forked from kubeflow/spark-operator
-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.go
272 lines (235 loc) · 10.8 KB
/
main.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
/*
Copyright 2017 Google LLC
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
https://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/
//go:generate hack/update-codegen.sh
package main
import (
"context"
"flag"
"fmt"
"os"
"os/signal"
"strings"
"syscall"
"time"
"github.com/golang/glog"
apiv1 "k8s.io/api/core/v1"
apiextensionsclient "k8s.io/apiextensions-apiserver/pkg/client/clientset/clientset"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/util/clock"
"k8s.io/client-go/informers"
clientset "k8s.io/client-go/kubernetes"
_ "k8s.io/client-go/plugin/pkg/client/auth/gcp"
"k8s.io/client-go/rest"
"k8s.io/client-go/tools/clientcmd"
"k8s.io/client-go/tools/leaderelection"
"k8s.io/client-go/tools/leaderelection/resourcelock"
"k8s.io/client-go/tools/record"
"github.com/GoogleCloudPlatform/spark-on-k8s-operator/pkg/batchscheduler"
crclientset "github.com/GoogleCloudPlatform/spark-on-k8s-operator/pkg/client/clientset/versioned"
crinformers "github.com/GoogleCloudPlatform/spark-on-k8s-operator/pkg/client/informers/externalversions"
operatorConfig "github.com/GoogleCloudPlatform/spark-on-k8s-operator/pkg/config"
"github.com/GoogleCloudPlatform/spark-on-k8s-operator/pkg/controller/scheduledsparkapplication"
"github.com/GoogleCloudPlatform/spark-on-k8s-operator/pkg/controller/sparkapplication"
"github.com/GoogleCloudPlatform/spark-on-k8s-operator/pkg/util"
"github.com/GoogleCloudPlatform/spark-on-k8s-operator/pkg/webhook"
)
var (
master = flag.String("master", "", "The address of the Kubernetes API server. Overrides any value in kubeconfig. Only required if out-of-cluster.")
kubeConfig = flag.String("kubeConfig", "", "Path to a kube config. Only required if out-of-cluster.")
controllerThreads = flag.Int("controller-threads", 10, "Number of worker threads used by the SparkApplication controller.")
resyncInterval = flag.Int("resync-interval", 30, "Informer resync interval in seconds.")
namespace = flag.String("namespace", apiv1.NamespaceAll, "The Kubernetes namespace to manage. Will manage custom resource objects of the managed CRD types for the whole cluster if unset.")
enableWebhook = flag.Bool("enable-webhook", false, "Whether to enable the mutating admission webhook for admitting and patching Spark pods.")
enableResourceQuotaEnforcement = flag.Bool("enable-resource-quota-enforcement", false, "Whether to enable ResourceQuota enforcement for SparkApplication resources. Requires the webhook to be enabled.")
enableMetrics = flag.Bool("enable-metrics", false, "Whether to enable the metrics endpoint.")
metricsPort = flag.String("metrics-port", "10254", "Port for the metrics endpoint.")
metricsEndpoint = flag.String("metrics-endpoint", "/metrics", "Metrics endpoint.")
metricsPrefix = flag.String("metrics-prefix", "", "Prefix for the metrics.")
ingressURLFormat = flag.String("ingress-url-format", "", "Ingress URL format.")
enableLeaderElection = flag.Bool("leader-election", false, "Enable Spark operator leader election.")
leaderElectionLockNamespace = flag.String("leader-election-lock-namespace", "spark-operator", "Namespace in which to create the ConfigMap for leader election.")
leaderElectionLockName = flag.String("leader-election-lock-name", "spark-operator-lock", "Name of the ConfigMap for leader election.")
leaderElectionLeaseDuration = flag.Duration("leader-election-lease-duration", 15*time.Second, "Leader election lease duration.")
leaderElectionRenewDeadline = flag.Duration("leader-election-renew-deadline", 14*time.Second, "Leader election renew deadline.")
leaderElectionRetryPeriod = flag.Duration("leader-election-retry-period", 4*time.Second, "Leader election retry period.")
enableBatchScheduler = flag.Bool("enable-batch-scheduler", false,
fmt.Sprintf("Enable batch schedulers for pods' scheduling, the available batch schedulers are: (%s).", strings.Join(batchscheduler.GetRegisteredNames(), ",")))
)
func main() {
var metricsLabels util.ArrayFlags
flag.Var(&metricsLabels, "metrics-labels", "Labels for the metrics")
flag.Parse()
// Create the client config. Use kubeConfig if given, otherwise assume in-cluster.
config, err := buildConfig(*master, *kubeConfig)
if err != nil {
glog.Fatal(err)
}
kubeClient, err := clientset.NewForConfig(config)
if err != nil {
glog.Fatal(err)
}
signalCh := make(chan os.Signal, 1)
signal.Notify(signalCh, syscall.SIGINT, syscall.SIGTERM)
stopCh := make(chan struct{}, 1)
startCh := make(chan struct{}, 1)
if *enableLeaderElection {
hostname, err := os.Hostname()
if err != nil {
glog.Fatal(err)
}
resourceLock, err := resourcelock.New(resourcelock.ConfigMapsResourceLock, *leaderElectionLockNamespace, *leaderElectionLockName, kubeClient.CoreV1(), resourcelock.ResourceLockConfig{
Identity: hostname,
// TODO: This is a workaround for a nil dereference in client-go. This line can be removed when that dependency is updated.
