TestON/tests/SCPF/SCPFscaleTopo/SCPFscaleTopo.py

# Testing network scalability, this test suite scales up a network topology
# using mininet and verifies ONOS stability

class SCPFscaleTopo:

    def __init__( self ):
        self.default = ''

    def CASE1( self, main ):
        import os
        import imp
        """
        - Construct tests variables
        - GIT ( optional )
            - Checkout ONOS master branch
            - Pull latest ONOS code
        - Building ONOS ( optional )
            - Install ONOS package
            - Build ONOS package
        """
        main.case( "Constructing test variables" )
        main.step( "Constructing test variables" )
        stepResult = main.FALSE
        # The variable to decide if the data should be written into data base.
        # 1 means Yes and -1 means No.
        main.writeData = 1
        main.searchTerm = main.params[ 'SearchTerm' ]
        main.testOnDirectory = os.path.dirname( os.getcwd ( ) )
        main.apps = main.params[ 'ENV' ][ 'cellApps' ]
        gitBranch = main.params[ 'GIT' ][ 'branch' ]
        main.dependencyPath = main.testOnDirectory + \
                              main.params[ 'DEPENDENCY' ][ 'path' ]
        main.tsharkResultPath = main.params[ 'TsharkPath' ]
        main.roleRequest = main.params[ 'SearchTerm' ]['roleRequest']
        main.multiovs = main.params[ 'DEPENDENCY' ][ 'multiovs' ]
        main.topoName = main.params[ 'TOPOLOGY' ][ 'topology' ]
        main.numCtrls = int( main.params[ 'CTRL' ][ 'numCtrls' ] )
        main.topoScale = ( main.params[ 'TOPOLOGY' ][ 'scale' ] ).split( "," )
        main.topoScaleSize = len( main.topoScale )
        wrapperFile1 = main.params[ 'DEPENDENCY' ][ 'wrapper1' ]
        wrapperFile2 = main.params[ 'DEPENDENCY' ][ 'wrapper2' ]
        wrapperFile3 = main.params[ 'DEPENDENCY' ][ 'wrapper3' ]
        main.topoCmpAttempts = int( main.params[ 'ATTEMPTS' ][ 'topoCmp' ] )
        main.pingallAttempts = int( main.params[ 'ATTEMPTS' ][ 'pingall' ] )
        main.startUpSleep = int( main.params[ 'SLEEP' ][ 'startup' ] )
        main.balanceSleep = int( main.params[ 'SLEEP' ][ 'balance' ] )
        main.nodeSleep = int( main.params[ 'SLEEP' ][ 'nodeSleep' ] )
        main.pingallSleep = int( main.params[ 'SLEEP' ][ 'pingall' ] )
        main.MNSleep = int( main.params[ 'SLEEP' ][ 'MNsleep' ] )
        main.pingTimeout = float( main.params[ 'TIMEOUT' ][ 'pingall' ] )
        main.hostDiscover = main.params[ 'TOPOLOGY' ][ 'host' ]
        main.hostDiscoverSleep = float( main.params['SLEEP']['host'] )
        if main.hostDiscover == 'True':
            main.hostDiscover = True
        else:
            main.hostDiscover = False
        gitPull = main.params[ 'GIT' ][ 'pull' ]
        main.homeDir = os.path.expanduser('~')
        main.cellData = {} # for creating cell file
        main.hostsData = {}
        main.CLIs = []
        main.ONOSip = []
        main.activeNodes = []
        main.ONOSip = main.ONOSbench.getOnosIps()

        for i in range(main.numCtrls):
                main.CLIs.append( getattr( main, 'ONOScli%s' % (i+1) ) )

        main.allinfo = {} # The dictionary to record all the data from karaf.log

        for i in range( 2 ):
            main.allinfo[ i ]={}
            for w in range ( 3 ):
                # Totaltime: the time from the new switchConnection to its end
                # swConnection: the time from the first new switchConnection to the last new switchConnection
                # lastSwToLastRr: the time from the last new switchConnection to the last role request
                # lastRrToLastTopology: the time form the last role request to the last topology
                # disconnectRate: the rate that shows how many switch disconnect after connection
                main.allinfo[ i ][ 'info' + str( w ) ]= { 'totalTime': 0, 'swConnection': 0, 'lastSwToLastRr': 0, 'lastRrToLastTopology': 0, 'disconnectRate': 0 }

