| """ |
| Copyright 2016 Open Networking Foundation ( ONF ) |
| |
| Please refer questions to either the onos test mailing list at <onos-test@onosproject.org>, |
| the System Testing Plans and Results wiki page at <https://wiki.onosproject.org/x/voMg>, |
| or the System Testing Guide page at <https://wiki.onosproject.org/x/WYQg> |
| |
| TestON is free software: you can redistribute it and/or modify |
| it under the terms of the GNU General Public License as published by |
| the Free Software Foundation, either version 2 of the License, or |
| ( at your option ) any later version. |
| |
| TestON is distributed in the hope that it will be useful, |
| but WITHOUT ANY WARRANTY; without even the implied warranty of |
| MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| GNU General Public License for more details. |
| |
| You should have received a copy of the GNU General Public License |
| along with TestON. If not, see <http://www.gnu.org/licenses/>. |
| """ |
| import os |
| import time |
| import json |
| import urllib |
| import re |
| from core import utilities |
| |
| |
| class Testcaselib: |
| |
| useSSH = True |
| |
| @staticmethod |
| def initTest( main ): |
| """ |
| - Construct tests variables |
| - GIT ( optional ) |
| - Checkout ONOS master branch |
| - Pull latest ONOS code |
| - Building ONOS ( optional ) |
| - Install ONOS package |
| - Build ONOS package |
| """ |
| try: |
| from tests.dependencies.ONOSSetup import ONOSSetup |
| main.testSetUp = ONOSSetup() |
| except ImportError: |
| main.log.error( "ONOSSetup not found. exiting the test" ) |
| main.cleanAndExit() |
| from tests.dependencies.Network import Network |
| main.Network = Network() |
| main.testSetUp.envSetupDescription( False ) |
| stepResult = main.FALSE |
| try: |
| # Test variables |
| main.cellName = main.params[ 'ENV' ][ 'cellName' ] |
| main.apps = main.params[ 'ENV' ][ 'cellApps' ] |
| main.path = os.path.dirname( main.testFile ) |
| main.useCommonTopo = main.params[ 'DEPENDENCY' ][ 'useCommonTopo' ] == 'True' |
| main.topoPath = main.path + ( "/.." if main.useCommonTopo else "" ) + "/dependencies/" |
| main.useCommonConf = main.params[ 'DEPENDENCY' ][ 'useCommonConf' ] == 'True' |
| main.configPath = main.path + ( "/.." if main.useCommonConf else "" ) + "/dependencies/" |
| main.forJson = "json/" |
| main.forChart = "chart/" |
| main.forConfig = "conf/" |
| main.forHost = "host/" |
| main.forSwitchFailure = "switchFailure/" |
| main.forLinkFailure = "linkFailure/" |
| main.forMulticast = "multicast/" |
| main.topology = main.params[ 'DEPENDENCY' ][ 'topology' ] |
| main.topologyLib = main.params[ 'DEPENDENCY' ][ 'lib' ] if 'lib' in main.params[ 'DEPENDENCY' ] else None |
| main.topologyConf = main.params[ 'DEPENDENCY' ][ 'conf' ] if 'conf' in main.params[ 'DEPENDENCY' ] else None |
| main.scale = ( main.params[ 'SCALE' ][ 'size' ] ).split( "," ) |
| main.maxNodes = int( main.params[ 'SCALE' ][ 'max' ] ) |
| |
| stepResult = main.testSetUp.envSetup( False ) |
| except Exception as e: |
| main.testSetUp.envSetupException( e ) |
| |
| main.testSetUp.evnSetupConclusion( stepResult ) |
| |
| @staticmethod |
| def installOnos( main, vlanCfg=True, skipPackage=False, cliSleep=10, |
| parallel=True ): |
| """ |
| - 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 |
| """ |
| # main.scale[ 0 ] determines the current number of ONOS controller |
| if not main.apps: |
| main.log.error( "App list is empty" ) |
| main.log.info( "Cluster size: " + str( main.Cluster.numCtrls ) ) |
