Merge branch 'gc-advens-fix-496' into launch

.env

@@ -41,7 +41,8 @@ HTTP_PROXY=
 HTTPS_PROXY=
 SHUFFLE_PASS_WORKER_PROXY=TRUE
 SHUFFLE_PASS_APP_PROXY=FALSE
-TZ=Europe/Amsterdam 					#Timezone-handler in Orborus, Worker and Apps
+TZ=Europe/Amsterdam 					# Timezone-handler in Orborus, Worker and Apps
+ORBORUS_CONTAINER_NAME=				# Used to FIND the containername. cgroup v2: issue 501
 
 SHUFFLE_BASE_IMAGE_REGISTRY=ghcr.io
 SHUFFLE_BASE_IMAGE_NAME=frikky

functions/onprem/orborus/build.sh

@@ -1,5 +1,5 @@
 NAME=shuffle-orborus
-VERSION=0.9.10
+VERSION=0.9.14
 
 echo "Running docker build with $NAME:$VERSION"
 #docker rmi frikky/shuffle:$NAME --force

functions/onprem/orborus/orborus.go

@@ -56,6 +56,7 @@ var dockerApiVersion = os.Getenv("DOCKER_API_VERSION")
 var runningMode = strings.ToLower(os.Getenv("RUNNING_MODE"))
 var cleanupEnv = strings.ToLower(os.Getenv("CLEANUP"))
 var timezone = os.Getenv("TZ")
+var containerName = os.Getenv("ORBORUS_CONTAINER_NAME")
 var executionIds = []string{}
 
 var dockercli *dockerclient.Client
@@ -101,19 +102,26 @@ func getThisContainerId() {
 		if err == nil {
 			containerId = strings.TrimSpace(string(out))
 
-			// cgroup error. Hardcoding this.
+			// cgroup error. Use fallback strategy below.
 			// https://github.com/moby/moby/issues/7015
 			//log.Printf("Checking if %s is in %s", ".scope", string(out))
 			if strings.Contains(string(out), ".scope") {
-				containerId = "shuffle-orborus"
+				containerId = ""
 				//docker-76c537e9a4b7c7233011f5d70e6b7f2d600b6413ac58a96519b8dca7a3f7117a.scope
 			}
 		} else {
-			if fCol == "0" {
-				containerId = "shuffle-orborus"
 			log.Printf("[WARNING] Failed getting container ID: %s", err)
 		}
 	}
+
+	if containerId == "" {
+		if containerName != "" {
+			containerId = containerName
+			log.Printf("[INFO] Falling back to CONTAINER_NAME as container ID")
+		} else {
+			containerId = "shuffle-orborus"
+			log.Printf(`[WARNING] CONTAINER_NAME is not set. Falling back to default name "%s" as container ID`, containerId)
+		}
 	}
 
 	log.Printf(`[INFO] Started with containerId "%s"`, containerId)
@@ -137,14 +145,7 @@ func deployWorker(image string, identifier string, env []string) {
 		Binds: []string{
 			"/var/run/docker.sock:/var/run/docker.sock:rw",
 		},
-	}
+		NetworkMode: container.NetworkMode(fmt.Sprintf("container:%s", containerId)),
-
-	// form container id and use it as network source if it's not empty
-	if containerId != "" {
-		//log.Printf("[INFO] Found container ID %s", containerId)
-		hostConfig.NetworkMode = container.NetworkMode(fmt.Sprintf("container:%s", containerId))
-	} else {
-		//log.Printf("[INFO] Empty self container id, continue without NetworkMode")
 	}
 
 	if cleanupEnv == "true" {

functions/onprem/worker/build.sh

@@ -1,5 +1,5 @@
 NAME=shuffle-worker
-VERSION=0.9.13
+VERSION=0.9.14
 
 echo "Running docker build with $NAME:$VERSION"
 #CGO_ENABLED=0 GOOS=linux go build -a -installsuffix cgo -o worker.bin .

functions/onprem/worker/worker.go

@@ -16,7 +16,6 @@ import (
 	"net/http"
 	"net/url"
 	"os"
-	"os/exec"
 	"strings"
 	"time"
 
@@ -42,7 +41,6 @@ var cleanupEnv = strings.ToLower(os.Getenv("CLEANUP"))
 var timezone = os.Getenv("TZ")
 var baseimagename = "frikky/shuffle"
 var registryName = "registry.hub.docker.com"
-var fallbackName = "shuffle-orborus"
 var sleepTime = 2
 var requestCache *cache.Cache
 var topClient *http.Client
@@ -61,38 +59,6 @@ var startAction string
 var results []shuffle.ActionResult
 var allLogs map[string]string
 
-var containerId string
-
-// form container id of current running container
-func getThisContainerId() string {
-	if len(containerId) > 0 {
-		return containerId
-	}
-
-	id := ""
-	cmd := fmt.Sprintf("cat /proc/self/cgroup | grep memory | tail -1 | cut -d/ -f3 | grep -o -E '[0-9A-z]{64}'")
-	out, err := exec.Command("bash", "-c", cmd).Output()
-	if err == nil {
-		id = strings.TrimSpace(string(out))
-
-		//log.Printf("Checking if %s is in %s", ".scope", string(out))
-		if strings.Contains(string(out), ".scope") {
-			id = fallbackName
-		}
-	}
-
-	return id
-}
-
-func init() {
-	containerId = getThisContainerId()
-	if len(containerId) == 0 {
-		log.Printf("[WARNING] No container ID found. Not running containerized? This should only show during testing")
-	} else {
-		log.Printf("[INFO] Found container ID for this worker: %s", containerId)
-	}
-}
-
 // removes every container except itself (worker)
 func shutdown(workflowExecution shuffle.WorkflowExecution, nodeId string, reason string, handleResultSend bool) {
 	log.Printf("[INFO] Shutdown (%s) started with reason %#v. Result amount: %d. ResultsSent: %d, Send result: %#v", workflowExecution.Status, reason, len(workflowExecution.Results), requestsSent, handleResultSend)
@@ -221,14 +187,7 @@ func deployApp(cli *dockerclient.Client, image string, identifier string, env []
 			Config: map[string]string{},
 		},
 		Resources: container.Resources{},
-	}
+		NetworkMode: container.NetworkMode(fmt.Sprintf("container:worker-%s", workflowExecution.ExecutionId)),
-
-	// form container id and use it as network source if it's not empty
-	containerId = getThisContainerId()
-	if containerId != "" {
-		hostConfig.NetworkMode = container.NetworkMode(fmt.Sprintf("container:%s", containerId))
-	} else {
-		log.Printf("[WARNING] Empty self container id, continue without NetworkMode")
 	}
 
 	// Removing because log extraction should happen first