feat(dbha-v2): add daemon-start guard for auto-restart on crash. issu…

dbm-services/common/dbha-v2/Makefile

@@ -50,15 +50,17 @@ PROTO_FILES = $(wildcard $(PROTO_DIR)/*.proto)
 # generate go code files
 GO_GEN_FILES=$(PROTO_FILES:$(PROTO_DIR)/%.proto=$(GEN_DIR)/%.pb.go)
 
-# services list
-SERVICES := admin analysis receiver probe
+# services list and binary names with dbha- prefix
+SERVICES   := admin analysis receiver probe
+BIN_PREFIX := dbha-
+BINARIES   := $(addprefix $(BIN_PREFIX),$(SERVICES))
 
 # define common go build command
 define GO_BUILD
 	@mkdir -p $(BUILD_DIR)
 	CGO_ENABLED=0 GOOS=$(GO_OS) GOARCH=amd64 go build -ldflags=$(BUILD_FLAG) \
 				-gcflags="all=-trimpath=$(PWD)" -asmflags="all=-trimpath=$(PWD)" \
-				-o $(BUILD_DIR)/$(1) cmd/$(1)/main.go
+				-o $(BUILD_DIR)/$(BIN_PREFIX)$(1) cmd/$(1)/main.go
 endef
 
 .PHONY: all proto $(SERVICES) clean format test package help
@@ -100,7 +102,7 @@ toolkits: cluster
 cluster:
 	@mkdir -p $(BUILD_DIR)
 	CGO_ENABLED=0 GOOS=${GO_OS} GOARCH=amd64 go build -ldflags=$(BUILD_FLAG) -gcflags="all=-trimpath=$(PWD)" \
-				-asmflags="all=-trimpath=$(PWD)" -o $(BUILD_DIR)/$@ tools/cmd/cluster.go
+				-asmflags="all=-trimpath=$(PWD)" -o $(BUILD_DIR)/$(BIN_PREFIX)$@ tools/cmd/cluster.go
 
 # build protobuf to go
 $(GEN_DIR)/%.pb.go: $(PROTO_DIR)/%.proto
@@ -119,8 +121,8 @@ package: $(SERVICES) toolkits
 	@echo "Packaging $(PKG_NAME)..."
 	@rm -rf $(PKG_DIR)
 	@mkdir -p $(PKG_DIR)/etc $(PKG_DIR)/logs $(PKG_DIR)/pids $(PKG_DIR)/toolkits
-	@cp $(addprefix $(BUILD_DIR)/,$(SERVICES)) $(PKG_DIR)/
-	@cp $(BUILD_DIR)/cluster $(PKG_DIR)/toolkits/
+	@cp $(addprefix $(BUILD_DIR)/,$(BINARIES)) $(PKG_DIR)/
+	@cp $(BUILD_DIR)/$(BIN_PREFIX)cluster $(PKG_DIR)/toolkits/
 	@-cp etc/*.yaml $(PKG_DIR)/etc/ 2>/dev/null || true
 	@cd $(BUILD_DIR) && tar -czvf $(PKG_NAME) $(PROJECT)
 	@rm -rf $(PKG_DIR)

dbm-services/common/dbha-v2/cmd/admin/main.go

@@ -46,6 +46,7 @@ func main() {
 	rootCmd.AddCommand(admin.MigrateCmd)
 	rootCmd.AddCommand(admin.HealthCmd)
 	rootCmd.AddCommand(admin.StartCmd)
+	rootCmd.AddCommand(admin.DaemonStartCmd)
 	rootCmd.AddCommand(admin.StopCmd)
 	rootCmd.AddCommand(admin.RestartCmd)
 	rootCmd.AddCommand(admin.ReloadCmd)

dbm-services/common/dbha-v2/cmd/analysis/main.go

@@ -45,6 +45,7 @@ func main() {
 	rootCmd.AddCommand(analysis.VersionCmd)
 	rootCmd.AddCommand(analysis.HealthCmd)
 	rootCmd.AddCommand(analysis.StartCmd)
+	rootCmd.AddCommand(analysis.DaemonStartCmd)
 	rootCmd.AddCommand(analysis.StopCmd)
 	rootCmd.AddCommand(analysis.RestartCmd)
 	rootCmd.AddCommand(analysis.ReloadCmd)