| main.log.info( "Cluster ips: " + ', '.join( main.Cluster.getIps() ) ) |
| main.dynamicHosts = [ 'in1', 'out1' ] |
| main.testSetUp.ONOSSetUp( main.Cluster, newCell=True, cellName=main.cellName, |
| skipPack=skipPackage, |
| useSSH=Testcaselib.useSSH, |
| installParallel=parallel, includeCaseDesc=False ) |
| ready = utilities.retry( main.Cluster.active( 0 ).CLI.summary, |
| main.FALSE, |
| sleep=cliSleep, |
| attempts=10 ) |
| if ready: |
| ready = main.TRUE |
| utilities.assert_equals( expect=main.TRUE, actual=ready, |
| onpass="ONOS summary command succeded", |
| onfail="ONOS summary command failed" ) |
| if not ready: |
| main.log.error( "ONOS startup failed!" ) |
| main.cleanAndExit() |
| |
| for ctrl in main.Cluster.active(): |
| ctrl.CLI.logSet( "DEBUG", "org.onosproject.segmentrouting" ) |
| ctrl.CLI.logSet( "DEBUG", "org.onosproject.driver" ) |
| ctrl.CLI.logSet( "DEBUG", "org.onosproject.net.flowobjective.impl" ) |
| ctrl.CLI.logSet( "DEBUG", "org.onosproject.routeservice.impl" ) |
| ctrl.CLI.logSet( "DEBUG", "org.onosproject.routeservice.store" ) |
| ctrl.CLI.logSet( "DEBUG", "org.onosproject.routing.fpm" ) |
| ctrl.CLI.logSet( "TRACE", "org.onosproject.events" ) |
| ctrl.CLI.logSet( "DEBUG", "org.onosproject.mcast" ) |
| |
| @staticmethod |
| def loadCount( main ): |
| with open("%s/count/%s.count" % (main.configPath, main.cfgName)) as count: |
| main.count = json.load(count) |
| |
| @staticmethod |
| def loadJson( main ): |
| with open( "%s%s.json" % ( main.configPath + main.forJson, |
| main.cfgName ) ) as cfg: |
| main.Cluster.active( 0 ).REST.setNetCfg( json.load( cfg ) ) |
| |
| @staticmethod |
| def loadChart( main ): |
| try: |
| with open( "%s%s.chart" % ( main.configPath + main.forChart, |
| main.cfgName ) ) as chart: |
| main.pingChart = json.load(chart) |
| except IOError: |
| main.log.warn( "No chart file found." ) |
| |
| @staticmethod |
| def loadHost( main ): |
| with open( "%s%s.host" % ( main.configPath + main.forHost, |
| main.cfgName ) ) as host: |
| main.expectedHosts = json.load( host ) |
| |
| @staticmethod |
| def loadSwitchFailureChart( main ): |
| with open( "%s%s.switchFailureChart" % ( main.configPath + main.forSwitchFailure, |
| main.cfgName ) ) as sfc: |
| main.switchFailureChart = json.load( sfc ) |
| |
| @staticmethod |
| def loadLinkFailureChart( main ): |
| with open( "%s%s.linkFailureChart" % ( main.configPath + main.forLinkFailure, |
| main.cfgName ) ) as lfc: |
| main.linkFailureChart = json.load( lfc ) |
| |
| @staticmethod |
| def loadMulticastConfig( main ): |
| with open( "%s%s.multicastConfig" % ( main.configPath + main.forMulticast, |
| main.cfgName ) ) as cfg: |
| main.multicastConfig = json.load( cfg ) |
| |
| @staticmethod |
| def startMininet( main, topology, args="" ): |
| copyResult = main.ONOSbench.scp( main.Mininet1, |
| main.topoPath + main.topology, |
| main.Mininet1.home + "custom", |
| direction="to" ) |
| if main.topologyLib: |
| for lib in main.topologyLib.split(","): |
| copyResult = copyResult and main.ONOSbench.scp( main.Mininet1, |
| main.topoPath + lib, |
| main.Mininet1.home + "custom", |
| direction="to" ) |
| if main.topologyConf: |
| import re |
| controllerIPs = [ ctrl.ipAddress for ctrl in main.Cluster.runningNodes ] |
| index = 0 |
| for conf in main.topologyConf.split(","): |
| # Update zebra configurations with correct ONOS instance IP |
| if conf in [ "zebradbgp1.conf", "zebradbgp2.conf" ]: |
| ip = controllerIPs[ index ] |
| index = ( index + 1 ) % len( controllerIPs ) |
