Merge pull request volcano-sh#4661 from mvinchoo/mvinchoo/config

Feature: Option to disable default config fallback

Author: volcano-sh-bot

cmd/scheduler/app/options/options.go

@@ -93,6 +93,10 @@ type ServerOption struct {
 	// not be counted in pod pvc resource request and node.Allocatable, because the spec.drivers of csinode resource
 	// is always null, these provisioners usually are host path csi controllers like rancher.io/local-path and hostpath.csi.k8s.io.
 	IgnoredCSIProvisioners []string
+
+	// DisableDefaultSchedulerConfig indicates if the scheduler should fallback to default
+	// config if the current scheduler config is invalid
+	DisableDefaultSchedulerConfig bool
 }
 
 // DecryptFunc is custom function to parse ca file
@@ -151,6 +155,7 @@ func (s *ServerOption) AddFlags(fs *pflag.FlagSet) {
 	fs.StringVar(&s.CacheDumpFileDir, "cache-dump-dir", "/tmp", "The target dir where the json file put at when dump cache info to json file")
 	fs.Uint32Var(&s.NodeWorkerThreads, "node-worker-threads", defaultNodeWorkers, "The number of threads syncing node operations.")
 	fs.StringSliceVar(&s.IgnoredCSIProvisioners, "ignored-provisioners", nil, "The provisioners that will be ignored during pod pvc request computation and preemption.")
+	fs.BoolVar(&s.DisableDefaultSchedulerConfig, "disable-default-scheduler-config", false, "The flag indicates whether the scheduler should avoid using the default configuration if the provided scheduler configuration is invalid.")
 }
 
 // CheckOptionOrDie check leader election flag when LeaderElection is enabled.

cmd/scheduler/app/options/options_test.go

@@ -76,13 +76,14 @@ func TestAddFlags(t *testing.T) {
 			QPS:        defaultQPS,
 			Burst:      defaultBurst,
 		},
-		PluginsDir:                 defaultPluginsDir,
-		HealthzBindAddress:         ":11251",
-		MinNodesToFind:             defaultMinNodesToFind,
-		MinPercentageOfNodesToFind: defaultMinPercentageOfNodesToFind,
-		PercentageOfNodesToFind:    defaultPercentageOfNodesToFind,
-		NodeWorkerThreads:          defaultNodeWorkers,
-		CacheDumpFileDir:           "/tmp",
+		PluginsDir:                    defaultPluginsDir,
+		HealthzBindAddress:            ":11251",
+		MinNodesToFind:                defaultMinNodesToFind,
+		MinPercentageOfNodesToFind:    defaultMinPercentageOfNodesToFind,
+		PercentageOfNodesToFind:       defaultPercentageOfNodesToFind,
+		NodeWorkerThreads:             defaultNodeWorkers,
+		CacheDumpFileDir:              "/tmp",
+		DisableDefaultSchedulerConfig: false,
 	}
 	expectedFeatureGates := map[featuregate.Feature]bool{
 		features.PodDisruptionBudgetsSupport: false,

pkg/scheduler/scheduler.go

@@ -52,12 +52,13 @@ type Scheduler struct {
 	schedulePeriod time.Duration
 	once           sync.Once
 
-	mutex          sync.Mutex
-	actions        []framework.Action
-	plugins        []conf.Tier
-	configurations []conf.Configuration
-	metricsConf    map[string]string
-	dumper         schedcache.Dumper
+	mutex              sync.Mutex
+	actions            []framework.Action
+	plugins            []conf.Tier
+	configurations     []conf.Configuration
+	metricsConf        map[string]string
+	dumper             schedcache.Dumper
+	disableDefaultConf bool
 }
 
