feat: add replay

Makefile

@@ -5,7 +5,7 @@ NAME = "github.com/odpf/optimus"
 LAST_COMMIT := $(shell git rev-parse --short HEAD)
 LAST_TAG := "$(shell git rev-list --tags --max-count=1)"
 OPMS_VERSION := "$(shell git describe --tags ${LAST_TAG})-next"
-PROTON_COMMIT := "fae8287656b163ae07a7f03edd3ea3f5df499dcb"
+PROTON_COMMIT := "31ac9046d1a8c95a2f4645b87bf0620a3e6bb8bc"
 
 
 .PHONY: build test test-ci generate-proto unit-test-ci integration-test vet coverage clean install lint
@@ -66,4 +66,4 @@ install: ## install required dependencies
 	go install github.com/bufbuild/buf/cmd/buf@v1.5.0
 	go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@v1.2.0
 	go install github.com/grpc-ecosystem/grpc-gateway/v2/protoc-gen-grpc-gateway@v2.5.0
-	go install github.com/grpc-ecosystem/grpc-gateway/v2/protoc-gen-openapiv2@v2.5.0
+	go install github.com/grpc-ecosystem/grpc-gateway/v2/protoc-gen-openapiv2@v2.5.0

client/cmd/commands.go

@@ -13,6 +13,7 @@ import (
 	"github.com/odpf/optimus/client/cmd/playground"
 	"github.com/odpf/optimus/client/cmd/plugin"
 	"github.com/odpf/optimus/client/cmd/project"
+	"github.com/odpf/optimus/client/cmd/replay"
 	"github.com/odpf/optimus/client/cmd/resource"
 	"github.com/odpf/optimus/client/cmd/scheduler"
 	"github.com/odpf/optimus/client/cmd/secret"
@@ -66,6 +67,7 @@ func New() *cli.Command {
 		version.NewVersionCommand(),
 		playground.NewPlaygroundCommand(),
 		scheduler.NewSchedulerCommand(),
+		replay.NewReplayCommand(),
 