| with open( main.configPath + main.forConfig + conf ) as f: |
| s = f.read() |
| s = re.sub( r"(fpm connection ip).*(port 2620)", r"\1 " + ip + r" \2", s ) |
| with open( main.configPath + main.forConfig + conf, "w" ) as f: |
| f.write( s ) |
| copyResult = copyResult and main.ONOSbench.scp( main.Mininet1, |
| main.configPath + main.forConfig + conf, |
| "~/", |
| direction="to" ) |
| stepResult = copyResult |
| utilities.assert_equals( expect=main.TRUE, |
| actual=stepResult, |
| onpass="Successfully copied topo files", |
| onfail="Failed to copy topo files" ) |
| main.step( "Starting Mininet Topology" ) |
| arg = "--onos-ip=%s %s" % (",".join([ctrl.ipAddress for ctrl in main.Cluster.runningNodes]), args) |
| main.topology = topology |
| topoResult = main.Mininet1.startNet( |
| topoFile=main.Mininet1.home + "custom/" + main.topology, args=arg ) |
| stepResult = topoResult |
| utilities.assert_equals( expect=main.TRUE, |
| actual=stepResult, |
| onpass="Successfully loaded topology", |
| onfail="Failed to load topology" ) |
| # Exit if topology did not load properly |
| if not topoResult: |
| main.cleanAndExit() |
| |
| @staticmethod |
| def connectToPhysicalNetwork( main ): |
| main.step( "Connecting to physical netowrk" ) |
| topoResult = main.NetworkBench.connectToNet() |
| stepResult = topoResult |
| utilities.assert_equals( expect=main.TRUE, |
| actual=stepResult, |
| onpass="Successfully loaded topology", |
| onfail="Failed to load topology" ) |
| # Exit if topology did not load properly |
| if not topoResult: |
| main.cleanAndExit() |
| |
| main.step( "Assign switches to controllers." ) |
| assignResult = main.TRUE |
| switches = main.NetworkBench.getSwitches() |
| pool = [] |
| for name in switches.keys(): |
| thread = main.Thread( target=main.NetworkBench.assignSwController, |
| name="assignSwitchToController", |
| args=[ name, main.Cluster.getIps(), '6653' ] ) |
| pool.append( thread ) |
| thread.start() |
| for thread in pool: |
| thread.join( 300 ) |
| if not thread.result: |
| stepResult = main.FALSE |
| utilities.assert_equals( expect=main.TRUE, |
| actual=stepResult, |
| onpass="Successfully assign switches to controllers", |
| onfail="Failed to assign switches to controllers" ) |
| |
| # Check devices |
| Testcaselib.checkDevices( main, switches=int( main.params[ 'TOPO' ][ 'switchNum' ] ) ) |
| time.sleep( float( main.params[ "timers" ][ "connectToNetSleep" ] ) ) |
| # Connecting to hosts that only have data plane connectivity |
| main.step( "Connecting inband hosts" ) |
| stepResult = main.Network.connectInbandHosts() |
| utilities.assert_equals( expect=main.TRUE, |
| actual=stepResult, |
| onpass="Successfully connected inband hosts", |
| onfail="Failed to connect inband hosts" ) |
| |
| @staticmethod |
| def saveOnosDiagnostics( main ): |
| """ |
| Get onos-diags.tar.gz and save it to the log directory. |
| suffix: suffix string of the file name. E.g. onos-diags-case1.tar.gz |
| """ |
| main.log.info( "Collecting onos-diags..." ) |
| main.ONOSbench.onosDiagnostics( [ctrl.ipAddress for ctrl in main.Cluster.runningNodes], |
| main.logdir, |
| "-CASE%d" % main.CurrentTestCaseNumber ) |
| |
| @staticmethod |
| def config( main, cfgName ): |
| main.spines = [] |
| |
| main.failures = int( main.params[ 'failures' ] ) |
| main.cfgName = cfgName |
| |
| if main.cfgName == '2x2': |
| spine = {} |
| spine[ 'name' ] = main.params[ 'switches' ][ 'spine1' ] |
| spine[ 'dpid' ] = main.params[ 'switches' ][ 'spinedpid1' ] |
