Skip to content

Commit

Permalink
[Addon kubevela#603] Add Apache Spark as a experimental addon
Browse files Browse the repository at this point in the history 
Signed-off-by: yanghua <[email protected]>
  • Loading branch information
@yanghua
yanghua committed Feb 27, 2023
1 parent 285efc3 commit 843319f
Show file tree
Hide file tree
Showing 5 changed files with 396 additions and 0 deletions.
170 changes: 170 additions & 0 deletions experimental/addons/spark-kubernetes-operator/README.md
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,170 @@
# spark-kubernetes-operator

A kubernetes operator for Apache Spark(https://github.com/GoogleCloudPlatform/spark-on-k8s-operator), it allows users to manage Spark applications and their lifecycle through native K8S tooling like `kubectl`.

> Note: It's not provided by Apache Spark. But widely used by a large number of companies(https://github.com/GoogleCloudPlatform/spark-on-k8s-operator/blob/master/docs/who-is-using.md).
# Install

```
#The following steps are for enabling fluxcd and spark-kubernetes-operator in namespace called "spark-operator".
vela addon enable fluxcd
vela addon enable spark-kubernetes-operator
```

# Uninstall

```
vela addon disable spark-kubernetes-operator
vela addon disable fluxcd
```

# To check the spark-kubernetes-operator running status

* Firstly, check the spark-kubernetes-operator (and the fluxcd and we need to deploy by helm) running status

```
vela addon status spark-kubernetes-operator
vela ls -A | grep spark
```

* Secondly, show the component type `spark-cluster`, so we know how to use it in one application. As a spark user, you can choose the parameter to set for your spark cluster.

```
vela show spark-application
# Specification
+---------------------+------------------------------------------------------------------------------------------------------+-------------------+----------+---------+
| NAME | DESCRIPTION | TYPE | REQUIRED | DEFAULT |
+---------------------+------------------------------------------------------------------------------------------------------+-------------------+----------+---------+
| name | Specify the spark application name. | string | true | |
| namespace | Specify the namespace for spark application to install. | string | true | |
| type | Specify the application language type, e.g. "Scala", "Python", "Java" or "R". | string | true | |
| pythonVersion | Specify the python version. | string | false | |
| mode | Specify the deploy mode, e.go "cluster", "client" or "in-cluster-client". | string | true | |
| image | Specify the container image for the driver, executor, and init-container. | string | true | |
| imagePullPolicy | Specify the image pull policy for the driver, executor, and init-container. | string | true | |
| mainClass | Specify the fully-qualified main class of the Spark application. | string | true | |
| mainApplicationFile | Specify the path to a bundled JAR, Python, or R file of the application. | string | true | |
| sparkVersion | Specify the version of Spark the application uses. | string | true | |
| driverCores | Specify the number of CPU cores to request for the driver pod. | int | true | |
| executorCores | Specify the number of CPU cores to request for the executor pod. | int | true | |
| arguments | Specify a list of arguments to be passed to the application. | []string | false | |
| sparkConf | Specify the config information carries user-specified Spark configuration properties as they would | map[string]string | false | |
| | use the "--conf" option in spark-submit. | | | |
| hadoopConf | Specify the config information carries user-specified Hadoop configuration properties as they would | map[string]string | false | |
| | use the the "--conf" option in spark-submit. The SparkApplication controller automatically adds | | | |
| | prefix "spark.hadoop." to Hadoop configuration properties. | | | |
| sparkConfigMap | Specify the name of the ConfigMap containing Spark configuration files such as log4j.properties. The | string | false | |
| | controller will add environment variable SPARK_CONF_DIR to the path where the ConfigMap is mounted | | | |
| | to. | | | |
| hadoopConfigMap | Specify the name of the ConfigMap containing Hadoop configuration files such as core-site.xml. The | string | false | |
| | controller will add environment variable HADOOP_CONF_DIR to the path where the ConfigMap is mounted | | | |
| | to. | | | |
+---------------------+------------------------------------------------------------------------------------------------------+-------------------+----------+---------+
```