        main.dbFilePath = main.params[ 'DATABASE' ][ 'dbPath' ]
        main.log.info( "Create Database file " + main.dbFilePath )
        resultDB = open(main.dbFilePath, 'w+' )
        resultDB.close()


        main.startUp = imp.load_source( wrapperFile1,
                                        main.dependencyPath +
                                        wrapperFile1 +
                                        ".py" )

        main.scaleTopoFunction = imp.load_source( wrapperFile2,
                                                  main.dependencyPath +
                                                  wrapperFile2 +
                                                  ".py" )

        main.topo = imp.load_source( wrapperFile3,
                                     main.dependencyPath +
                                     wrapperFile3 +
                                     ".py" )

        main.ONOSbench.scp( main.Mininet1,
                            main.dependencyPath +
                            main.multiovs,
                            main.Mininet1.home,
                            direction="to" )

        if main.CLIs:
                stepResult = main.TRUE
        else:
            main.log.error( "Did not properly created list of " +
                            "ONOS CLI handle" )
            stepResult = main.FALSE

        utilities.assert_equals( expect=main.TRUE,
                                 actual=stepResult,
                                 onpass="Successfully construct " +
                                        "test variables ",
                                 onfail="Failed to construct test variables" )


    def CASE2( self, main):
        """
        - Set up cell
            - Create cell file
            - Set cell file
            - Verify cell file
        - Kill ONOS process
        - Uninstall ONOS cluster
        - Verify ONOS start up
        - Install ONOS cluster
        - Connect to cli
        """
        import time
        main.log.info( "Checking if mininet is already running" )
        if len( main.topoScale ) < main.topoScaleSize:
            main.log.info( "Mininet is already running. Stopping mininet." )
            main.Mininet1.stopNet()
            time.sleep(main.MNSleep)
        else:
            main.log.info( "Mininet was not running" )

        main.commit = main.ONOSbench.getVersion(report=True)
        main.commit = main.commit.split(" ")[1]

        main.case( "Starting up " + str( main.numCtrls ) +
                   " node(s) ONOS cluster" )
        main.caseExplanation = "Set up ONOS with " + str( main.numCtrls ) +\
                                " node(s) ONOS cluster"

        #kill off all onos processes
        main.log.info( "Safety check, killing all ONOS processes" +
                       " before initiating environment setup" )

        for i in range( main.numCtrls ):
            main.ONOSbench.onosStop( main.ONOSip[ i ] )
            main.ONOSbench.onosKill( main.ONOSip[ i ] )

        tempOnosIp = []
        for i in range( main.numCtrls ):
            tempOnosIp.append( main.ONOSip[i] )

        main.ONOSbench.createCellFile( main.ONOSbench.ip_address,
                                       "temp", main.Mininet1.ip_address,
                                       main.apps, tempOnosIp, main.ONOScli1.user_name )

        main.step( "Apply cell to environment" )
        cellResult = main.ONOSbench.setCell( "temp" )
        verifyResult = main.ONOSbench.verifyCell()
        stepResult = cellResult and verifyResult
        utilities.assert_equals( expect=main.TRUE,
                                 actual=stepResult,
                                 onpass="Successfully applied cell to " + \
                                        "environment",
                                 onfail="Failed to apply cell to environment " )

        main.step( "Creating ONOS package" )
        packageResult = main.ONOSbench.buckBuild()
        stepResult = packageResult
        utilities.assert_equals( expect=main.TRUE,
                                 actual=stepResult,
                                 onpass="Successfully created ONOS package",
                                 onfail="Failed to create ONOS package" )