dbm-services/common/dbha-v2/cmd/probe/main.go

@@ -46,6 +46,7 @@ func main() {
 	rootCmd.AddCommand(probe.VersionCmd)
 	rootCmd.AddCommand(probe.HealthCmd)
 	rootCmd.AddCommand(probe.StartCmd)
+	rootCmd.AddCommand(probe.DaemonStartCmd)
 	rootCmd.AddCommand(probe.StopCmd)
 	rootCmd.AddCommand(probe.RestartCmd)
 	rootCmd.AddCommand(probe.ReloadCmd)

dbm-services/common/dbha-v2/cmd/receiver/main.go

@@ -44,6 +44,7 @@ func main() {
 	rootCmd.AddCommand(receiver.VersionCmd)
 	rootCmd.AddCommand(receiver.HealthCmd)
 	rootCmd.AddCommand(receiver.StartCmd)
+	rootCmd.AddCommand(receiver.DaemonStartCmd)
 	rootCmd.AddCommand(receiver.StopCmd)
 	rootCmd.AddCommand(receiver.RestartCmd)
 	rootCmd.AddCommand(receiver.ReloadCmd)

dbm-services/common/dbha-v2/internal/admin/cmds/cmds.go

@@ -56,6 +56,14 @@ func ReloadCmdRunE(cmd *cobra.Command, args []string) error {
 	return process.ReloadCmdRunE(cmd, args, config.Cfg.PidFile, process.NameAdmin, StopTimeout, ForceStop)
 }
 
+func DaemonStartCmdRunE(cmd *cobra.Command, args []string) error {
+	configPath, _ := cmd.Root().PersistentFlags().GetString("config")
+	if err := config.Load(configPath); err != nil {
+		return err
+	}
+	return process.DaemonStartCmdRunE(cmd, args, config.Cfg.PidFile, process.NameAdmin, process.DefaultGuardRestartDelay)
+}
+
 func HealthCmdRunE(cmd *cobra.Command, _ []string) error {
 	baseHealth := process.GetBaseHealthInfo(config.Cfg.PidFile, process.NameAdmin)
 

dbm-services/common/dbha-v2/internal/admin/command.go

@@ -89,6 +89,13 @@ var StartCmd = &cobra.Command{
 	RunE:  cmds.StartCmdRunE,
 }
 
+// DaemonStartCmd is used to start this process with a guard that restarts it on abnormal exit.
+var DaemonStartCmd = &cobra.Command{
+	Use:   "daemon-start",
+	Short: "Start this process with guard (auto-restart on crash).",
+	RunE:  cmds.DaemonStartCmdRunE,
+}
+
 // StopCmd is used to stop this process.
 var StopCmd = &cobra.Command{
 	Use:   "stop",

dbm-services/common/dbha-v2/internal/admin/config/config.go

@@ -29,6 +29,8 @@ import (
 	"time"
 
 	"dbm-services/common/dbha-v2/pkg/logger"
+
+	"github.com/spf13/viper"
 )
 
 var Cfg = Configuration{
@@ -107,3 +109,20 @@ type Configuration struct {
 	Storage    StorageConfig   `yaml:"storage"    mapstructure:"storage"`
 	Log        LogConfig       `yaml:"log"        mapstructure:"log"`
 }
+
+// Load loads admin configuration from file
+func Load(configFilePath string) error {
+	viper.SetConfigName("admin")
+	viper.SetConfigType("yaml")
+	viper.AddConfigPath("./etc")
+
+	if configFilePath != "" {
+		viper.SetConfigFile(configFilePath)
+	}
+
+	if err := viper.ReadInConfig(); err != nil {
+		return err
+	}
+
+	return viper.Unmarshal(&Cfg)
+}

dbm-services/common/dbha-v2/internal/analysis/cmds/cmds.go

@@ -56,6 +56,14 @@ func ReloadCmdRunE(cmd *cobra.Command, args []string) error {
 	return process.ReloadCmdRunE(cmd, args, config.Cfg.PidFile, process.NameAnalysis, StopTimeout, ForceStop)
 }
 
