[core] Make new ddsched timeout only apply to GetData status calls

Author: teo

core/config.go

@@ -102,7 +102,7 @@ func setDefaults() error {
 	viper.SetDefault("ccdbEndpoint", "http://ccdb-test.cern.ch:8080")
 	viper.SetDefault("dcsServiceEndpoint", "//127.0.0.1:50051")
 	viper.SetDefault("dcsServiceUseSystemProxy", false)
-	viper.SetDefault("ddSchedulergRPCTimeout", "30s")
+	viper.SetDefault("ddSchedulerStatusTimeout", "30s")
 	viper.SetDefault("ddSchedulerEndpoint", "//127.0.0.1:50052")
 	viper.SetDefault("ddSchedulerUseSystemProxy", false)
 	viper.SetDefault("trgServiceEndpoint", "//127.0.0.1:50060")
@@ -171,7 +171,7 @@ func setFlags() error {
 	pflag.String("dcsServiceEndpoint", viper.GetString("dcsServiceEndpoint"), "Endpoint of the DCS gRPC service (`host:port`)")
 	pflag.Bool("dcsServiceUseSystemProxy", viper.GetBool("dcsServiceUseSystemProxy"), "When true the https_proxy, http_proxy and no_proxy environment variables are obeyed")
 	pflag.String("ddSchedulerEndpoint", viper.GetString("ddSchedulerEndpoint"), "Endpoint of the DD scheduler gRPC service (`host:port`)")
-	pflag.Duration("ddSchedulergRPCTimeout", viper.GetDuration("ddSchedulergRPCTimeout"), "Timeout for gRPC calls in ddshed plugin")
+	pflag.Duration("ddSchedulerStatusTimeout", viper.GetDuration("ddSchedulerStatusTimeout"), "Timeout for status calls in ddshed plugin")
 	pflag.Bool("ddSchedulerUseSystemProxy", viper.GetBool("ddSchedulerUseSystemProxy"), "When true the https_proxy, http_proxy and no_proxy environment variables are obeyed")
 	pflag.String("trgServiceEndpoint", viper.GetString("trgServiceEndpoint"), "Endpoint of the TRG gRPC service (`host:port`)")
 	pflag.String("trgPollingInterval", viper.GetString("trgPollingInterval"), "How often to query the TRG gRPC service for run status (default: 3s)")

core/integration/ddsched/client.go

@@ -29,7 +29,6 @@ import (
 	"time"
 
 	"github.com/AliceO2Group/Control/common/logger"
-	"github.com/AliceO2Group/Control/core/integration"
 	ddpb "github.com/AliceO2Group/Control/core/integration/ddsched/protos"
 	"github.com/sirupsen/logrus"
 	"github.com/spf13/viper"
@@ -68,7 +67,6 @@ func NewClient(cxt context.Context, cancel context.CancelFunc, endpoint string)
 			Timeout:             time.Second,
 			PermitWithoutStream: true,
 		}),
-		grpc.WithUnaryInterceptor(integration.UnaryTimeoutInterceptor(viper.GetDuration("ddSchedulergRPCTimeout"), "ddsched gRPC call failed")),
 	}
 	if !viper.GetBool("ddSchedulerUseSystemProxy") {
 		dialOptions = append(dialOptions, grpc.WithNoProxy())

core/integration/ddsched/plugin.go

@@ -52,11 +52,10 @@ import (
 )
 
 const (
-	DDSCHED_DIAL_TIMEOUT            = 2 * time.Second
-	DDSCHED_INITIALIZE_TIMEOUT      = 30 * time.Second
-	DDSCHED_TERMINATE_TIMEOUT       = 30 * time.Second
-	DDSCHED_DEFAULT_POLLING_TIMEOUT = 30 * time.Second
-	TOPIC                           = topic.IntegratedService + topic.Separator + "ddsched"
+	DDSCHED_DIAL_TIMEOUT       = 2 * time.Second
+	DDSCHED_INITIALIZE_TIMEOUT = 30 * time.Second
+	DDSCHED_TERMINATE_TIMEOUT  = 30 * time.Second
+	TOPIC                      = topic.IntegratedService + topic.Separator + "ddsched"
 )
 
 type Plugin struct {
@@ -161,7 +160,9 @@ func (p *Plugin) partitionStatesForEnvs(envIds []uid.ID) map[uid.ID]map[string]s
 			PartitionId:   envId.String(),
 			EnvironmentId: envId.String(),
 		}
-		state, err := p.ddSchedClient.PartitionStatus(context.Background(), &in, grpc.EmptyCallOption{})
+		ctx, cancel := context.WithTimeout(context.Background(), viper.GetDuration("ddschedStatusTimeout"))
+		state, err := p.ddSchedClient.PartitionStatus(ctx, &in, grpc.EmptyCallOption{})
+		cancel()
 		if err != nil {
 			continue
 		}