EventRecorder: &record.FakeRecorder{},
})
if err != nil {
glog.Fatal(err)
}
electionCfg := leaderelection.LeaderElectionConfig{
Lock: resourceLock,
LeaseDuration: *leaderElectionLeaseDuration,
RenewDeadline: *leaderElectionRenewDeadline,
RetryPeriod: *leaderElectionRetryPeriod,
Callbacks: leaderelection.LeaderCallbacks{
OnStartedLeading: func(c context.Context) {
close(startCh)
},
OnStoppedLeading: func() {
close(stopCh)
},
},
}
elector, err := leaderelection.NewLeaderElector(electionCfg)
if err != nil {
glog.Fatal(err)
}
go elector.Run(context.Background())
}
glog.Info("Starting the Spark Operator")
crClient, err := crclientset.NewForConfig(config)
if err != nil {
glog.Fatal(err)
}
apiExtensionsClient, err := apiextensionsclient.NewForConfig(config)
if err != nil {
glog.Fatal(err)
}
var batchSchedulerMgr *batchscheduler.SchedulerManager
if *enableBatchScheduler {
if !*enableWebhook {
glog.Fatal(
"failed to initialize the batch scheduler manager as it requires the webhook to be enabled")
}
batchSchedulerMgr = batchscheduler.NewSchedulerManager(config)
}
crInformerFactory := buildCustomResourceInformerFactory(crClient)
podInformerFactory := buildPodInformerFactory(kubeClient)
var metricConfig *util.MetricConfig
if *enableMetrics {
metricConfig = &util.MetricConfig{
MetricsEndpoint: *metricsEndpoint,
MetricsPort: *metricsPort,
MetricsPrefix: *metricsPrefix,
MetricsLabels: metricsLabels,
}
glog.Info("Enabling metrics collecting and exporting to Prometheus")
util.InitializeMetrics(metricConfig)
}
applicationController := sparkapplication.NewController(
crClient, kubeClient, crInformerFactory, podInformerFactory, metricConfig, *namespace, *ingressURLFormat, batchSchedulerMgr)
scheduledApplicationController := scheduledsparkapplication.NewController(
crClient, kubeClient, apiExtensionsClient, crInformerFactory, clock.RealClock{})
// Start the informer factory that in turn starts the informer.
go crInformerFactory.Start(stopCh)
go podInformerFactory.Start(stopCh)
var hook *webhook.WebHook
if *enableWebhook {
var coreV1InformerFactory informers.SharedInformerFactory
if *enableResourceQuotaEnforcement {
coreV1InformerFactory = buildCoreV1InformerFactory(kubeClient)
}
var err error
// Don't deregister webhook on exit if leader election enabled (i.e. multiple webhooks running)
hook, err = webhook.New(kubeClient, crInformerFactory, *namespace, !*enableLeaderElection, *enableResourceQuotaEnforcement, coreV1InformerFactory)
if err != nil {
glog.Fatal(err)
}
if *enableResourceQuotaEnforcement {
go coreV1InformerFactory.Start(stopCh)
}
if err = hook.Start(stopCh); err != nil {
glog.Fatal(err)
}
} else if *enableResourceQuotaEnforcement {
glog.Fatal("Webhook must be enabled to use resource quota enforcement.")
}
if *enableLeaderElection {
glog.Info("Waiting to be elected leader before starting application controller goroutines")
<-startCh
}
glog.Info("Starting application controller goroutines")
if err = applicationController.Start(*controllerThreads, stopCh); err != nil {
glog.Fatal(err)
}
if err = scheduledApplicationController.Start(*controllerThreads, stopCh); err != nil {
glog.Fatal(err)
}
select {
case <-signalCh:
close(stopCh)
case <-stopCh:
}
glog.Info("Shutting down the Spark Operator")
applicationController.Stop()
scheduledApplicationController.Stop()
if *enableWebhook {
if err := hook.Stop(); err != nil {
glog.Fatal(err)
}
}
}
func buildConfig(masterURL string, kubeConfig string) (*rest.Config, error) {
if kubeConfig != "" {
return clientcmd.BuildConfigFromFlags(masterURL, kubeConfig)
}
return rest.InClusterConfig()
}
func buildCustomResourceInformerFactory(crClient crclientset.Interface) crinformers.SharedInformerFactory {
var factoryOpts []crinformers.SharedInformerOption
if *namespace != apiv1.NamespaceAll {
factoryOpts = append(factoryOpts, crinformers.WithNamespace(*namespace))
}
return crinformers.NewSharedInformerFactoryWithOptions(
crClient,
// resyncPeriod. Every resyncPeriod, all resources in the cache will re-trigger events.
time.Duration(*resyncInterval)*time.Second,
factoryOpts...)
}
func buildPodInformerFactory(kubeClient clientset.Interface) informers.SharedInformerFactory {
var podFactoryOpts []informers.SharedInformerOption
if *namespace != apiv1.NamespaceAll {
podFactoryOpts = append(podFactoryOpts, informers.WithNamespace(*namespace))
}
tweakListOptionsFunc := func(options *metav1.ListOptions) {
options.LabelSelector = fmt.Sprintf("%s,%s", operatorConfig.SparkRoleLabel, operatorConfig.LaunchedBySparkOperatorLabel)
}
podFactoryOpts = append(podFactoryOpts, informers.WithTweakListOptions(tweakListOptionsFunc))
return informers.NewSharedInformerFactoryWithOptions(kubeClient, time.Duration(*resyncInterval)*time.Second, podFactoryOpts...)
}
func buildCoreV1InformerFactory(kubeClient clientset.Interface) informers.SharedInformerFactory {
var coreV1FactoryOpts []informers.SharedInformerOption
if *namespace != apiv1.NamespaceAll {
coreV1FactoryOpts = append(coreV1FactoryOpts, informers.WithNamespace(*namespace))
}
return informers.NewSharedInformerFactoryWithOptions(kubeClient, time.Duration(*resyncInterval)*time.Second, coreV1FactoryOpts...)
}