ },
+ "default": {
+ "description": "An unexpected error response.",
+ "schema": {
+ "$ref": "#/definitions/google.rpc.Status"
+ }
+ }
+ },
+ "parameters": [
+ {
+ "name": "namespace",
+ "in": "path",
+ "required": true,
+ "type": "string"
+ },
+ {
+ "name": "name",
+ "in": "path",
+ "required": true,
+ "type": "string"
+ }
+ ],
+ "tags": [
+ "InferenceService"
+ ]
+ },
+ "delete": {
+ "operationId": "DeleteInferenceService",
+ "responses": {
+ "200": {
+ "description": "A successful response.",
+ "schema": {
+ "properties": {}
+ }
+ },
+ "default": {
+ "description": "An unexpected error response.",
+ "schema": {
+ "$ref": "#/definitions/google.rpc.Status"
+ }
+ }
+ },
+ "parameters": [
+ {
+ "name": "namespace",
+ "in": "path",
+ "required": true,
+ "type": "string"
+ },
+ {
+ "name": "name",
+ "in": "path",
+ "required": true,
+ "type": "string"
+ }
+ ],
+ "tags": [
+ "InferenceService"
+ ]
+ }
+ },
  "/apis/v1beta1/{namespace}/secrets": {
  "get": {
  "operationId": "ListSecrets",
@@ -3260,6 +3368,43 @@
  }
  }
  },
+ "Container": {
+ "type": "object",
+ "properties": {
+ "image": {
+ "type": "string"
+ },
+ "name": {
+ "type": "string"
+ },
+ "env": {
+ "type": "array",
+ "items": {
+ "$ref": "#/definitions/Env"
+ }
+ }
+ }
+ },
+ "CreateInferenceServiceRequest": {
+ "type": "object",
+ "properties": {
+ "namespace": {
+ "type": "string"
+ },
+ "name": {
+ "type": "string"
+ },
+ "defaultTransformerImage": {
+ "type": "string"
+ },
+ "predictor": {
+ "$ref": "#/definitions/InferenceServicePredictor"
+ },
+ "transformer": {
+ "$ref": "#/definitions/InferenceServiceTransformer"
+ }
+ }
+ },
  "CreateWorkflowExecutionBody": {
  "type": "object",
  "properties": {
@@ -3362,6 +3507,17 @@
  }
  }
  },
+ "Env": {
+ "type": "object",
+ "properties": {
+ "name": {
+ "type": "string"
+ },
+ "value": {
+ "type": "string"
+ }
+ }
+ },
  "File": {
  "type": "object",
  "properties": {
@@ -3431,6 +3587,23 @@
  }
  }
  },
+ "GetInferenceServiceResponse": {
+ "type": "object",
+ "properties": {
+ "ready": {
+ "type": "boolean"
+ },
+ "conditions": {
+ "type": "array",
+ "items": {
+ "$ref": "#/definitions/InferenceServiceCondition"
+ }
+ },
+ "predictUrl": {
+ "type": "string"
+ }
+ }
+ },
  "GetLabelsResponse": {
  "type": "object",
  "properties": {
@@ -3497,6 +3670,72 @@
  }
  }
  },
+ "InferenceServiceCondition": {
+ "type": "object",
+ "properties": {
+ "lastTransitionTime": {
+ "type": "string"
+ },
+ "status": {
+ "type": "string"
+ },
+ "type": {
+ "type": "string"
+ }
+ }
+ },
+ "InferenceServicePredictor": {
+ "type": "object",
+ "properties": {
+ "name": {
+ "type": "string"
+ },
+ "runtimeVersion": {
+ "type": "string"
+ },
+ "storageUri": {
+ "type": "string"
+ },
+ "nodeSelector": {
+ "type": "string"
+ },
+ "minCpu": {
+ "type": "string"
+ },
+ "minMemory": {
+ "type": "string"
+ },
+ "maxCpu": {
+ "type": "string"
+ },
+ "maxMemory": {
+ "type": "string"
+ }
+ }
+ },
+ "InferenceServiceTransformer": {
+ "type": "object",
+ "properties": {
+ "containers": {
+ "type": "array",
+ "items": {
+ "$ref": "#/definitions/Container"
+ }
+ },
+ "minCpu": {
+ "type": "string"
+ },
+ "minMemory": {
+ "type": "string"
+ },
+ "maxCpu": {
+ "type": "string"
+ },
+ "maxMemory": {
+ "type": "string"
+ }
+ }
+ },
  "IsAuthorized": {
  "type": "object",
  "properties": {