+func DaemonStartCmdRunE(cmd *cobra.Command, args []string) error {
+	configPath, _ := cmd.Root().PersistentFlags().GetString("config")
+	if err := config.Load(configPath); err != nil {
+		return err
+	}
+	return process.DaemonStartCmdRunE(cmd, args, config.Cfg.PidFile, process.NameAnalysis, process.DefaultGuardRestartDelay)
+}
+
 func HealthCmdRunE(cmd *cobra.Command, _ []string) error {
 	baseHealth := process.GetBaseHealthInfo(config.Cfg.PidFile, process.NameAnalysis)
 

dbm-services/common/dbha-v2/internal/analysis/command.go

@@ -53,6 +53,13 @@ var StartCmd = &cobra.Command{
 	RunE:  cmds.StartCmdRunE,
 }
 
+// DaemonStartCmd is used to start this process with a guard that restarts it on abnormal exit.
+var DaemonStartCmd = &cobra.Command{
+	Use:   "daemon-start",
+	Short: "Start this process with guard (auto-restart on crash).",
+	RunE:  cmds.DaemonStartCmdRunE,
+}
+
 // StopCmd is used to stop this process.
 var StopCmd = &cobra.Command{
 	Use:   "stop",

dbm-services/common/dbha-v2/internal/analysis/config/config.go

@@ -29,6 +29,8 @@ import (
 	"time"
 
 	"dbm-services/common/dbha-v2/pkg/logger"
+
+	"github.com/spf13/viper"
 )
 
 var Cfg = Configuration{
@@ -162,6 +164,23 @@ type Configuration struct {
 	Log       LogConfig       `yaml:"log"       mapstructure:"log"`
 }
 
+// Load loads analysis configuration from file
+func Load(configFilePath string) error {
+	viper.SetConfigName("analysis")
+	viper.SetConfigType("yaml")
+	viper.AddConfigPath("./etc")
+
+	if configFilePath != "" {
+		viper.SetConfigFile(configFilePath)
+	}
+
+	if err := viper.ReadInConfig(); err != nil {
+		return err
+	}
+
+	return viper.Unmarshal(&Cfg)
+}
+
 func init() {
 	Cfg.Detector.Ssh.Port = 22
 	Cfg.Detector.Ssh.User = "root"

dbm-services/common/dbha-v2/internal/analysis/detector/detector.go

@@ -22,6 +22,7 @@
  * SOFTWARE.
  */
 
+// Package detector provides probe health detection and target selection for DBHA analysis.
 package detector
 
 import (

dbm-services/common/dbha-v2/internal/analysis/storage/dbha_data.go

@@ -44,6 +44,7 @@ type DbhaData struct {
 	DB *hamysql.GormDB
 }
 
+// GetBizIDs returns all distinct business IDs from DBM metadata.
 func (ha *DbhaData) GetBizIDs() ([]int, error) {
 	bkBizIDs := []int{}
 
@@ -58,6 +59,7 @@ func (ha *DbhaData) GetBizIDs() ([]int, error) {
 	return bkBizIDs, nil
 }
 
+// ReadMetadataCacheWithBizID reads metadata cache in batches for the given bizID.
 func (ha *DbhaData) ReadMetadataCacheWithBizID(bizID int, batchCnt int,
 	offsetDuration time.Duration) (metaData []*hamodel.DbmMetadata, err error) {
 
@@ -164,6 +166,7 @@ func (ha *DbhaData) SaveSwitchingLog(ctx context.Context, records ...*hamodel.Db
 	return err
 }
 
+// ReadSwitchingStrategyWithBkBizId returns switching strategies for the given business ID.
 func (ha *DbhaData) ReadSwitchingStrategyWithBkBizId(bkBizId int) ([]*hamodel.DbSwitchingStrategy, error) {
 	var strategies []*hamodel.DbSwitchingStrategy
 

dbm-services/common/dbha-v2/internal/probe/cmds/cmds.go

@@ -74,6 +74,14 @@ func ReloadCmdRunE(cmd *cobra.Command, args []string) error {
 	return process.ReloadCmdRunE(cmd, args, config.Cfg.PidFile, process.NameProbe, StopTimeout, ForceStop)
 }
 