| main.spines.append( spine ) |
| |
| spine = {} |
| spine[ 'name' ] = main.params[ 'switches' ][ 'spine2' ] |
| spine[ 'dpid' ] = main.params[ 'switches' ][ 'spinedpid2' ] |
| main.spines.append( spine ) |
| |
| elif main.cfgName == '4x4': |
| spine = {} |
| spine[ 'name' ] = main.params[ 'switches' ][ 'spine1' ] |
| spine[ 'dpid' ] = main.params[ 'switches' ][ 'spinedpid1' ] |
| main.spines.append( spine ) |
| |
| spine = {} |
| spine[ 'name' ] = main.params[ 'switches' ][ 'spine2' ] |
| spine[ 'dpid' ] = main.params[ 'switches' ][ 'spinedpid2' ] |
| main.spines.append( spine ) |
| |
| spine = {} |
| spine[ 'name' ] = main.params[ 'switches' ][ 'spine3' ] |
| spine[ 'dpid' ] = main.params[ 'switches' ][ 'spinedpid3' ] |
| main.spines.append( spine ) |
| |
| spine = {} |
| spine[ 'name' ] = main.params[ 'switches' ][ 'spine4' ] |
| spine[ 'dpid' ] = main.params[ 'switches' ][ 'spinedpid4' ] |
| main.spines.append( spine ) |
| |
| else: |
| main.log.error( "Configuration failed!" ) |
| main.cleanAndExit() |
| |
| @staticmethod |
| def addStaticOnosRoute( main, subnet, intf): |
| """ |
| Adds an ONOS static route with the use route-add command. |
| """ |
| routeResult = main.Cluster.active( 0 ).addStaticRoute(subnet, intf) |
| |
| @staticmethod |
| def checkGroupsForBuckets( main, deviceId, subnetDict, routingTable=30 ): |
| """ |
| Check number of groups for each subnet on device deviceId and matches |
| it with an expected value. subnetDict is a dictionarty containing values |
| of the type "10.0.1.0/24" : 5. |
| """ |
| main.step( "Checking if number of groups for subnets in device {0} is as expected.".format( deviceId ) ) |
| groups = main.Cluster.active( 0 ).CLI.getGroups( deviceId, groupType="select" ) |
| flows = main.Cluster.active( 0 ).CLI.flows( jsonFormat=False, device=deviceId ) |
| |
| result = main.TRUE |
| for subnet, numberInSelect in subnetDict.iteritems(): |
| for flow in flows.splitlines(): |
| if "tableId={0}".format( routingTable ) in flow and subnet in flow: |
| # this will match the group id that this flow entry points to, for example : |
| # 0x70000041 in flow entry which contains "deferred=[GROUP:0x70000041], transition=TABLE:60," |
| groupId = re.search( r".*GROUP:(0x.*)], transition.*", flow ).groups()[0] |
| count = 0 |
| for group in groups.splitlines(): |
| if 'id={0}'.format( groupId ) in group: |
| count += 1 |
| if count - 1 != numberInSelect: |
| result = main.FALSE |
| main.log.warn( "Mismatch in number of buckets of select group, found {0}, expected {1} for subnet {2} on device {3}".format( count - 1, numberInSelect, subnet, deviceId ) ) |
| utilities.assert_equals( expect=main.TRUE, actual=result, |
| onpass="All bucket numbers are as expected", |
| onfail="Some bucket numbers are not as expected" ) |
| |
| @staticmethod |
| def checkFlows( main, minFlowCount, tag="", dumpflows=True, sleep=10 ): |
| main.step( |
| "Check whether the flow count is bigger than %s" % minFlowCount ) |
| if tag == "": |
| tag = 'CASE%d' % main.CurrentTestCaseNumber |
| count = utilities.retry( main.Cluster.active( 0 ).CLI.checkFlowCount, |
| main.FALSE, |
| kwargs={ 'min': minFlowCount }, |
| attempts=10, |
| sleep=sleep ) |
| if count == main.FALSE: |
| count = main.Cluster.active( 0 ).CLI.checkFlowCount() |
| utilities.assertEquals( |
| expect=True, |
| actual=( count > minFlowCount ), |
| onpass="Flow count looks correct: " + str( count ), |
| onfail="Flow count looks wrong: " + str( count ) ) |
| |
| main.step( "Check whether all flow status are ADDED" ) |