        time.sleep( main.startUpSleep )
        main.step( "Uninstalling ONOS package" )
        onosUninstallResult = main.TRUE
        for ip in main.ONOSip:
            onosUninstallResult = onosUninstallResult and \
                    main.ONOSbench.onosUninstall( nodeIp=ip )
        stepResult = onosUninstallResult
        utilities.assert_equals( expect=main.TRUE,
                                 actual=stepResult,
                                 onpass="Successfully uninstalled ONOS package",
                                 onfail="Failed to uninstall ONOS package" )

        time.sleep( main.startUpSleep )
        main.step( "Installing ONOS package" )
        onosInstallResult = main.TRUE
        for i in range( main.numCtrls ):
            onosInstallResult = onosInstallResult and \
                    main.ONOSbench.onosInstall( node=main.ONOSip[ i ] )
        stepResult = onosInstallResult
        utilities.assert_equals( expect=main.TRUE,
                                 actual=stepResult,
                                 onpass="Successfully installed ONOS package",
                                 onfail="Failed to install ONOS package" )

        main.step( "Set up ONOS secure SSH" )
        secureSshResult = main.TRUE
        for i in range( int( main.numCtrls ) ):
            secureSshResult = secureSshResult and main.ONOSbench.onosSecureSSH( node=main.ONOSip[i] )
        utilities.assert_equals( expect=main.TRUE, actual=secureSshResult,
                                 onpass="Test step PASS",
                                 onfail="Test step FAIL" )

        time.sleep( main.startUpSleep )
        main.step( "Starting ONOS service" )
        stopResult = main.TRUE
        startResult = main.TRUE
        onosIsUp = main.TRUE

        for i in range( main.numCtrls ):
            onosIsUp = onosIsUp and main.ONOSbench.isup( main.ONOSip[ i ] )
        if onosIsUp == main.TRUE:
            main.log.report( "ONOS instance is up and ready" )
        else:
            main.log.report( "ONOS instance may not be up, stop and " +
                             "start ONOS again " )

            for i in range( main.numCtrls ):
                stopResult = stopResult and \
                        main.ONOSbench.onosStop( main.ONOSip[ i ] )
            for i in range( main.numCtrls ):
                startResult = startResult and \
                        main.ONOSbench.onosStart( main.ONOSip[ i ] )
        stepResult = onosIsUp and stopResult and startResult
        utilities.assert_equals( expect=main.TRUE,
                                 actual=stepResult,
                                 onpass="ONOS service is ready",
                                 onfail="ONOS service did not start properly" )

        main.step( "Start ONOS cli" )
        cliResult = main.TRUE
        main.activeNodes = []
        for i in range( main.numCtrls ):
            cliResult = cliResult and \
                        main.CLIs[ i ].startOnosCli( main.ONOSip[ i ] )
            main.activeNodes.append( i )
        stepResult = cliResult
        utilities.assert_equals( expect=main.TRUE,
                                 actual=stepResult,
                                 onpass="Successfully start ONOS cli",
                                 onfail="Failed to start ONOS cli" )
        time.sleep( main.startUpSleep )

    def CASE10( self, main ):
        """
            Starting up torus topology
        """

        main.case( "Starting up Mininet and verifying topology" )
        main.caseExplanation = "Starting Mininet with a scalling topology and " +\
                "comparing topology elements between Mininet and ONOS"
        if main.topoScale:
            main.currScale = main.topoScale.pop(0)
        else: main.log.error( "topology scale is empty" )
        main.step( "Starting up TORUS %sx%s topology" % (main.currScale, main.currScale) )

        main.log.info( "Constructing Mininet command" )
        mnCmd = " mn --custom " + main.Mininet1.home + main.multiovs + \
                " --switch ovsm --topo " + main.topoName + "," + main.currScale + "," + main.currScale
        for i in range( main.numCtrls ):
                mnCmd += " --controller remote,ip=" + main.ONOSip[ i ]
        stepResult = main.Mininet1.startNet( mnCmd=mnCmd )
        utilities.assert_equals( expect=main.TRUE,
                                 actual=stepResult,
                                  onpass=main.topoName +
                                    " topology started successfully",
                                 onfail=main.topoName +
                                    " topology failed to start" )

