From c1e29e8ce22e6136ae27697cbd6a2b9d5af53f1c Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Gr=C3=A9goire=20C?=
 <82955277+gc-advens@users.noreply.github.com>
Date: Mon, 30 Aug 2021 13:13:53 +0200
Subject: [PATCH 1/4] Fix #496: compatibility with cgroup v2

---
 functions/onprem/orborus/orborus.go | 67 +++--------------------------
 functions/onprem/orborus/run.sh     |  1 -
 functions/onprem/worker/worker.go   | 43 +-----------------
 3 files changed, 8 insertions(+), 103 deletions(-)

diff --git a/functions/onprem/orborus/orborus.go b/functions/onprem/orborus/orborus.go
index 6ca42b24..73faafeb 100644
--- a/functions/onprem/orborus/orborus.go
+++ b/functions/onprem/orborus/orborus.go
@@ -16,7 +16,6 @@ import (
 	"log"
 	"net/http"
 	"os"
-	"os/exec"
 	"strconv"
 	"strings"
 	"time"
@@ -53,13 +52,12 @@ var orgId = os.Getenv("ORG_ID")
 var baseUrl = os.Getenv("BASE_URL")
 var environment = os.Getenv("ENVIRONMENT_NAME")
 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("CONTAINER_NAME")
 var executionIds = []string{}
 
 var dockercli *dockerclient.Client
-var containerId string
 
 func init() {
 	var err error
@@ -68,60 +66,16 @@ func init() {
 	if err != nil {
 		panic(fmt.Sprintf("Unable to create docker client: %s", err))
 	}
-
-	getThisContainerId()
-}
-
-// form id of current running container
-func getThisContainerId() {
-	fCol := ""
-
-	// some adjusting based on current running mode
-	switch runningMode {
-	case "kubernetes":
-		// cgroup will be like:
-		// 11:net_cls,net_prio:/kubepods/besteffort/podf132b44d-cfcf-43f7-9906-79f58e268333/851466f8b5ed5aa0f265b1c95c6d2bafbc51a38dd5c5a1621b6e586572150009
-		fCol = "5"
-		log.Printf("[INFO] Running containerized in Kubernetes!")
-
-	case "docker":
-		// cgroup will be like:
-		// 12:perf_event:/docker/0f06810364f52a2cd6e80bfba27419cb8a29758a204cd676388f4913bb366f2b
-		fCol = "3"
-		log.Printf("[INFO] Running containerized in Docker!")
-
-	default:
-		fCol = "3" // for backward-compatibility with production
-		log.Printf("[WARNING] RUNNING_MODE not set - defaulting to Docker (NOT Kubernetes).")
-	}
-
-	if fCol != "" {
-		cmd := fmt.Sprintf("cat /proc/self/cgroup | grep memory | tail -1 | cut -d/ -f%s | grep -o -E '[0-9A-z]{64}'", fCol)
-		out, err := exec.Command("bash", "-c", cmd).Output()
-		if err == nil {
-			containerId = strings.TrimSpace(string(out))
-
-			// cgroup error. Hardcoding this.
-			// 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"
-				//docker-76c537e9a4b7c7233011f5d70e6b7f2d600b6413ac58a96519b8dca7a3f7117a.scope
-			}
-		} else {
-			if fCol == "0" {
-				containerId = "shuffle-orborus"
-				log.Printf("[WARNING] Failed getting container ID: %s", err)
-			}
-		}
-	}
-
-	log.Printf(`[INFO] Started with containerId "%s"`, containerId)
 }
 
 // Deploys the internal worker whenever something happens
 // https://docs.docker.com/engine/api/sdk/examples/
 func deployWorker(image string, identifier string, env []string) {
+	if containerName == "" {
+		log.Printf("[INFO] CONTAINER_NAME is not set, defaulting to 'shuffle-orborus'")
+		containerName = "shuffle-orborus"
+	}
+
 	// Binds is the actual "-v" volume.
 	// Max 20% CPU every second
 