 		// Will decide later, to add it server side or not
 		plugin.NewPluginCommand(),

client/cmd/replay/create.go

@@ -0,0 +1,159 @@
+package replay
+
+import (
+	"errors"
+	"fmt"
+	"time"
+
+	"github.com/odpf/salt/log"
+	"github.com/spf13/cobra"
+	"golang.org/x/net/context"
+	"google.golang.org/protobuf/types/known/timestamppb"
+
+	"github.com/odpf/optimus/client/cmd/internal"
+	"github.com/odpf/optimus/client/cmd/internal/connectivity"
+	"github.com/odpf/optimus/client/cmd/internal/logger"
+	"github.com/odpf/optimus/config"
+	pb "github.com/odpf/optimus/protos/odpf/optimus/core/v1beta1"
+)
+
+const (
+	replayTimeout = time.Minute * 1
+	ISOTimeLayout = time.RFC3339
+)
+
+type createCommand struct {
+	logger         log.Logger
+	configFilePath string
+
+	parallel    bool
+	description string
+	jobConfig   string
+
+	projectName   string
+	namespaceName string
+	host          string
+}
+
+// CreateCommand initializes command for creating a replay request
+func CreateCommand() *cobra.Command {
+	refresh := &createCommand{
+		logger: logger.NewClientLogger(),
+	}
+
+	cmd := &cobra.Command{
+		Use:   "create",
+		Short: "Run replay operation on a dag based on provided start and end time range",
+		Long: "This operation takes three arguments, first is DAG name[required]\nused in optimus specification, " +
+			"second is start time[required] of\nreplay, third is end time[optional] of replay. \nDate ranges are inclusive.",
+		Example: "optimus replay create <job_name> <2023-01-01T02:30:00Z00:00> [2023-01-02T02:30:00Z00:00]",
+		Args: func(cmd *cobra.Command, args []string) error {
+			if len(args) < 1 {
+				return errors.New("job name is required")
+			}
+			if len(args) < 2 { //nolint: gomnd
+				return errors.New("replay start time is required")
+			}
+			return nil
+		},
+		RunE:    refresh.RunE,
+		PreRunE: refresh.PreRunE,
+	}
+
+	refresh.injectFlags(cmd)
+	return cmd
+}
+
+func (r *createCommand) injectFlags(cmd *cobra.Command) {
+	// Config filepath flag
+	cmd.Flags().StringVarP(&r.configFilePath, "config", "c", config.EmptyPath, "File path for client configuration")
+	cmd.Flags().StringVarP(&r.namespaceName, "namespace-name", "n", "", "Name of the optimus namespace")
+
+	cmd.Flags().BoolVarP(&r.parallel, "parallel", "", false, "Backfill job runs in parallel")
+	cmd.Flags().StringVarP(&r.description, "description", "d", "", "Description of why backfill is needed")
+	cmd.Flags().StringVarP(&r.jobConfig, "job-config", "", "", "additional job configurations")
+
+	// Mandatory flags if config is not set
+	cmd.Flags().StringVarP(&r.projectName, "project-name", "p", "", "Name of the optimus project")
+	cmd.Flags().StringVar(&r.host, "host", "", "Optimus service endpoint url")
+}
+
+func (r *createCommand) PreRunE(cmd *cobra.Command, _ []string) error {
+	conf, err := internal.LoadOptionalConfig(r.configFilePath)
+	if err != nil {
+		return err
+	}
+
+	if conf == nil {
+		internal.MarkFlagsRequired(cmd, []string{"project-name", "host"})
+		return nil
+	}
+
+	if r.projectName == "" {
+		r.projectName = conf.Project.Name
+	}
+	if r.host == "" {
+		r.host = conf.Host
+	}
+	return nil
+}
+
+func (r *createCommand) RunE(_ *cobra.Command, args []string) error {
+	jobName := args[0]
+	startTime := args[1]
+	endTime := args[1]
+	if len(args) >= 3 { //nolint: gomnd
+		endTime = args[2]
+	}
+
+	replayID, err := r.createReplayRequest(jobName, startTime, endTime, r.jobConfig)
+	if err != nil {
+		return err
+	}
+	r.logger.Info("Replay request created with id %s", replayID)
+	return nil
+}
+
+func (r *createCommand) createReplayRequest(jobName, startTimeStr, endTimeStr, jobConfig string) (string, error) {
+	conn, err := connectivity.NewConnectivity(r.host, replayTimeout)
+	if err != nil {
+		return "", err
+	}
+	defer conn.Close()
+
+	replayService := pb.NewReplayServiceClient(conn.GetConnection())
+
+	startTime, err := getTimeProto(startTimeStr)
+	if err != nil {
+		return "", err
+	}
+	endTime, err := getTimeProto(endTimeStr)
+	if err != nil {
+		return "", err
+	}
+	respStream, err := replayService.Replay(conn.GetContext(), &pb.ReplayRequest{
+		ProjectName:   r.projectName,
+		JobName:       jobName,
+		NamespaceName: r.namespaceName,
+		StartTime:     startTime,
+		EndTime:       endTime,
+		Parallel:      r.parallel,
+		Description:   r.description,
+		JobConfig:     jobConfig,
+	})
+	if err != nil {
+		if errors.Is(err, context.DeadlineExceeded) {
+			r.logger.Error("Replay creation took too long, timing out")
+		}
+		return "", fmt.Errorf("replay request failed: %w", err)
+	}
+	return respStream.Id, nil
+}
+
+func getTimeProto(timeStr string) (*timestamppb.Timestamp, error) {
+	parsedTime, err := time.Parse(ISOTimeLayout, timeStr)
+	if err != nil {
+		return nil, err
+	}
+	return timestamppb.New(parsedTime), nil
+}

client/cmd/replay/replay.go

@@ -0,0 +1,21 @@
+package replay
+
+import (
+	"github.com/spf13/cobra"
+)
+
+// NewReplayCommand initializes command for replay
+func NewReplayCommand() *cobra.Command {
+	cmd := &cobra.Command{
+		Use:   "replay",
+		Short: "replay related functions",
+		Annotations: map[string]string{
+			"group:core": "false",
+		},
+	}
+
+	cmd.AddCommand(
+		CreateCommand(),
+	)
+	return cmd
+}

config/config_server.go

@@ -1,5 +1,7 @@
 package config
 
+import "time"
+
 type ServerConfig struct {
 	Version          Version           `mapstructure:"version"`
 	Log              LogConfig         `mapstructure:"log"`
@@ -8,6 +10,7 @@ type ServerConfig struct {
 	Telemetry        TelemetryConfig   `mapstructure:"telemetry"`
 	ResourceManagers []ResourceManager `mapstructure:"resource_managers"`
 	Plugin           PluginConfig      `mapstructure:"plugin"`
+	Replay           ReplayConfig      `mapstructure:"replay"`
 }
 
 type Serve struct {
@@ -47,3 +50,8 @@ type ResourceManagerConfigOptimus struct {
 type PluginConfig struct {
 	Artifacts []string `mapstructure:"artifacts"`
 }
+
+// TODO: add worker interval
+type ReplayConfig struct {
+	ReplayTimeout time.Duration `mapstructure:"replay_timeout" default:"3h"`
+}

config/loader_test.go

@@ -7,6 +7,7 @@ import (
 	"path/filepath"
 	"strings"
 	"testing"
+	"time"
 