# Example for how to run a component typed spark-cluster in application

1. Firstly, copy the following example to "spark-app-v1.yaml":

> The addon will create a namespace named "spark-cluster"
```
apiVersion: core.oam.dev/v1beta1
kind: Application
metadata:
name: spark-app-v1
namespace: spark-cluster
spec:
components:
- name: my-spark-application-component
type: spark-application
properties:
name: my-spark-app
namespace: spark-cluster
type: Scala
mode: cluster
image: "gcr.io/spark-operator/spark:v3.1.1"
imagePullPolicy: Always
mainClass: org.apache.spark.examples.streaming.JavaQueueStream
mainApplicationFile: "local:///opt/spark/examples/jars/spark-examples_2.12-3.1.1.jar"
sparkVersion: "3.1.1"
driverCores: 1
executorCores: 1
```

2. Secondly, start the application:

```
vela up -f spark-app-v1.yaml
```

You will see the stdout like this:

```
Applying an application in vela K8s object format...
I0227 16:54:37.069480 361176 apply.go:121] "creating object" name="spark-app-v1" resource="core.oam.dev/v1beta1, Kind=Application"
✅ App has been deployed 🚀🚀🚀
Port forward: vela port-forward spark-app-v1 -n spark-cluster
SSH: vela exec spark-app-v1 -n spark-cluster
Logging: vela logs spark-app-v1 -n spark-cluster
App status: vela status spark-app-v1 -n spark-cluster
Endpoint: vela status spark-app-v1 -n spark-cluster --endpoint
Application spark-cluster/spark-app-v1 applied.
```

3. Then, you can use the native command to check the status of the Spark applicaiton:

```
$ kubectl get sparkapplications -n spark-cluster
NAME STATUS ATTEMPTS START FINISH AGE
my-spark-app RUNNING 1 2023-02-27T08:54:40Z <no value> 2m33s
```

or get the application detail via this command:

```
$ kubectl describe sparkapplication my-spark-app -n spark-cluster
Name: my-spark-app
Namespace: spark-cluster
Labels: app.oam.dev/app-revision-hash=4e5592aea53a5961
app.oam.dev/appRevision=spark-app-v1-v1
app.oam.dev/cluster=local
app.oam.dev/component=my-spark-application-component
app.oam.dev/name=spark-app-v1
app.oam.dev/namespace=spark-cluster
app.oam.dev/resourceType=TRAIT
app.oam.dev/revision=
oam.dev/render-hash=640a3298d803274e
trait.oam.dev/resource=spark
trait.oam.dev/type=AuxiliaryWorkload
Annotations: app.oam.dev/last-applied-configuration:
{"apiVersion":"sparkoperator.k8s.io/v1beta2","kind":"SparkApplication","metadata":{"annotations":{"app.oam.dev/last-applied-time":"2023-02...
app.oam.dev/last-applied-time: 2023-02-27T16:54:37+08:00
oam.dev/kubevela-version: v1.7.0
API Version: sparkoperator.k8s.io/v1beta2
Kind: SparkApplication
Metadata:
......
```

or get the general purpose detail information via this command:

```
$ kubectl get app spark-app-v1 -n spark-cluster -oyaml
apiVersion: core.oam.dev/v1beta1
kind: Application
metadata:
......
```

4. Show the service of spark application via this command:

```
$ kubectl get svc -n spark-cluster
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
my-spark-app-c58a1c869214bfe5-driver-svc ClusterIP None <none> 7078/TCP,7079/TCP,4040/TCP 19m
my-spark-app-ui-svc ClusterIP xx.xx.xx.xx <none> 4040/TCP 19m
```
90 changes: 90 additions & 0 deletions experimental/addons/spark-kubernetes-operator/definitions/spark-application.cue
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,90 @@
"spark-application": {
annotations: {}
attributes: workload: type: "autodetects.core.oam.dev"
description: "Describes a containerized spark application that can specify resource spec."
labels: {}
type: "component"
}