@@ -137,14 +91,7 @@ func deployWorker(image string, identifier string, env []string) {
 		Binds: []string{
 			"/var/run/docker.sock:/var/run/docker.sock:rw",
 		},
-	}
-
-	// 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")
+		NetworkMode: container.NetworkMode(fmt.Sprintf("container:%s", containerName))
 	}
 
 	if cleanupEnv == "true" {
diff --git a/functions/onprem/orborus/run.sh b/functions/onprem/orborus/run.sh
index 48e5b701..9c6d8db5 100644
--- a/functions/onprem/orborus/run.sh
+++ b/functions/onprem/orborus/run.sh
@@ -3,7 +3,6 @@ docker run \
 	--env ENVIRONMENT_NAME="Shuffle" \
 	--env BASE_URL=http://shuffle-backend:5001 \
 	--env DOCKER_API_VERSION=1.42 \
-	--env RUNNING_MODE="Docker" \
 	--network "shuffle_shuffle" \
 	-v /var/run/docker.sock:/var/run/docker.sock \
 	frikky/shuffle:orborus
diff --git a/functions/onprem/worker/worker.go b/functions/onprem/worker/worker.go
index dfbe8312..b9ae4e59 100644
--- a/functions/onprem/worker/worker.go
+++ b/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{},
-	}
-
-	// 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")
+		NetworkMode: container.NetworkMode(fmt.Sprintf("container:worker-%s", workflowExecution.ExecutionId)),
 	}
 
 	// Removing because log extraction should happen first

From 9f3889c76d06f8e1c47b01168a032df809a09e52 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Gr=C3=A9goire=20C?=
 <82955277+gc-advens@users.noreply.github.com>
Date: Tue, 31 Aug 2021 08:20:14 +0200
Subject: [PATCH 2/4] Revert changes

---
 functions/onprem/orborus/orborus.go | 67 ++++++++++++++++++++++++++---
 functions/onprem/orborus/run.sh     |  1 +
 2 files changed, 61 insertions(+), 7 deletions(-)

diff --git a/functions/onprem/orborus/orborus.go b/functions/onprem/orborus/orborus.go
index 73faafeb..6ca42b24 100644
--- a/functions/onprem/orborus/orborus.go
+++ b/functions/onprem/orborus/orborus.go
@@ -16,6 +16,7 @@ import (
 	"log"
 	"net/http"
 	"os"
+	"os/exec"
 	"strconv"
 	"strings"
 	"time"
@@ -52,12 +53,13 @@ var orgId = os.Getenv("ORG_ID")
 var baseUrl = os.Getenv("BASE_URL")
 var environment = os.Getenv("ENVIRONMENT_NAME")
 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("CONTAINER_NAME")
 var executionIds = []string{}
 
 var dockercli *dockerclient.Client
+var containerId string
 
 func init() {
 	var err error
@@ -66,16 +68,60 @@ func init() {
 	if err != nil {
 		panic(fmt.Sprintf("Unable to create docker client: %s", err))
 	}
+
+	getThisContainerId()
+}
+
+// form id of current running container
+func getThisContainerId() {
+	fCol := ""
+
+	// some adjusting based on current running mode
+	switch runningMode {
+	case "kubernetes":
+		// cgroup will be like:
+		// 11:net_cls,net_prio:/kubepods/besteffort/podf132b44d-cfcf-43f7-9906-79f58e268333/851466f8b5ed5aa0f265b1c95c6d2bafbc51a38dd5c5a1621b6e586572150009
+		fCol = "5"
+		log.Printf("[INFO] Running containerized in Kubernetes!")
+
+	case "docker":
+		// cgroup will be like:
+		// 12:perf_event:/docker/0f06810364f52a2cd6e80bfba27419cb8a29758a204cd676388f4913bb366f2b
+		fCol = "3"
+		log.Printf("[INFO] Running containerized in Docker!")
+
+	default:
+		fCol = "3" // for backward-compatibility with production
+		log.Printf("[WARNING] RUNNING_MODE not set - defaulting to Docker (NOT Kubernetes).")
+	}
+
+	if fCol != "" {
+		cmd := fmt.Sprintf("cat /proc/self/cgroup | grep memory | tail -1 | cut -d/ -f%s | grep -o -E '[0-9A-z]{64}'", fCol)
+		out, err := exec.Command("bash", "-c", cmd).Output()
+		if err == nil {
+			containerId = strings.TrimSpace(string(out))
+
+			// cgroup error. Hardcoding this.
+			// 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"
+				//docker-76c537e9a4b7c7233011f5d70e6b7f2d600b6413ac58a96519b8dca7a3f7117a.scope
+			}
+		} else {
+			if fCol == "0" {
+				containerId = "shuffle-orborus"
+				log.Printf("[WARNING] Failed getting container ID: %s", err)
+			}
+		}
+	}
+
+	log.Printf(`[INFO] Started with containerId "%s"`, containerId)
 }
 