| flowCheck = utilities.retry( main.Cluster.active( 0 ).CLI.checkFlowsState, |
| main.FALSE, |
| kwargs={ 'isPENDING': False }, |
| attempts=5, |
| sleep=sleep ) |
| utilities.assertEquals( |
| expect=main.TRUE, |
| actual=flowCheck, |
| onpass="Flow status is correct!", |
| onfail="Flow status is wrong!" ) |
| if dumpflows: |
| main.ONOSbench.dumpONOSCmd( main.Cluster.active( 0 ).ipAddress, |
| "flows", |
| main.logdir, |
| tag + "_FlowsBefore" ) |
| main.ONOSbench.dumpONOSCmd( main.Cluster.active( 0 ).ipAddress, |
| "groups", |
| main.logdir, |
| tag + "_GroupsBefore" ) |
| |
| @staticmethod |
| def checkDevices( main, switches, tag="", sleep=10 ): |
| main.step( |
| "Check whether the switches count is equal to %s" % switches ) |
| if tag == "": |
| tag = 'CASE%d' % main.CurrentTestCaseNumber |
| result = utilities.retry( main.Cluster.active( 0 ).CLI.checkStatus, |
| main.FALSE, |
| kwargs={ 'numoswitch': switches}, |
| attempts=10, |
| sleep=sleep ) |
| utilities.assert_equals( expect=main.TRUE, actual=result, |
| onpass="Device up successful", |
| onfail="Failed to boot up devices?" ) |
| |
| @staticmethod |
| def checkFlowsByDpid( main, dpid, minFlowCount, sleep=10 ): |
| main.step( |
| " Check whether the flow count of device %s is bigger than %s" % ( dpid, minFlowCount ) ) |
| count = utilities.retry( main.Cluster.active( 0 ).CLI.checkFlowAddedCount, |
| main.FALSE, |
| args=( dpid, minFlowCount ), |
| attempts=5, |
| sleep=sleep ) |
| if count == main.FALSE: |
| count = main.Cluster.active( 0 ).CLI.checkFlowAddedCount( dpid ) |
| utilities.assertEquals( |
| expect=True, |
| actual=( count > minFlowCount ), |
| onpass="Flow count looks correct: " + str( count ), |
| onfail="Flow count looks wrong: " + str( count ) ) |
| |
| @staticmethod |
| def checkFlowEqualityByDpid( main, dpid, flowCount, sleep=10 ): |
| main.step( |
| " Check whether the flow count of device %s is equal to %s" % ( dpid, flowCount ) ) |
| count = utilities.retry( main.Cluster.active( 0 ).CLI.checkFlowAddedCount, |
| main.FALSE, |
| args=( dpid, flowCount, False, 1 ), |
| attempts=5, |
| sleep=sleep ) |
| if count == main.FALSE: |
| count = main.Cluster.active( 0 ).CLI.checkFlowAddedCount( dpid ) |
| utilities.assertEquals( |
| expect=True, |
| actual=( count == flowCount ), |
| onpass="Flow count looks correct: " + str( count ) , |
| onfail="Flow count looks wrong. found {}, should be {}.".format( count, flowCount ) ) |
| |
| @staticmethod |
| def checkGroupEqualityByDpid( main, dpid, groupCount, sleep=10): |
| main.step( |
| " Check whether the group count of device %s is equal to %s" % ( dpid, groupCount ) ) |
| count = utilities.retry( main.Cluster.active( 0 ).CLI.checkGroupAddedCount, |
| main.FALSE, |
| args=( dpid, groupCount, False, 1), |
| attempts=5, |
| sleep=sleep ) |
| if count == main.FALSE: |
| count = main.Cluster.active( 0 ).CLI.checkGroupAddedCount( dpid ) |
| utilities.assertEquals( |
| expect=True, |
| actual=( count == groupCount ), |
| onpass="Group count looks correct: " + str( count ) , |
| onfail="Group count looks wrong. found {}, should be {}.".format( count, groupCount ) ) |
| |
| @staticmethod |
| def checkFlowsGroupsFromFile( main ): |
| |
| for dpid, values in main.count.items(): |
| flowCount = values["flows"] |
| groupCount = values["groups"] |
| main.log.report( "Check flow count for dpid " + str( dpid ) + |
| ", should be " + str( flowCount ) ) |
| Testcaselib.checkFlowEqualityByDpid( main, dpid, flowCount ) |