        time.sleep( main.MNSleep )
        main.log.info( "Clean up Tshark" )
        with open(main.tsharkResultPath, "w" ) as tshark:
            tshark.write( "" )
        main.log.info( "Starting Tshark capture" )
        main.ONOSbench.tsharkGrep( main.roleRequest, main.tsharkResultPath, grepOptions='-E' )
        main.CLIs[ 0 ].activateApp( "org.onosproject.openflow" )
        time.sleep( main.MNSleep )
        main.log.info( "Stop Tshark" )
        main.ONOSbench.tsharkStop()
        main.log.info( "Get role request time" )
        with open( main.tsharkResultPath, "r" ) as resultFile:
            resultText = resultFile.readlines()
            resultFile.close()

    def CASE11( self, main ):
        """
            Compare topo, and sending Arping package
            if the topology is same, then Pass.
        """
        import json
        import time
        # First capture
        for i in range( 3 ):
            # Calculate total time
            main.allinfo[ 0 ][ 'info' + str( i )][ 'totalTime' ] = main.scaleTopoFunction.getInfoFromLog( main, main.searchTerm[ 'start' ], 'first', main.searchTerm[ 'end' ], 'last', index=i, funcMode='TD' )
            # Calculate switch connection time
            main.allinfo[ 0 ][ 'info' + str( i )][ 'swConnection' ] = main.scaleTopoFunction.getInfoFromLog( main, main.searchTerm[ 'start' ], 'first', main.searchTerm[ 'start' ], 'last', index=i, funcMode='TD' )
            # Calculate the time from last switch connection to the last role request
            main.allinfo[ 0 ][ 'info' + str( i )][ 'lastSwToLastRr' ] = main.scaleTopoFunction.compareTimeDiffWithRoleRequest( main, main.searchTerm[ 'start' ], 'last', index=i )
            # Calculate the time from the last role request to the last topology
            main.allinfo[ 0 ][ 'info' + str( i )][ 'lastRrToLastTopology' ] = main.scaleTopoFunction.compareTimeDiffWithRoleRequest( main, main.searchTerm[ 'end' ], 'last', index=i )
            # Calculate the disconnecti rate
            main.allinfo[ 0 ][ 'info' + str( i )][ 'disconnectRate' ] = main.scaleTopoFunction.getInfoFromLog( main, main.searchTerm[ 'Disconnect' ], 'num', main.searchTerm[ 'start' ], 'num', index=i, funcMode='DR' )
        main.log.debug( "The data is " + str( main.allinfo[ 0 ] ) )

        main.case( "Verifying topology: TORUS %sx%s" % ( main.currScale, main.currScale ) )
        main.caseExplanation = "Pinging all hosts and comparing topology " +\
                "elements between Mininet and ONOS"

        main.log.info( "Gathering topology information")
        time.sleep( main.MNSleep )
        stepResult = main.TRUE
        main.step( "Comparing MN topology to ONOS topology" )
        compareRetry = 0
        while compareRetry < 3:
            #While loop for retry
            devices = main.topo.getAllDevices( main )
            ports = main.topo.getAllPorts( main )
            links = main.topo.getAllLinks( main)
            mnSwitches = main.Mininet1.getSwitches()
            mnLinks = main.Mininet1.getLinks(timeout=180)

            for controller in range(len(main.activeNodes)):
                # controllerStr = str( main.activeNodes[controller] + 1 )
                if devices[ controller ] and ports[ controller ] and \
                                "Error" not in devices[ controller ] and \
                                "Error" not in ports[ controller ]:
                    currentDevicesResult = main.Mininet1.compareSwitches(
                            mnSwitches,
                            json.loads( devices[ controller ] ),
                            json.loads( ports[ controller ] ) )
                else:
                    currentDevicesResult = main.FALSE

                if links[ controller ] and "Error" not in links[ controller ]:
                    currentLinksResult = main.Mininet1.compareLinks(
                            mnSwitches, mnLinks,
                            json.loads( links[ controller ] ) )
                else:
                    currentLinksResult = main.FALSE