+func DaemonStartCmdRunE(cmd *cobra.Command, args []string) error {
+	configPath, _ := cmd.Root().PersistentFlags().GetString("config")
+	if err := config.Load(configPath); err != nil {
+		return err
+	}
+	return process.DaemonStartCmdRunE(cmd, args, config.Cfg.PidFile, process.NameProbe, process.DefaultGuardRestartDelay)
+}
+
 func HealthCmdRunE(cmd *cobra.Command, _ []string) error {
 	if err := config.Load(ConfigFilePath); err != nil {
 		baseHealth := process.GetBaseHealthInfo(config.Cfg.PidFile, process.NameProbe)

dbm-services/common/dbha-v2/internal/probe/command.go

@@ -54,6 +54,13 @@ var StartCmd = &cobra.Command{
 	RunE:  cmds.StartCmdRunE,
 }
 
+// DaemonStartCmd is used to start this process with a guard that restarts it on abnormal exit.
+var DaemonStartCmd = &cobra.Command{
+	Use:   "daemon-start",
+	Short: "Start this process with guard (auto-restart on crash).",
+	RunE:  cmds.DaemonStartCmdRunE,
+}
+
 // StopCmd is used to stop this process.
 var StopCmd = &cobra.Command{
 	Use:   "stop",

dbm-services/common/dbha-v2/internal/receiver/cmds/cmds.go

@@ -56,6 +56,14 @@ func ReloadCmdRunE(cmd *cobra.Command, args []string) error {
 	return process.ReloadCmdRunE(cmd, args, config.Cfg.PidFile, process.NameReceiver, StopTimeout, ForceStop)
 }
 
+func DaemonStartCmdRunE(cmd *cobra.Command, args []string) error {
+	configPath, _ := cmd.Root().PersistentFlags().GetString("config")
+	if err := config.Load(configPath); err != nil {
+		return err
+	}
+	return process.DaemonStartCmdRunE(cmd, args, config.Cfg.PidFile, process.NameReceiver, process.DefaultGuardRestartDelay)
+}
+
 func HealthCmdRunE(cmd *cobra.Command, _ []string) error {
 	baseHealth := process.GetBaseHealthInfo(config.Cfg.PidFile, process.NameReceiver)
 

dbm-services/common/dbha-v2/internal/receiver/command.go

@@ -53,6 +53,13 @@ var StartCmd = &cobra.Command{
 	RunE:  cmds.StartCmdRunE,
 }
 
+// DaemonStartCmd is used to start this process with a guard that restarts it on abnormal exit.
+var DaemonStartCmd = &cobra.Command{
+	Use:   "daemon-start",
+	Short: "Start this process with guard (auto-restart on crash).",
+	RunE:  cmds.DaemonStartCmdRunE,
+}
+
 // StopCmd is used to stop this process.
 var StopCmd = &cobra.Command{
 	Use:   "stop",

dbm-services/common/dbha-v2/internal/receiver/config/config.go

@@ -28,6 +28,8 @@ import (
 	"time"
 
 	"dbm-services/common/dbha-v2/pkg/logger"
+
+	"github.com/spf13/viper"
 )
 
 var Cfg = Configuration{
@@ -102,3 +104,20 @@ type Configuration struct {
 	Service   ServiceConfig   `yaml:"service"   mapstructure:"service"`
 	Log       LogConfig       `yaml:"log"       mapstructure:"log"`
 }
+
+// Load loads receiver configuration from file
+func Load(configFilePath string) error {
+	viper.SetConfigName("receiver")
+	viper.SetConfigType("yaml")
+	viper.AddConfigPath("./etc")
+
+	if configFilePath != "" {
+		viper.SetConfigFile(configFilePath)
+	}
+
+	if err := viper.ReadInConfig(); err != nil {
+		return err
+	}
+
+	return viper.Unmarshal(&Cfg)
+}