| |
| main.log.report( "Check group count for dpid " + str( dpid ) + |
| ", should be " + str( groupCount ) ) |
| Testcaselib.checkGroupEqualityByDpid( main, dpid, groupCount ) |
| |
| return |
| |
| @staticmethod |
| def pingAll( main, tag="", dumpflows=True, acceptableFailed=0, basedOnIp=False, |
| sleep=10, retryAttempts=1, skipOnFail=False ): |
| ''' |
| Verify connectivity between hosts according to the ping chart |
| acceptableFailed: max number of acceptable failed pings. |
| basedOnIp: if True, run ping or ping6 based on suffix of host names |
| retryAttempts: the number of retry ping. Only works for IPv4 hosts. |
| ''' |
| main.log.report( "Check host connectivity" ) |
| main.log.debug( "Ping chart: %s" % main.pingChart ) |
| if tag == "": |
| tag = 'CASE%d' % main.CurrentTestCaseNumber |
| for entry in main.pingChart.itervalues(): |
| main.log.debug( "Entry in ping chart: %s" % entry ) |
| expect = entry[ 'expect' ] |
| if expect == "Unidirectional": |
| # Verify ping from each src host to each dst host |
| src = entry[ 'src' ] |
| dst = entry[ 'dst' ] |
| expect = main.TRUE |
| main.step( "Verify unidirectional connectivity from %s to %s with tag %s" % ( str( src ), str( dst ), tag ) ) |
| if basedOnIp: |
| if ("v4" in src[0]): |
| pa = main.Network.pingallHostsUnidirectional( src, dst, acceptableFailed=acceptableFailed ) |
| utilities.assert_equals( expect=expect, actual=pa, |
| onpass="IPv4 connectivity successfully tested", |
| onfail="IPv4 connectivity failed" ) |
| if ("v6" in src[0]): |
| pa = main.Network.pingallHostsUnidirectional( src, dst, ipv6=True, acceptableFailed=acceptableFailed ) |
| utilities.assert_equals( expect=expect, actual=pa, |
| onpass="IPv6 connectivity successfully tested", |
| onfail="IPv6 connectivity failed" ) |
| else: |
| pa = main.Network.pingallHostsUnidirectional( src, dst, acceptableFailed=acceptableFailed ) |
| utilities.assert_equals( expect=expect, actual=pa, |
| onpass="IP connectivity successfully tested", |
| onfail="IP connectivity failed" ) |
| else: |
| # Verify ping between each host pair |
| hosts = entry[ 'hosts' ] |
| try: |
| expect = main.TRUE if str(expect).lower() == 'true' else main.FALSE |
| except: |
| expect = main.FALSE |
| main.step( "Verify full connectivity for %s with tag %s" % ( str( hosts ), tag ) ) |
| if basedOnIp: |
| if ("v4" in hosts[0]): |
| pa = utilities.retry( main.Network.pingallHosts, |
| main.FALSE if expect else main.TRUE, |
| args=(hosts,), |
| attempts=retryAttempts, |
| sleep=sleep ) |
| utilities.assert_equals( expect=expect, actual=pa, |
| onpass="IPv4 connectivity successfully tested", |
| onfail="IPv4 connectivity failed" ) |
| if ("v6" in hosts[0]): |
| pa = main.Network.pingIpv6Hosts( hosts, acceptableFailed=acceptableFailed ) |
| utilities.assert_equals( expect=expect, actual=pa, |
| onpass="IPv6 connectivity successfully tested", |
| onfail="IPv6 connectivity failed" ) |
| else: |
| pa = main.Network.pingallHosts( hosts ) |
| utilities.assert_equals( expect=expect, actual=pa, |
| onpass="IP connectivity successfully tested", |
| onfail="IP connectivity failed" ) |
| if skipOnFail and pa != expect: |
| Testcaselib.saveOnosDiagnostics( main ) |
| Testcaselib.cleanup( main, copyKarafLog=False ) |
| main.skipCase() |
| |
| if dumpflows: |
| main.ONOSbench.dumpONOSCmd( main.Cluster.active( 0 ).ipAddress, |
| "flows", |
| main.logdir, |
| tag + "_FlowsOn" ) |
| main.ONOSbench.dumpONOSCmd( main.Cluster.active( 0 ).ipAddress, |
| "groups", |
| main.logdir, |
