Merge pull request #43 from uditgaurav/az-down-non-default-vpc

Enhac(exp): Adds support for non-default vpc for the az down experiment

Author: uditgaurav

.github/workflows/push.yml

@@ -41,4 +41,4 @@ jobs:
           push: true
           file: Dockerfile
           platforms: linux/amd64,linux/arm64
-          tags: litmuschaos/py-runner:ci
+          tags: litmuschaos/py-runner:ci

build/go-multiarch-build.sh

@@ -55,10 +55,16 @@ def injectChaosInSerialMode(experimentsDetails , chaosDetails , eventsDetails ,
 		# Detaching the target zones from loa balancer 
 		for azone in targetZones:
 
-			logging.info("[Info]: Detaching the following zone(s), Zone Name : %s", azone)
-			err = statusAws.detachAZfromLB(experimentsDetails, azone)
+			logging.info("[Info]: Detaching the following zone, Zone Name %s", azone)
+			targetSubnet, err = statusAws.getTargetSubnet(experimentsDetails, azone)
+			subnetList = list(targetSubnet.split(" "))
 			if err != None:
 				return err
+			logging.info("[Info]: Detaching the following subnet, %s", subnetList)
+			err = statusAws.detachSubnet(experimentsDetails, subnetList)
+			if err != None:
+				return err
+
 			if chaosDetails.Randomness:
 				err = common.RandomInterval(experimentsDetails.ChaosInterval)
 				if err != None:
@@ -85,19 +91,17 @@ def injectChaosInSerialMode(experimentsDetails , chaosDetails , eventsDetails ,
 			duration = (datetime.now() - ChaosStartTimeStamp).seconds
 
 	logging.info("[Completion]: %s chaos is done",(experimentsDetails.ExperimentName))
-
 	return None
 
 # injectChaosInParallelMode disable the target available zone from loadbalancer in parallel mode (all at once)
 def injectChaosInParallelMode(experimentsDetails , chaosDetails , eventsDetails , resultDetails, clients, statusAws):
 
-
 	#ChaosStartTimeStamp contains the start timestamp, when the chaos injection begin
 	ChaosStartTimeStamp = datetime.now()
 	duration = (datetime.now() - ChaosStartTimeStamp).seconds
-	
+	subnet = []
+
 	while duration < experimentsDetails.ChaosDuration:
-		
   		# Get the target available zone details for the chaos execution
 		targetZones = experimentsDetails.LoadBalancerZones.split(",")
 		logging.info("[Info]: Target available zone list, %s", targetZones)
@@ -107,12 +111,17 @@ def injectChaosInParallelMode(experimentsDetails , chaosDetails , eventsDetails
 			types.SetEngineEventAttributes(eventsDetails, types.ChaosInject, msg, "Normal", chaosDetails)
 			events.GenerateEvents(eventsDetails, chaosDetails, "ChaosEngine",clients)
 
-		# Detaching the target zones from loa balancer
+		# Detaching the target zones from load balancer
 		for azone in targetZones:
-			logging.info("[Info]: Detaching the following zone(s), Zone Name %s", azone)
-			err = statusAws.detachAZfromLB(experimentsDetails, azone)
+			logging.info("[Info]: Detaching the following zone, Zone Name %s", azone)
+			targetSubnet, err = statusAws.getTargetSubnet(experimentsDetails, azone)
+			subnet.append(targetSubnet)
 			if err != None:
-				return err	
+				return err
+		logging.info("[Info]: Detaching the following subnet(s), %s", subnet)
+		err = statusAws.detachSubnet(experimentsDetails,subnet)
+		if err != None:
+			return err
 
 		if chaosDetails.Randomness:
 			err = common.RandomInterval(experimentsDetails.ChaosInterval)
@@ -128,10 +137,10 @@ def injectChaosInParallelMode(experimentsDetails , chaosDetails , eventsDetails
 		# Attaching the target available zone after the chaos injection
 		logging.info("[Status]: Attach the available zone back to load balancer")
 		for azone in targetZones:
-			err = statusAws.attachAZtoLB(experimentsDetails, azone)
+			err = statusAws.attachSubnet(experimentsDetails, subnet)
 			if err != None:
 				return err
-			
+     		
 		#Verify the status of available zone after the chaos injection
 		logging.info("[Status]: Checking AWS load balancer's AZ status")		
 		err = statusAws.CheckAWSStatus(experimentsDetails)

chaosLib/litmus/aws_az_chaos/lib/aws_az_chaos.py

@@ -8,8 +8,8 @@ def GetENV(experimentDetails):
 	experimentDetails.ExperimentName =  os.getenv("EXPERIMENT_NAME", "aws-az-chaos")
 	experimentDetails.ChaosNamespace = os.getenv("CHAOS_NAMESPACE", "")
 	experimentDetails.EngineName = os.getenv("CHAOSENGINE", "")
-	experimentDetails.ChaosDuration = maths.atoi(os.getenv("TOTAL_CHAOS_DURATION", ""))
-	experimentDetails.ChaosInterval = os.getenv("CHAOS_INTERVAL", "")
+	experimentDetails.ChaosDuration = maths.atoi(os.getenv("TOTAL_CHAOS_DURATION", "60"))
+	experimentDetails.ChaosInterval = os.getenv("CHAOS_INTERVAL", "30")
 	experimentDetails.RampTime = maths.atoi(os.getenv("RAMP_TIME", ""))
 	experimentDetails.ChaosLib = os.getenv("LIB", "litmus")
 	experimentDetails.ChaosUID = os.getenv("CHAOS_UID", "")
@@ -22,7 +22,7 @@ def GetENV(experimentDetails):
 	experimentDetails.LoadBalancerName = os.getenv("LOAD_BALANCER_NAME", "")
 	experimentDetails.LoadBalancerZones = os.getenv("LOAD_BALANCER_ZONES", "")
 	experimentDetails.LoadBalancerNameARN = os.getenv("LOAD_BALANCERNAME_ARN", "na")
-  
+
 #InitialiseChaosVariables initialise all the global variables
 def InitialiseChaosVariables(chaosDetails, experimentDetails):
 	appDetails = types.AppDetails()