template: {
parameter: {
// +usage=Specify the spark application name
name: string
// +usage=Specify the namespace for spark application to install
namespace: string
// +usage=Specify the application language type, e.g. "Scala", "Python", "Java" or "R"
type: string
// +usage=Specify the python version
pythonVersion ?: string
// +usage=Specify the deploy mode, e.go "cluster", "client" or "in-cluster-client"
mode: string
// +usage=Specify the container image for the driver, executor, and init-container
image: string
// +usage=Specify the image pull policy for the driver, executor, and init-container
imagePullPolicy: string
// +usage=Specify the fully-qualified main class of the Spark application
mainClass: string
// +usage=Specify the path to a bundled JAR, Python, or R file of the application
mainApplicationFile: string
// +usage=Specify the version of Spark the application uses
sparkVersion: string
// +usage=Specify the number of CPU cores to request for the driver pod
driverCores: int
// +usage=Specify the number of CPU cores to request for the executor pod
executorCores: int
// +usage=Specify a list of arguments to be passed to the application
arguments ?: [...string]
// +usage=Specify the config information carries user-specified Spark configuration properties as they would use the "--conf" option in spark-submit
sparkConf ?: [string]: string
// +usage=Specify the config information carries user-specified Hadoop configuration properties as they would use the the "--conf" option in spark-submit. The SparkApplication controller automatically adds prefix "spark.hadoop." to Hadoop configuration properties
hadoopConf ?: [string]: string
// +usage=Specify the name of the ConfigMap containing Spark configuration files such as log4j.properties. The controller will add environment variable SPARK_CONF_DIR to the path where the ConfigMap is mounted to
sparkConfigMap ?: string
// +usage=Specify the name of the ConfigMap containing Hadoop configuration files such as core-site.xml. The controller will add environment variable HADOOP_CONF_DIR to the path where the ConfigMap is mounted to
hadoopConfigMap ?: string

}

output: {
kind: "ClusterRoleBinding"
apiVersion: "rbac.authorization.k8s.io/v1"
metadata: name: parameter.name
roleRef: {
name: "edit"
apiGroup: "rbac.authorization.k8s.io"
kind: "ClusterRole"
}
subjects: [{
name: "default"
kind: "ServiceAccount"
namespace: parameter.namespace
}]
}

outputs: {

"spark": {
kind: "SparkApplication"
apiVersion: "sparkoperator.k8s.io/v1beta2"
metadata: {
name: parameter.name
namespace: parameter.namespace
}
spec: {
type: parameter.type
mode: parameter.mode
image: parameter.image
imagePullPolicy: parameter.imagePullPolicy
mainClass: parameter.mainClass
mainApplicationFile: parameter.mainApplicationFile
sparkVersion: parameter.sparkVersion
driver: {
cores: parameter.driverCores
}
executor: {
cores: parameter.executorCores
}
}
}
}
}
15 changes: 15 additions & 0 deletions experimental/addons/spark-kubernetes-operator/metadata.yaml
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,15 @@
description: A kubernetes operator for Apache Spark
icon: "https://spark.apache.org/images/spark-logo.png"
invisible: false
name: spark-kubernetes-operator
tags:
- GoogleCloudPlatform/spark-on-k8s-operator
version: v1beta2-1.3.8-3.1.1
url: https://github.com/GoogleCloudPlatform/spark-on-k8s-operator

dependencies:
- name: fluxcd

system:
vela: ">=1.5.0-beta.3"
kubernetes: ">=1.16"
20 changes: 20 additions & 0 deletions experimental/addons/spark-kubernetes-operator/parameter.cue
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,20 @@
// parameter.cue is used to store addon parameters.
//
// You can use these parameters in template.cue or in resources/ by 'parameter.myparam'
//
// For example, you can use parameters to allow the user to customize
// container images, ports, and etc.
parameter: {
// +usage=Deploy to specified clusters. Leave empty to deploy to all clusters.
clusters?: [...string]
// +usage=Namespace to deploy to
namespace: *"spark-operator" | string
// +usage=Specify if create the webhook or not
"createWebhook": *false | bool
// +usage=Specify the image repository
"imageRepository": *"ghcr.io/googlecloudplatform/spark-operator" | string
// +usage=Specify the image tag
"imageTag": *"v1beta2-1.3.8-3.1.1" | string
// +usage=Specify if create the sa for job or not
"createSparkServiceAccount": *false|bool
}
Loading

0 comments on commit 843319f

Please sign in to comment.