                stepResult = stepResult and currentDevicesResult and currentLinksResult
            if stepResult:
                break
            compareRetry += 1
        utilities.assert_equals(expect=main.TRUE,
                                actual=stepResult,
                                onpass=" Topology match Mininet",
                                onfail="ONOS Topology doesn't match Mininet")

        if stepResult:
            if main.hostDiscover:
                hostList = []
                for i in range( 1, int( main.currScale ) + 1 ):
                    for j in range( 1, int( main.currScale ) + 1) :
                        # Generate host list
                        hoststr = "h" + str(i) + "x" + str(j)
                        hostList.append(hoststr)
                for i in range( len(hostList) ):
                    totalHost = main.topo.sendArpPackage( main, hostList[i] )
                    time.sleep( main.hostDiscoverSleep )
                    if totalHost < 0:
                        # if totalHost less than 0 which means dependence function has exception.
                        main.log.info( "Error when discover host!" )
                        break
                if totalHost == int( main.currScale ) *  int( main.currScale ):
                    main.log.info( "Discovered all hosts" )
                    stepResult = stepResult and main.TRUE
                else:
                    main.log.warn( "Some hosts ware not discovered by ONOS... Topology doesn't match!" )
                    stepResult = main.FALSE
                utilities.assert_equals(expect=main.TRUE,
                                        actual=stepResult,
                                        onpass=" Topology match Mininet",
                                        onfail="ONOS Topology doesn't match Mininet")
            main.log.info( "Finished this iteration, continue to scale next topology." )
        else:
            main.log.info( "Clean up and exit TestON. Finished this test." )
            main.cleanup()
            main.exit()

    def CASE100( self, main ):
        '''
           Bring Down node 3
        '''

        main.case("Bring ONOS node 3 down: TORUS %sx%s" % (main.currScale, main.currScale))
        main.caseExplanation = "Balance masters to make sure " +\
                        "each controller has some devices and " +\
                        "stop ONOS node 3 service. "

        stepResult = main.FALSE
        main.step( "Bringing down node 3" )
        # Always bring down the third node
        main.deadNode = 2
        # Printing purposes
        node = main.deadNode + 1
        main.log.info( "Stopping node %s" % node )
        stepResult = main.ONOSbench.onosStop( main.ONOSip[ main.deadNode ] )
        main.log.info( "Removing dead node from list of active nodes" )
        main.activeNodes.pop( main.deadNode )

        utilities.assert_equals( expect=main.TRUE,
                                 actual=stepResult,
                                 onpass="Successfully bring down node 3",
                                 onfail="Failed to bring down node 3" )

    def CASE200( self, main ):
        '''
            Bring up onos node
        '''

        main.case("Bring ONOS node 3 up: TORUS %sx%s" % (main.currScale, main.currScale))
        main.caseExplanation = "Bring node 3 back up and balance the masters"

        node = main.deadNode + 1
        main.log.info( "Starting node %s" % node )
        stepResult = main.ONOSbench.onosStart( main.ONOSip[ main.deadNode ] )
        main.log.info( "Starting onos cli" )
        stepResult = stepResult and main.CLIs[ main.deadNode ].startOnosCli( main.ONOSip[ main.deadNode ] )

        main.log.info( "Adding previously dead node to list of active nodes" )
        main.activeNodes.append( main.deadNode )

        utilities.assert_equals( expect=main.TRUE,
                                 actual=stepResult,
                                 onpass="Successfully brought up onos node %s" % node,
                                 onfail="Failed to bring up onos node %s" % node )


        time.sleep(main.nodeSleep)

    def CASE300( self, main ):
        '''

