ecs-init: custom health check server

ecs-init/chc/custom_healthcheck_server.go

@@ -0,0 +1,100 @@
+package chc
+
+import (
+	"context"
+	"errors"
+	"net"
+	"os"
+	"os/exec"
+	"strconv"
+	"time"
+
+	log "github.com/cihub/seelog"
+)
+
+const (
+	ecsInitSocket = "/var/run/ecs.sock"
+	hcCmdTimeout  = time.Second * 30
+)
+
+func StartCustomHealthCheckServer() {
+	err := os.Remove(ecsInitSocket) // remove any previous socket file
+	if err != nil {
+		log.Errorf("Error cleaning up old unix socket file: %v, err: %v", ecsInitSocket, err)
+		return
+	}
+
+	// start listening
+	l, err := net.ListenUnix("unix", &net.UnixAddr{ecsInitSocket, "unix"})
+	if err != nil {
+		log.Errorf("Error listening on the unix socket: %v, err: %v", ecsInitSocket, err)
+		return
+	}
+	defer l.Close()
+
+	log.Infof("Listening on %v ...", ecsInitSocket)
+
+	for {
+		conn, err := l.Accept()
+		if err != nil {
+			log.Errorf("Error accepting new client: %v", err)
+			continue
+		}
+		log.Infof("New client connected.")
+		handleConnectionErr := handleConnection(conn)
+		if handleConnectionErr != nil {
+			log.Errorf("Error handling connection: %v", err)
+			continue
+		}
+	}
+}
+
+func handleConnection(conn net.Conn) error {
+	defer conn.Close()
+
+	// read the request
+	buf := make([]byte, 1024)
+	n, err := conn.Read(buf)
+	if err != nil {
+		log.Errorf("Error reading: %v", err)
+		return err
+	}
+	request := string(buf[0:n])
+	log.Infof("Received a custom healthcheck cmd to run: %v", request)
+
+	// execute the custom healthcheck command
+	ctx, cancel := context.WithTimeout(context.Background(), hcCmdTimeout)
+	cmd := exec.CommandContext(ctx, request)
+	defer cancel()
+	err = cmd.Start()
+	if err != nil {
+		log.Errorf("Error starting the custom healthcheck command: %v, err: %v", cmd, err)
+		return err
+	}
+
+	// parse the custom health check output/exit code and send a response back to the ECS agent
+	response := "99"
+	err = cmd.Wait()
+	if err != nil {
+		var exitErr *exec.ExitError
+		if errors.As(err, &exitErr) {
+			response = strconv.Itoa(exitErr.ExitCode())
+		} else {
+			log.Errorf("invalid error during cmd.Wait(): %v", err)
+		}
+	} else {
+		response = "0"
+	}
+
+	log.Infof("Custom healthcheck command exit code: %v", response)
+	log.Infof("Sending response: %v", response)
+
+	_, err = conn.Write([]byte(response))
+	if err != nil {
+		log.Errorf("Error sending response: %v", err)
+		return err
+	}
+
+	log.Infof("Finished processing the custom healthcheck command: %v", request)
+	return nil
+}

ecs-init/engine/engine.go

@@ -24,6 +24,7 @@ import (
 	"github.com/aws/amazon-ecs-agent/ecs-init/apparmor"
 	"github.com/aws/amazon-ecs-agent/ecs-init/backoff"
 	"github.com/aws/amazon-ecs-agent/ecs-init/cache"
+	"github.com/aws/amazon-ecs-agent/ecs-init/chc"
 	"github.com/aws/amazon-ecs-agent/ecs-init/config"
 	"github.com/aws/amazon-ecs-agent/ecs-init/docker"
 	"github.com/aws/amazon-ecs-agent/ecs-init/exec"
@@ -276,6 +277,10 @@ func (e *Engine) StartSupervised() error {
 			return engineError("could not remove existing Agent container", err)
 		}
 
+		// start the custom healthcheck server,
+		// this will receive request from the ECS agent to run custom health check commands on the container instance
+		go chc.StartCustomHealthCheckServer()
+
 		log.Info("Starting Amazon Elastic Container Service Agent")
 		agentExitCode, err = docker.StartAgent()
 		if err != nil {