feat: Support customizable deployment strategy for RawDeployment mode. …

…Fixes kserve#3452 (kserve#3603) * feat: Support customizable deployment strategy for RawDeployment mode Signed-off-by: Yuan Tang <[email protected]> * regen Signed-off-by: Yuan Tang <[email protected]> * lint Signed-off-by: Yuan Tang <[email protected]> * Correctly apply rollingupdate Signed-off-by: Yuan Tang <[email protected]> * address comments Signed-off-by: Yuan Tang <[email protected]> * Add validation Signed-off-by: Yuan Tang <[email protected]> --------- Signed-off-by: Yuan Tang <[email protected]> Signed-off-by: asd981256 <[email protected]>
asd981256 · May 14, 2024 · 2a22f5a · 2a22f5a
1 parent 4437792
commit 2a22f5a
Show file tree

Hide file tree

Showing 31 changed files with 1,126 additions and 303 deletions.
diff --git a/config/crd/full/serving.kserve.io_inferenceservices.yaml b/config/crd/full/serving.kserve.io_inferenceservices.yaml
@@ -1688,6 +1688,24 @@ spec:
  - name
  type: object
  type: array
+ deploymentStrategy:
+ properties:
+ rollingUpdate:
+ properties:
+ maxSurge:
+ anyOf:
+ - type: integer
+ - type: string
+ x-kubernetes-int-or-string: true
+ maxUnavailable:
+ anyOf:
+ - type: integer
+ - type: string
+ x-kubernetes-int-or-string: true
+ type: object
+ type:
+ type: string
+ type: object
  dnsConfig:
  properties:
  nameservers:
@@ -4333,6 +4351,24 @@ spec:
  - name
  type: object
  type: array
+ deploymentStrategy:
+ properties:
+ rollingUpdate:
+ properties:
+ maxSurge:
+ anyOf:
+ - type: integer
+ - type: string
+ x-kubernetes-int-or-string: true
+ maxUnavailable:
+ anyOf:
+ - type: integer
+ - type: string
+ x-kubernetes-int-or-string: true
+ type: object
+ type:
+ type: string
+ type: object
  dnsConfig:
  properties:
  nameservers:
@@ -13886,6 +13922,24 @@ spec:
  - name
  type: object
  type: array
+ deploymentStrategy:
+ properties:
+ rollingUpdate:
+ properties:
+ maxSurge:
+ anyOf:
+ - type: integer
+ - type: string
+ x-kubernetes-int-or-string: true
+ maxUnavailable:
+ anyOf:
+ - type: integer
+ - type: string
+ x-kubernetes-int-or-string: true
+ type: object
+ type:
+ type: string
+ type: object
  dnsConfig:
  properties:
  nameservers:

diff --git a/docs/samples/client/kfserving_sdk_v1beta1_sample.ipynb b/docs/samples/client/kfserving_sdk_v1beta1_sample.ipynb
@@ -22,7 +22,7 @@
  "metadata": {},
  "outputs": [],
  "source": [
- "from kubernetes import client \n",
+ "from kubernetes import client\n",
  "from kfserving import KFServingClient\n",
  "from kfserving import constants\n",
  "from kfserving import utils\n",
@@ -45,8 +45,8 @@
  "metadata": {},
  "outputs": [],
  "source": [
- "#namespace = utils.get_default_target_namespace()\n",
- "namespace = 'kfserving-test'"
+ "# namespace = utils.get_default_target_namespace()\n",
+ "namespace = \"kfserving-test\""
  ]
  },
  {
@@ -69,16 +69,21 @@
  "metadata": {},
  "outputs": [],
  "source": [
- "api_version = constants.KFSERVING_GROUP + '/' + kfserving_version\n",
+ "api_version = constants.KFSERVING_GROUP + \"/\" + kfserving_version\n",
  "\n",
- "isvc = V1beta1InferenceService(api_version=api_version,\n",
- " kind=constants.KFSERVING_KIND,\n",
- " metadata=client.V1ObjectMeta(\n",
- " name='flower-sample', namespace=namespace),\n",
- " spec=V1beta1InferenceServiceSpec(\n",
- " predictor=V1beta1PredictorSpec(\n",
- " tensorflow=(V1beta1TFServingSpec(\n",
- " storage_uri='gs://kfserving-examples/models/tensorflow/flowers'))))\n",
+ "isvc = V1beta1InferenceService(\n",
+ " api_version=api_version,\n",
+ " kind=constants.KFSERVING_KIND,\n",
+ " metadata=client.V1ObjectMeta(name=\"flower-sample\", namespace=namespace),\n",
+ " spec=V1beta1InferenceServiceSpec(\n",
+ " predictor=V1beta1PredictorSpec(\n",
+ " tensorflow=(\n",
+ " V1beta1TFServingSpec(\n",
+ " storage_uri=\"gs://kfserving-examples/models/tensorflow/flowers\"\n",
+ " )\n",
+ " )\n",
+ " )\n",
+ " ),\n",
  ")"
  ]
  },
@@ -152,7 +157,7 @@
  }
  ],
  "source": [
- "KFServing.get('flower-sample', namespace=namespace, watch=True, timeout_seconds=120)"
+ "KFServing.get(\"flower-sample\", namespace=namespace, watch=True, timeout_seconds=120)"
  ]
  },
  {
@@ -223,18 +228,23 @@
  }
  ],
  "source": [
- "isvc = V1beta1InferenceService(api_version=api_version,\n",
- " kind=constants.KFSERVING_KIND,\n",
- " metadata=client.V1ObjectMeta(\n",
- " name='flower-sample', namespace=namespace),\n",
- " spec=V1beta1InferenceServiceSpec(\n",
- " predictor=V1beta1PredictorSpec(\n",
- " canary_traffic_percent=20,\n",
- " tensorflow=(V1beta1TFServingSpec(\n",
- " storage_uri='gs://kfserving-examples/models/tensorflow/flowers-2'))))\n",
+ "isvc = V1beta1InferenceService(\n",
+ " api_version=api_version,\n",
+ " kind=constants.KFSERVING_KIND,\n",
+ " metadata=client.V1ObjectMeta(name=\"flower-sample\", namespace=namespace),\n",
+ " spec=V1beta1InferenceServiceSpec(\n",
+ " predictor=V1beta1PredictorSpec(\n",
+ " canary_traffic_percent=20,\n",
+ " tensorflow=(\n",
+ " V1beta1TFServingSpec(\n",
+ " storage_uri=\"gs://kfserving-examples/models/tensorflow/flowers-2\"\n",
+ " )\n",
+ " ),\n",
+ " )\n",
+ " ),\n",
  ")\n",
  "\n",
- "KFServing.patch('flower-sample', isvc, namespace=namespace)"
+ "KFServing.patch(\"flower-sample\", isvc, namespace=namespace)"
  ]
  },
  {
@@ -250,7 +260,7 @@
  "metadata": {},
  "outputs": [],
  "source": [
- "KFServing.wait_isvc_ready('flower-sample', namespace=namespace)"
+ "KFServing.wait_isvc_ready(\"flower-sample\", namespace=namespace)"
  ]
  },
  {
@@ -268,7 +278,7 @@
  }
  ],
  "source": [
- "KFServing.get('flower-sample', namespace=namespace, watch=True)"
+ "KFServing.get(\"flower-sample\", namespace=namespace, watch=True)"
  ]
  },
  {
@@ -313,7 +323,7 @@
  }
  ],
  "source": [
- "KFServing.delete('flower-sample', namespace=namespace)"
+ "KFServing.delete(\"flower-sample\", namespace=namespace)"
  ]
  },
  {