            Balancing Masters
        '''
        time.sleep(main.balanceSleep)
        main.step( "Balancing Masters" )

        stepResult = main.FALSE
        if main.activeNodes:
            controller = main.activeNodes[0]
            stepResult = utilities.retry( main.CLIs[controller].balanceMasters,
                                          main.FALSE,
                                          [],
                                          sleep=3,
                                          attempts=3 )

        else:
            main.log.error( "List of active nodes is empty" )
        utilities.assert_equals( expect=main.TRUE,
                                 actual=stepResult,
                                 onpass="Balance masters was successfull",
                                 onfail="Failed to balance masters")
        time.sleep(main.balanceSleep)

    def CASE1000( self, main ):
        '''
            Report errors/warnings/exceptions
        '''
        # Compare the slowest Node through total time of each node
        slowestNode = 0
        slowestTotalTime = 0
        # Second capture
        for i in range( 3 ):
            # Calculate total time
            main.allinfo[ 1 ][ 'info' + str( i )][ 'totalTime' ] = main.scaleTopoFunction.getInfoFromLog( main, main.searchTerm[ 'start' ], 'first', main.searchTerm[ 'end' ], 'last', index=i, funcMode='TD' )
            # Compare the total time
            if main.allinfo[ 1 ][ 'info' + str( i ) ][ 'totalTime' ] > slowestTotalTime:
                slowestTotalTime = main.allinfo[ 1 ][ 'info' + str( i ) ][ 'totalTime' ]
                slowestNode = i
            # Calculate switch connection time
            main.allinfo[ 1 ][ 'info' + str( i )][ 'swConnection' ] = main.scaleTopoFunction.getInfoFromLog( main, main.searchTerm[ 'start' ], 'first', main.searchTerm[ 'start' ], 'last', index=i, funcMode='TD' )
            # Calculate the time from last switch connection to the last role request
            main.allinfo[ 1 ][ 'info' + str( i )][ 'lastSwToLastRr' ] = main.scaleTopoFunction.compareTimeDiffWithRoleRequest( main, main.searchTerm[ 'start' ], 'last', index=i )
            # Calculate the time from the last role request to the last topology
            main.allinfo[ 1 ][ 'info' + str( i )][ 'lastRrToLastTopology' ] = main.scaleTopoFunction.compareTimeDiffWithRoleRequest( main, main.searchTerm[ 'end' ], 'last', index=i )
            # Calculate the disconnecti rate
            main.allinfo[ 1 ][ 'info' + str( i )][ 'disconnectRate' ] = main.scaleTopoFunction.getInfoFromLog( main, main.searchTerm[ 'Disconnect' ], 'num', main.searchTerm[ 'start' ],'num', index=i, funcMode='DR' )

        if ( main.allinfo[ 0 ] != main.allinfo[ 1 ] ):
            main.log.error( "The results of two capture are different!" )
        main.log.debug( "The data is " + str( main.allinfo ) )
        if main.writeData != -1:
            main.log.info( "Write the date into database" )
            # write the date into data base
            with open( main.dbFilePath, "a" ) as dbFile:
                temp = str( main.currScale )
                temp += ",'baremetal1'"
                # put result from second capture into data base
                temp += "," + str( "%.2f" % main.allinfo[ 1 ][ 'info' + str( slowestNode )][ 'totalTime' ] )
                temp += "," + str( "%.2f" % main.allinfo[ 1 ][ 'info' + str( slowestNode )][ 'swConnection' ] )
                temp += "," + str( "%.2f" % main.allinfo[ 1 ][ 'info' + str( slowestNode )][ 'lastSwToLastRr' ] )
                temp += "," + str( "%.2f" % main.allinfo[ 1 ][ 'info' + str( slowestNode )][ 'lastRrToLastTopology' ] )
                temp += "," + str( "%.2f" % main.allinfo[ 1 ][ 'info' + str( slowestNode )][ 'disconnectRate' ] )
                temp += "\n"
                dbFile.write( temp )
        else:
            main.log.error( "The data from log is wrong!" )
        main.writeData = 1
        main.case( "Checking logs for errors, warnings, and exceptions" )
        main.log.info( "Error report: \n" )
        main.ONOSbench.logReport( main.ONOSip[ 0 ],
                                                            [ "INFO",
                                                              "FOLLOWER",
                                                              "WARN",
                                                              "flow",
                                                              "ERROR",
                                                              "Except" ],
                                                              "s" )