 // Deploys the internal worker whenever something happens
 // https://docs.docker.com/engine/api/sdk/examples/
 func deployWorker(image string, identifier string, env []string) {
-	if containerName == "" {
-		log.Printf("[INFO] CONTAINER_NAME is not set, defaulting to 'shuffle-orborus'")
-		containerName = "shuffle-orborus"
-	}
-
 	// Binds is the actual "-v" volume.
 	// Max 20% CPU every second
 
@@ -91,7 +137,14 @@ 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", containerName))
+	}
+
+	// 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" {
diff --git a/functions/onprem/orborus/run.sh b/functions/onprem/orborus/run.sh
index 9c6d8db5..48e5b701 100644
--- a/functions/onprem/orborus/run.sh
+++ b/functions/onprem/orborus/run.sh
@@ -3,6 +3,7 @@ docker run \
 	--env ENVIRONMENT_NAME="Shuffle" \
 	--env BASE_URL=http://shuffle-backend:5001 \
 	--env DOCKER_API_VERSION=1.42 \
+	--env RUNNING_MODE="Docker" \
 	--network "shuffle_shuffle" \
 	-v /var/run/docker.sock:/var/run/docker.sock \
 	frikky/shuffle:orborus

From 233477407c8ec69b148ba6dc7c721ea26f8fbc8b Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Gr=C3=A9goire=20C?=
 <82955277+gc-advens@users.noreply.github.com>
Date: Tue, 31 Aug 2021 08:45:41 +0200
Subject: [PATCH 3/4] Add fallback strategies when orborus container id was not
 found

---
 functions/onprem/orborus/orborus.go | 29 +++++++++++++++--------------
 1 file changed, 15 insertions(+), 14 deletions(-)

diff --git a/functions/onprem/orborus/orborus.go b/functions/onprem/orborus/orborus.go
index 6ca42b24..7bc275d0 100644
--- a/functions/onprem/orborus/orborus.go
+++ b/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("CONTAINER_NAME")
 var executionIds = []string{}
 
 var dockercli *dockerclient.Client
@@ -101,18 +102,25 @@ 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)
-			}
+			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)
 		}
 	}
 
@@ -137,14 +145,7 @@ func deployWorker(image string, identifier string, env []string) {
 		Binds: []string{
 			"/var/run/docker.sock:/var/run/docker.sock:rw",
 		},
-	}
-
-	// 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")
+		NetworkMode: container.NetworkMode(fmt.Sprintf("container:%s", containerId)),
 	}
 
 	if cleanupEnv == "true" {

From 80ee585e819e6b8c0b6c37e33c84051e54ddbe3b Mon Sep 17 00:00:00 2001
From: frikky <fredrik_9490@hotmail.com>
Date: Thu, 2 Sep 2021 01:46:21 +0200
Subject: [PATCH 4/4] #501: Temporary fix for issues with cgroup v2

---
 .env                                | 3 ++-
 functions/onprem/orborus/build.sh   | 2 +-
 functions/onprem/orborus/orborus.go | 2 +-
 functions/onprem/worker/build.sh    | 2 +-
 4 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/.env b/.env
index 8610fe2b..ae66b902 100644
--- a/.env
+++ b/.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
diff --git a/functions/onprem/orborus/build.sh b/functions/onprem/orborus/build.sh
index eb1d12db..3beae1f3 100644
--- a/functions/onprem/orborus/build.sh
+++ b/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
diff --git a/functions/onprem/orborus/orborus.go b/functions/onprem/orborus/orborus.go
index 7bc275d0..eb46a708 100644
--- a/functions/onprem/orborus/orborus.go
+++ b/functions/onprem/orborus/orborus.go
@@ -56,7 +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("CONTAINER_NAME")
+var containerName = os.Getenv("ORBORUS_CONTAINER_NAME")
 var executionIds = []string{}
 
 var dockercli *dockerclient.Client
diff --git a/functions/onprem/worker/build.sh b/functions/onprem/worker/build.sh
index b2f269c3..2dbd7f04 100644
--- a/functions/onprem/worker/build.sh
+++ b/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 .