 // NewScheduler returns a Scheduler
@@ -74,11 +75,12 @@ func NewScheduler(config *rest.Config, opt *options.ServerOption) (*Scheduler, e
 
 	cache := schedcache.New(config, opt.SchedulerNames, opt.DefaultQueue, opt.NodeSelector, opt.NodeWorkerThreads, opt.IgnoredCSIProvisioners, opt.ResyncPeriod)
 	scheduler := &Scheduler{
-		schedulerConf:  opt.SchedulerConf,
-		fileWatcher:    watcher,
-		cache:          cache,
-		schedulePeriod: opt.SchedulePeriod,
-		dumper:         schedcache.Dumper{Cache: cache, RootDir: opt.CacheDumpFileDir},
+		schedulerConf:      opt.SchedulerConf,
+		fileWatcher:        watcher,
+		cache:              cache,
+		schedulePeriod:     opt.SchedulePeriod,
+		dumper:             schedcache.Dumper{Cache: cache, RootDir: opt.CacheDumpFileDir},
+		disableDefaultConf: opt.DisableDefaultSchedulerConfig,
 	}
 
 	return scheduler, nil
@@ -136,22 +138,30 @@ func (pc *Scheduler) loadSchedulerConf() {
 	klog.V(4).Infof("Start loadSchedulerConf ...")
 	defer func() {
 		actions, plugins := pc.getSchedulerConf()
-		klog.V(2).Infof("Successfully loaded Scheduler conf, actions: %v, plugins: %v", actions, plugins)
+		klog.V(2).Infof("Finished loading scheduler config. Final state: actions=%v, plugins=%v", actions, plugins)
 	}()
 
+	if pc.disableDefaultConf && len(pc.schedulerConf) == 0 {
+		klog.Fatalf("No --scheduler-conf path provided and default configuration fallback is disabled")
+	}
+
 	var err error
-	pc.once.Do(func() {
-		pc.actions, pc.plugins, pc.configurations, pc.metricsConf, err = UnmarshalSchedulerConf(DefaultSchedulerConf)
-		if err != nil {
-			klog.Errorf("unmarshal Scheduler config %s failed: %v", DefaultSchedulerConf, err)
-			panic("invalid default configuration")
-		}
-	})
+	if !pc.disableDefaultConf {
+		pc.once.Do(func() {
+			pc.actions, pc.plugins, pc.configurations, pc.metricsConf, err = UnmarshalSchedulerConf(DefaultSchedulerConf)
+			if err != nil {
+				klog.Fatalf("Invalid default configuration: unmarshal Scheduler config %s failed: %v", DefaultSchedulerConf, err)
+			}
+		})
+	}
 
 	var config string
 	if len(pc.schedulerConf) != 0 {
 		confData, err := os.ReadFile(pc.schedulerConf)
 		if err != nil {
+			if pc.disableDefaultConf {
+				klog.Fatalf("Failed to read scheduler config and default configuration fallback is disabled")
+			}
 			klog.Errorf("Failed to read the Scheduler config in '%s', using previous configuration: %v",
 				pc.schedulerConf, err)
 			return
@@ -161,6 +171,9 @@ func (pc *Scheduler) loadSchedulerConf() {
 
 	actions, plugins, configurations, metricsConf, err := UnmarshalSchedulerConf(config)
 	if err != nil {
+		if pc.disableDefaultConf {
+			klog.Fatalf("Invalid scheduler configuration and default configuration fallback is disabled")
+		}
 		klog.Errorf("Scheduler config %s is invalid: %v", config, err)
 		return
 	}

pkg/scheduler/util.go

@@ -28,7 +28,6 @@ import (
 	"strings"
 
 	"gopkg.in/yaml.v2"
-	"k8s.io/klog/v2"
 
 	"volcano.sh/volcano/pkg/scheduler/conf"
 	"volcano.sh/volcano/pkg/scheduler/framework"
@@ -86,7 +85,7 @@ func UnmarshalSchedulerConf(confStr string) ([]framework.Action, []conf.Tier, []
 		if action, found := framework.GetAction(strings.TrimSpace(actionName)); found {
 			actions = append(actions, action)
 		} else {
-			klog.Errorf("Failed to find Action %s, ignore it", actionName)
+			return nil, nil, nil, nil, fmt.Errorf("failed to find Action %s", actionName)
 		}
 	}