Merge remote-tracking branch 'origin/ci-container-restart' into integration_2024_w49 (!3141)

CI: Allow restart of the container if deployment fails

In CI we sometimes encounter fail of the gNB/UE deployment (and fail of the test
scenario) caused by unsuccessful initialization of the USRP N310.

This MR enables to optionally restart the gNB or UE container, if the deployment
fails on the health check during start up. By default, restarts of the container
are not allowed, but we can enable them by setting num_attemps > 1 for a given
deployment in the XML file.

Logs from failed deployment attempts are collected.

This MR aims to avoid known CI failure caused by "USRP N310 Initialization Failure"
mentioned in #871.

ci-scripts/cls_containerize.py

@@ -145,10 +145,6 @@ def AnalyzeBuildLogs(buildRoot, images, globalStatus):
 		collectInfo[image] = files
 	return collectInfo
 
-def GetContainerName(ssh, svcName, file):
-	ret = ssh.run(f"docker compose -f {file} config --format json {svcName}  | jq -r '.services.\"{svcName}\".container_name'", silent=True)
-	return ret.stdout
-
 def GetImageName(ssh, svcName, file):
 	ret = ssh.run(f"docker compose -f {file} config --format json {svcName}  | jq -r '.services.\"{svcName}\".image'", silent=True)
 	if ret.returncode != 0:
@@ -156,17 +152,18 @@ def GetImageName(ssh, svcName, file):
 	else:
 		return ret.stdout.strip()
 
-def GetContainerHealth(ssh, containerName):
+def GetServiceHealth(ssh, svcName, file):
-	if containerName is None:
+	if svcName is None:
-		return False
+		return False, f"Service {svcName} not found in {file}"
-	if 'db_init' in containerName or 'db-init' in containerName: # exits with 0, there cannot be healthy
+	image = GetImageName(ssh, svcName, file)
-		return True
+	if 'db_init' in svcName or 'db-init' in svcName: # exits with 0, there cannot be healthy
+		return True, f"Service {svcName} healthy, image {image}"
 	for _ in range(8):
-		result = ssh.run(f'docker inspect --format="{{{{.State.Health.Status}}}}" {containerName}', silent=True)
+		result = ssh.run(f"docker compose -f {file} ps --format json {svcName}  | jq -r 'if type==\"array\" then .[0].Health else .Health end'", silent=True)
 		if result.stdout == 'healthy':
-			return True
+			return True, f"Service {svcName} healthy, image {image}"
 		time.sleep(5)
-	return False
+	return False, f"Failed to deploy: service {svcName}"
 
 def ExistEnvFilePrint(ssh, wd, prompt='env vars in existing'):
 	ret = ssh.run(f'cat {wd}/.env', silent=True, reportNonZero=False)
@@ -218,9 +215,9 @@ def GetServices(ssh, requested, file):
     else:
         return requested
 
-def CopyinContainerLog(ssh, lSourcePath, yaml, containerName, filename):
+def CopyinServiceLog(ssh, lSourcePath, yaml, svcName, wd_yaml, filename):
 	remote_filename = f"{lSourcePath}/cmake_targets/log/{filename}"
-	ssh.run(f'docker logs {containerName} &> {remote_filename}')
+	ssh.run(f'docker compose -f {wd_yaml} logs {svcName} --no-log-prefix &> {remote_filename}')
 	local_dir = f"{os.getcwd()}/../cmake_targets/log/{yaml}"
 	local_filename = f"{local_dir}/{filename}"
 	return ssh.copyin(remote_filename, local_filename)
@@ -336,6 +333,7 @@ class Containerize():
 		self.imageToCopy = ''
 		#checkers from xml
 		self.ran_checkers={}
+		self.num_attempts = 1
 
 		self.flexricTag = ''
 
@@ -843,50 +841,55 @@ class Containerize():
 