 	saltConfig "github.com/odpf/salt/config"
 	"github.com/spf13/afero"
@@ -277,6 +278,8 @@ func (s *ConfigTestSuite) initExpectedServerConfig() {
 		},
 	}
 	s.expectedServerConfig.Plugin = config.PluginConfig{}
+
+	s.expectedServerConfig.Replay.ReplayTimeout = time.Hour * 3
 }
 
 func (*ConfigTestSuite) initServerConfigEnv() {

core/job/handler/v1beta1/job.go

@@ -240,7 +240,7 @@ func (jh *JobHandler) ListJobSpecification(ctx context.Context, req *pb.ListJobS
 
 func (*JobHandler) GetWindow(_ context.Context, req *pb.GetWindowRequest) (*pb.GetWindowResponse, error) {
 	// TODO: the default version to be deprecated & made mandatory in future releases
-	version := 1
+	version := 2
 	if err := req.GetScheduledAt().CheckValid(); err != nil {
 		return nil, fmt.Errorf("%w: failed to parse schedule time %s", err, req.GetScheduledAt())
 	}

core/scheduler/handler/v1beta1/replay.go

@@ -0,0 +1,83 @@
+package v1beta1
+
+import (
+	"fmt"
+	"strings"
+
+	"github.com/google/uuid"
+	"github.com/odpf/salt/log"
+	"golang.org/x/net/context"
+
+	"github.com/odpf/optimus/core/scheduler"
+	"github.com/odpf/optimus/core/tenant"
+	"github.com/odpf/optimus/internal/errors"
+	pb "github.com/odpf/optimus/protos/odpf/optimus/core/v1beta1"
+)
+
+type ReplayService interface {
+	CreateReplay(ctx context.Context, tenant tenant.Tenant, jobName scheduler.JobName, config *scheduler.ReplayConfig) (replayID uuid.UUID, err error)
+}
+
+type ReplayHandler struct {
+	l       log.Logger
+	service ReplayService
+
+	pb.UnimplementedReplayServiceServer
+}
+
+func (h ReplayHandler) Replay(ctx context.Context, req *pb.ReplayRequest) (*pb.ReplayResponse, error) {
+	replayTenant, err := tenant.NewTenant(req.GetProjectName(), req.NamespaceName)
+	if err != nil {
+		return nil, errors.GRPCErr(err, "unable to start replay for "+req.GetJobName())
+	}
+
+	jobName, err := scheduler.JobNameFrom(req.GetJobName())
+	if err != nil {
+		return nil, errors.GRPCErr(err, "unable to start replay for "+req.GetJobName())
+	}
+
+	if err = req.GetStartTime().CheckValid(); err != nil {
+		return nil, errors.GRPCErr(errors.InvalidArgument(scheduler.EntityJobRun, "invalid start_time"), "unable to start replay for "+req.GetJobName())
+	}
+
+	if req.GetEndTime() != nil {
+		if err = req.GetEndTime().CheckValid(); err != nil {
+			return nil, errors.GRPCErr(errors.InvalidArgument(scheduler.EntityJobRun, "invalid end_time"), "unable to start replay for "+req.GetJobName())
+		}
+	}
+
+	jobConfig := make(map[string]string)
+	if req.JobConfig != "" {
+		jobConfig, err = parseJobConfig(req.JobConfig)
+		if err != nil {
+			return nil, errors.GRPCErr(err, "unable to parse replay job config for "+req.JobName)
+		}
+	}
+
+	replayConfig := scheduler.NewReplayConfig(req.GetStartTime().AsTime(), req.GetEndTime().AsTime(), req.Parallel, jobConfig, req.Description)
+	replayID, err := h.service.CreateReplay(ctx, replayTenant, jobName, replayConfig)
+	if err != nil {
+		return nil, errors.GRPCErr(err, "unable to start replay for "+req.GetJobName())
+	}
+
+	return &pb.ReplayResponse{Id: replayID.String()}, nil
+}
+
+func parseJobConfig(jobConfig string) (map[string]string, error) {
+	configs := map[string]string{}
+	for _, config := range strings.Split(jobConfig, ",") {
+		keyValue := strings.Split(config, "=")
+		valueLen := 2
+		if len(keyValue) != valueLen {
+			return nil, fmt.Errorf("error on job config value, %s", config)
+		}
+		key := strings.TrimSpace(strings.ToUpper(keyValue[0]))
+		value := keyValue[1]
+		configs[key] = value
+	}
+	return configs, nil
+}
+
+func NewReplayHandler(l log.Logger, service ReplayService) *ReplayHandler {
+	return &ReplayHandler{l: l, service: service}
+}