pkg/aws_az/environment/environment.py

@@ -2,7 +2,7 @@
 class ExperimentDetails(object):
 	def __init__(self, ExperimentName=None, EngineName=None, ChaosDuration=None, ChaosInterval=None, RampTime=None, Force=None, ChaosLib=None, 
 		AWSZones=None, AWSRegion=None, AppNS=None, AppLabel=None, AppKind=None, InstanceID=None, ChaosNamespace=None, ChaosPodName=None, Timeout=None, 
-		Delay=None, LoadBalancerName= None, LIBImagePullPolicy=None, LoadBalancerNameARN=None, LoadBalancerZones=None, UID=None):
+		Delay=None, LoadBalancerName= None, LIBImagePullPolicy=None, LoadBalancerNameARN=None, LoadBalancerZones=None, UID=None, VPCType=None):
 		self.ExperimentName      = ExperimentName 
 		self.EngineName          = EngineName
 		self.ChaosDuration       = ChaosDuration
@@ -23,4 +23,3 @@ def __init__(self, ExperimentName=None, EngineName=None, ChaosDuration=None, Cha
 		self.LoadBalancerName	 = LoadBalancerName
 		self.LoadBalancerZones 	 = LoadBalancerZones
 		self.LoadBalancerNameARN = LoadBalancerNameARN
-

pkg/aws_az/types/types.py

@@ -1,3 +1,5 @@
+from distutils.log import error
+from re import subn
 import boto3, logging
 import pkg.utils.client.client as client
 
@@ -20,29 +22,56 @@ def CheckAWSStatus(self, experimentsDetails):
             return ValueError(exp)
         logging.info("[Info]: LoadBalancer and Availablity of zone has been checked")
 
-    # detachAZfromLB detaching availablity zone from load balancer
-    def detachAZfromLB(self, experimentsDetails, zone): 
-        self.clients = client.AWSClient().clientElb 
+    def getSubnetFromVPC(self, experimentsDetails): 
+        client = boto3.client('elb')
         try:
-            self.clients.disable_availability_zones_for_load_balancer(
-                LoadBalancerName=experimentsDetails.LoadBalancerName,
-                AvailabilityZones=[
-                    zone,
+            response = client.describe_load_balancers(
+                LoadBalancerNames=[
+                    experimentsDetails.LoadBalancerName,
                 ]
             )
+            return (response['LoadBalancerDescriptions'][0]['Subnets'])
+        except (self.clients.exceptions.AccessPointNotFoundException, self.clients.exceptions.InvalidConfigurationRequestException) as exp:
+            return ValueError(exp)
 
+    def getTargetSubnet(self, experimentsDetails, zone):
+        client = boto3.client('ec2')
+        try:
+            lst=self.getSubnetFromVPC(experimentsDetails)
+            i=0
+            for i in range(len(lst)):
+                response = client.describe_subnets(
+                    SubnetIds=[
+                        lst[i],
+                    ],
+                )
+                if(response['Subnets'][0]['AvailabilityZone']) == zone:
+                    return lst[i], None
+        except (self.clients.exceptions.AccessPointNotFoundException, self.clients.exceptions.InvalidConfigurationRequestException) as exp:
+            return lst[i], ValueError(exp)
+
+
+    def detachSubnet(self, experimentsDetails, subnet): 
+        client = boto3.client('elb')
+        try:
+                response = client.detach_load_balancer_from_subnets(
+                LoadBalancerName=experimentsDetails.LoadBalancerName,
+                Subnets=subnet
+            )
+                if (response['ResponseMetadata']['HTTPStatusCode']) != "200":
+                    ValueError("[Error]: Fail to detach the target subnet %s", subnet)
         except (self.clients.exceptions.AccessPointNotFoundException, self.clients.exceptions.InvalidConfigurationRequestException) as exp:
             return ValueError(exp)
-    
-    # attachAZtoLB attaching availablity zone from load balancer
-    def attachAZtoLB(self, experimentsDetails, zone):
-        self.clients = client.AWSClient().clientElb
+
+    def attachSubnet(self, experimentsDetails, subnet): 
+        client = boto3.client('elb')
         try:
-            self.clients.enable_availability_zones_for_load_balancer(
+                response = client.attach_load_balancer_to_subnets(
                 LoadBalancerName=experimentsDetails.LoadBalancerName,
-                AvailabilityZones=[
-                    zone,
-                ]
+                Subnets=subnet
             )
+                if (response['ResponseMetadata']['HTTPStatusCode']) != "200":
+                    ValueError("[Error]: Fail to attach the target subnet %s", subnet)
         except (self.clients.exceptions.AccessPointNotFoundException, self.clients.exceptions.InvalidConfigurationRequestException) as exp:
             return ValueError(exp)
+