 	def DeployObject(self, HTML):
 		svr = self.eNB_serverId[self.eNB_instance]
+		num_attempts = self.num_attempts
 		lIpAddr, lSourcePath = self.GetCredentials(svr)
-		logging.debug('\u001B[1m Deploying OAI Object on server: ' + lIpAddr + '\u001B[0m')
+		logging.debug(f'Deploying OAI Object on server: {lIpAddr}')
 		yaml = self.yamlPath[self.eNB_instance].strip('/')
 		# creating the log folder by default
 		local_dir = f"{os.getcwd()}/../cmake_targets/log/{yaml.split('/')[-1]}"
 		os.system(f'mkdir -p {local_dir}')
 		wd = f'{lSourcePath}/{yaml}'
-
+		wd_yaml = f'{wd}/docker-compose.y*ml'
+		yaml_dir = yaml.split('/')[-1]
 		with cls_cmd.getConnection(lIpAddr) as ssh:
-			services = GetServices(ssh, self.services[self.eNB_instance], f"{wd}/docker-compose.y*ml")
+			services = GetServices(ssh, self.services[self.eNB_instance], wd_yaml)
 			if services == [] or services == ' ' or services == None:
-				msg = "Cannot determine services to start"
+				msg = 'Cannot determine services to start'
 				logging.error(msg)
 				HTML.CreateHtmlTestRowQueue('N/A', 'KO', [msg])
 				return False
-
 			ExistEnvFilePrint(ssh, wd)
 			WriteEnvFile(ssh, services, wd, self.deploymentTag, self.flexricTag)
-
+			if num_attempts <= 0:
-			logging.info(f"will start services {services}")
+				raise ValueError(f'Invalid value for num_attempts: {num_attempts}, must be greater than 0')
-			status = ssh.run(f'docker compose -f {wd}/docker-compose.y*ml up -d -- {services}')
+			for attempt in range(num_attempts):
-			if status.returncode != 0:
+				imagesInfo = []
-				msg = f"cannot deploy services {services}: {status.stdout}"
+				healthInfo = []
-				logging.error(msg)
+				logging.info(f'will start services {services}')
-				HTML.CreateHtmlTestRowQueue('N/A', 'KO', [msg])
+				status = ssh.run(f'docker compose -f {wd_yaml} up -d -- {services}')
-				return False
+				if status.returncode != 0:
-
+					msg = f'cannot deploy services {services}: {status.stdout}'
-			imagesInfo = []
+					logging.error(msg)
-			fstatus = True
+					HTML.CreateHtmlTestRowQueue('N/A', 'NOK', [msg])
-			for svc in services.split():
+					return False
-				containerName = GetContainerName(ssh, svc, f"{wd}/docker-compose.y*ml")
+				for svc in services.split():
-				healthy = GetContainerHealth(ssh, containerName)
+					health, msg = GetServiceHealth(ssh, svc, f'{wd_yaml}')
-				if not healthy:
+					logging.info(msg)
-					imagesInfo += [f"Failed to deploy: service {svc}"]
+					imagesInfo.append(msg)
-					fstatus = False
+					healthInfo.append(health)
-				else:
+				deployed = all(healthInfo)
-					image = GetImageName(ssh, svc, f"{wd}/docker-compose.y*ml")
+				if deployed:
-					logging.info(f"service {svc} healthy, container {containerName}, image {image}")
+					break
-					imagesInfo += [f"service {svc} healthy, image {image}"]
+				elif (attempt < num_attempts - 1):
-		if fstatus:
+					logging.warning(f'Failed to deploy on attempt {attempt}, restart services {services}')
+					for svc in services.split():
+						CopyinServiceLog(ssh, lSourcePath, yaml_dir, svc, wd_yaml, f'{svc}-{HTML.testCase_id}-attempt{attempt}.log')
+					ssh.run(f'docker compose -f {wd_yaml} down -- {services}')
+		if deployed:
 			HTML.CreateHtmlTestRowQueue('N/A', 'OK', ['\n'.join(imagesInfo)])
 		else:
 			HTML.CreateHtmlTestRowQueue('N/A', 'KO', ['\n'.join(imagesInfo)])
-		return fstatus
+		return deployed
 
 	def UndeployObject(self, HTML, RAN):
 		svr = self.eNB_serverId[self.eNB_instance]
@@ -902,7 +905,7 @@ class Containerize():
 			if services is not None:
 				all_serv = " ".join([s for s, _ in services])
 				ssh.run(f'docker compose -f {wd}/docker-compose.y*ml stop -- {all_serv}')
-				copyin_res = all(CopyinContainerLog(ssh, lSourcePath, yaml_dir, c, f'{s}-{HTML.testCase_id}.log') for s, c in services)
+				copyin_res = all(CopyinServiceLog(ssh, lSourcePath, yaml_dir, s, f"{wd}/docker-compose.y*ml", f'{s}-{HTML.testCase_id}.log') for s, c in services)
 			else:
 				logging.warning('could not identify services to stop => no log file')
 			ssh.run(f'docker compose -f {wd}/docker-compose.y*ml down -v')

