add parameter recover-cmd for process kill (#122)

* add parameter recover-cmd for process kill

Signed-off-by: FingerLeader <[email protected]>

* add parameter recover-cmd for process kill

Signed-off-by: FingerLeader <[email protected]>

* resolve confilct

Signed-off-by: FingerLeader <[email protected]>

* add test for recover-cmd

Signed-off-by: FingerLeader <[email protected]>

* fix some detail

Signed-off-by: FingerLeader <[email protected]>

* modify details

Signed-off-by: FingerLeader <[email protected]>

* modify details

Signed-off-by: FingerLeader <[email protected]>

* modify details

Signed-off-by: FingerLeader <[email protected]>

Author: FingerLeader

cmd/attack/process.go

@@ -62,6 +62,7 @@ func NewProcessKillCommand(dep fx.Option, options *core.ProcessCommand) *cobra.C
 
 	cmd.Flags().StringVarP(&options.Process, "process", "p", "", "The process name or the process ID")
 	cmd.Flags().IntVarP(&options.Signal, "signal", "s", 9, "The signal number to send")
+	cmd.Flags().StringVarP(&options.RecoverCmd, "recover-cmd", "r", "", "The command to be executed when recovering experiment")
 
 	return cmd
 }

pkg/core/process.go

@@ -30,9 +30,10 @@ type ProcessCommand struct {
 	CommonAttackConfig
 
 	// Process defines the process name or the process ID.
-	Process string `json:"process,omitempty"`
-	Signal  int    `json:"signal,omitempty"`
-	PIDs    []int
+	Process    string `json:"process,omitempty"`
+	Signal     int    `json:"signal,omitempty"`
+	PIDs       []int
+	RecoverCmd string `json:"recoverCmd,omitempty"`
 	// TODO: support these feature
 	// Newest       bool
 	// Oldest       bool

pkg/server/chaosd/process.go

@@ -15,11 +15,15 @@ package chaosd
 
 import (
 	"fmt"
+	"os/exec"
 	"strconv"
 	"syscall"
 
 	"github.com/mitchellh/go-ps"
 	"github.com/pingcap/errors"
+	"go.uber.org/zap"
+
+	"github.com/pingcap/log"
 
 	"github.com/chaos-mesh/chaosd/pkg/core"
 )
@@ -65,13 +69,23 @@ func (processAttack) Recover(exp core.Experiment, _ Environment) error {
 	}
 	pcmd := config.(*core.ProcessCommand)
 	if pcmd.Signal != int(syscall.SIGSTOP) {
-		return core.ErrNonRecoverableAttack.New("only SIGSTOP process attack is supported to recover")
-	}
+		if pcmd.RecoverCmd == "" {
+			return core.ErrNonRecoverableAttack.New("only SIGSTOP process attack and process attack with the recover-cmd are supported to recover")
+		}
 
-	for _, pid := range pcmd.PIDs {
-		if err := syscall.Kill(pid, syscall.SIGCONT); err != nil {
+		rcmd := exec.Command("bash", "-c", pcmd.RecoverCmd)
+		if err := rcmd.Start(); err != nil {
 			return errors.WithStack(err)
 		}
+
+		log.Info("Execute recover-cmd successfully", zap.String("recover-cmd", pcmd.RecoverCmd))
+
+	} else {
+		for _, pid := range pcmd.PIDs {
+			if err := syscall.Kill(pid, syscall.SIGCONT); err != nil {
+				return errors.WithStack(err)
+			}
+		}
 	}
 
 	return nil

test/integration_test/process/run.sh

@@ -58,23 +58,36 @@ if [[ ${stat} != "S" ]]; then
     exit 1
 fi
 
-# test kill attack
-${bin_path}/chaosd attack process kill -p ${pid} > proc.out
+# test kill attack and recover-cmd
+${bin_path}/chaosd attack process kill -p ${pid} -r "echo \"recover process\" >> proc.out" > proc.out
 
 sleep 1
 
+uid=$(cat proc.out | grep "Attack process ${pid} successfully" | awk -F: '{print $2}')
 stat=$(ps axo pid | grep ${pid})
 
+# test kill attack
 if [[ -n ${stat} ]]; then
     echo "target process is not killed by processed kill attack"
     rm dummy.out
     rm proc.out
+    kill ${pid}
+    exit 1
+fi
+
+# test recover-cmd
+${bin_path}/chaosd recover ${uid}
+recover_result=$(cat proc.out | awk 'END {print}')
+
+if [[ ${recover_result} != "recover process" ]]; then
+    echo "target recover process is not executed"
+    rm dummy.out
+    rm proc.out
     exit 1
 fi
 
+
 rm dummy.out
 rm proc.out
 
-kill ${pid}
-
 exit 0