ci-scripts/main.py

@@ -349,6 +349,7 @@ def ExecuteActionWithParam(action):
 		string_field = test.findtext('services')
 		if string_field is not None:
 			CONTAINERS.services[CONTAINERS.eNB_instance] = string_field
+		CONTAINERS.num_attempts = int(test.findtext('num_attempts') or 1)
 		CONTAINERS.deploymentTag = cls_containerize.CreateTag(CONTAINERS.ranCommitID, CONTAINERS.ranBranch, CONTAINERS.ranAllowMerge)
 		if action == 'Deploy_Object':
 			success = CONTAINERS.DeployObject(HTML)

ci-scripts/tests/deployment.py

@@ -47,6 +47,7 @@ class TestDeploymentMethods(unittest.TestCase):
 		self.cont.eNBUserName = None
 		self.cont.eNBPassword = None
 		self.cont.eNBSourceCodePath = os.getcwd()
+		self.cont.num_attempts = 3
 
 	def test_deploy(self):
 		self.cont.yamlPath[0] = 'tests/simple-dep/'
@@ -65,6 +66,15 @@ class TestDeploymentMethods(unittest.TestCase):
 		self.assertFalse(deploy)
 		self.cont.yamlPath = old
 
+	def test_deployfails_2svc(self):
+		# fails reliably
+		old = self.cont.yamlPath
+		self.cont.yamlPath[0] = 'tests/simple-fail-2svc/'
+		deploy = self.cont.DeployObject(self.html)
+		self.cont.UndeployObject(self.html, self.ran)
+		self.assertFalse(deploy)
+		self.cont.yamlPath = old
+
 	def test_deploy_ran(self):
 		self.cont.yamlPath[0] = 'yaml_files/5g_rfsimulator_tdd_dora'
 		self.cont.services[0] = "oai-gnb"

ci-scripts/tests/simple-fail-2svc/docker-compose.yml

@@ -0,0 +1,23 @@
+services:
+  test1:
+    image: ubuntu:jammy
+    container_name: test_container1
+    cap_drop:
+      - ALL
+    entrypoint: /bin/bash -c "false"
+    healthcheck:
+      test: /bin/bash -c "true"
+      interval: 1s
+      timeout: 1s
+      retries: 1
+  test2:
+    image: ubuntu:jammy
+    container_name: test_container2
+    cap_drop:
+      - ALL
+    entrypoint: /bin/bash -c "sleep infinity"
+    healthcheck:
+      test: /bin/bash -c "true"
+      interval: 1s
+      timeout: 1s
+      retries: 1

ci-scripts/xml_files/container_sa_n310_2X2_100MHz_quectel.xml

@@ -89,6 +89,7 @@
     <yaml_path>ci-scripts/yaml_files/5g_sa_n310_2x2_100MHz</yaml_path>
     <eNB_instance>0</eNB_instance>
     <eNB_serverId>0</eNB_serverId>
+    <num_attempts>3</num_attempts>
   </testCase>
 
   <testCase id="000001">

ci-scripts/xml_files/container_sa_n310_2X2_60MHz_quectel.xml

@@ -89,6 +89,7 @@
       <yaml_path>ci-scripts/yaml_files/5g_sa_n310_2x2_60MHz</yaml_path>
       <eNB_instance>0</eNB_instance>
       <eNB_serverId>0</eNB_serverId>
+      <num_attempts>3</num_attempts>
   </testCase>
 
   <testCase id="000001">

ci-scripts/xml_files/container_sa_n310_4X4_60MHz_quectel.xml

@@ -89,6 +89,7 @@
       <yaml_path>ci-scripts/yaml_files/5g_sa_n310_4x4_60MHz</yaml_path>
       <eNB_instance>0</eNB_instance>
       <eNB_serverId>0</eNB_serverId>
+      <num_attempts>3</num_attempts>
   </testCase>
 
   <testCase id="000001">

ci-scripts/xml_files/container_sa_n310_nrue.xml

@@ -80,6 +80,7 @@
     <yaml_path>ci-scripts/yaml_files/5g_sa_n310_gnb</yaml_path>
     <eNB_instance>0</eNB_instance>
     <eNB_serverId>0</eNB_serverId>
+    <num_attempts>3</num_attempts>
   </testCase>
   <testCase id="800814">
     <class>Create_Workspace</class>
@@ -93,6 +94,7 @@
     <yaml_path>ci-scripts/yaml_files/5g_sa_n310_nrue</yaml_path>
     <eNB_instance>1</eNB_instance>
     <eNB_serverId>1</eNB_serverId>
+    <num_attempts>3</num_attempts>
   </testCase>
  
   <testCase id="000001">