| tag + "_GroupsOn" ) |
| |
| @staticmethod |
| def killLink( main, end1, end2, switches, links, sleep=None ): |
| """ |
| end1,end2: identify the switches, ex.: 'leaf1', 'spine1' |
| switches, links: number of expected switches and links after linkDown, ex.: '4', '6' |
| Kill a link and verify ONOS can see the proper link change |
| """ |
| if sleep is None: |
| sleep = float( main.params[ 'timers' ][ 'LinkDiscovery' ] ) |
| else: |
| sleep = float( sleep ) |
| main.step( "Kill link between %s and %s" % ( end1, end2 ) ) |
| linkDown = main.Network.link( END1=end1, END2=end2, OPTION="down" ) |
| linkDown = linkDown and main.Network.link( END2=end1, END1=end2, OPTION="down" ) |
| # TODO: Can remove this, since in the retry we will wait anyways if topology is incorrect |
| main.log.info( |
| "Waiting %s seconds for link down to be discovered" % sleep ) |
| time.sleep( sleep ) |
| topology = utilities.retry( main.Cluster.active( 0 ).CLI.checkStatus, |
| main.FALSE, |
| kwargs={ 'numoswitch': switches, |
| 'numolink': links }, |
| attempts=10, |
| sleep=sleep ) |
| result = topology and linkDown |
| utilities.assert_equals( expect=main.TRUE, actual=result, |
| onpass="Link down successful", |
| onfail="Failed to turn off link?" ) |
| |
| @staticmethod |
| def killLinkBatch( main, links, linksAfter, switches, sleep=None ): |
| """ |
| links = list of links (src, dst) to bring down. |
| """ |
| |
| main.step("Killing a batch of links {0}".format(links)) |
| if sleep is None: |
| sleep = float( main.params[ 'timers' ][ 'LinkDiscovery' ] ) |
| else: |
| sleep = float( sleep ) |
| |
| for end1, end2 in links: |
| main.Network.link( END1=end1, END2=end2, OPTION="down") |
| main.Network.link( END1=end2, END2=end1, OPTION="down") |
| |
| # TODO: Can remove this, since in the retry we will wait anyways if topology is incorrect |
| main.log.info( |
| "Waiting %s seconds for links down to be discovered" % sleep ) |
| time.sleep( sleep ) |
| |
| topology = utilities.retry( main.Cluster.active( 0 ).CLI.checkStatus, |
| main.FALSE, |
| kwargs={ 'numoswitch': switches, |
| 'numolink': linksAfter }, |
| attempts=10, |
| sleep=sleep ) |
| |
| utilities.assert_equals( expect=main.TRUE, actual=topology, |
| onpass="Link batch down successful", |
| onfail="Link batch down failed" ) |
| |
| @staticmethod |
| def restoreLinkBatch( main, links, linksAfter, switches, sleep=None ): |
| """ |
| links = list of link (src, dst) to bring up again. |
| """ |
| |
| main.step("Restoring a batch of links {0}".format(links)) |
| if sleep is None: |
| sleep = float( main.params[ 'timers' ][ 'LinkDiscovery' ] ) |
| else: |
| sleep = float( sleep ) |
| |
| for end1, end2 in links: |
| main.Network.link( END1=end1, END2=end2, OPTION="up") |
| main.Network.link( END1=end2, END2=end1, OPTION="up") |
| |
| main.log.info( |
| "Waiting %s seconds for links up to be discovered" % sleep ) |
| time.sleep( sleep ) |
| |
| topology = utilities.retry( main.Cluster.active( 0 ).CLI.checkStatus, |
| main.FALSE, |
| kwargs={ 'numoswitch': switches, |
| 'numolink': linksAfter }, |
| attempts=10, |
| sleep=sleep ) |
| |
| utilities.assert_equals( expect=main.TRUE, actual=topology, |
| onpass="Link batch up successful", |
| onfail="Link batch up failed" ) |
| |
| @staticmethod |
| def disablePortBatch( main, ports, switches=None, links=None, sleep=None ): |
| """ |
| Disable a list of switch ports using 'portstate' and verify ONOS can see the proper link change |
| ports: a list of ports to disable ex. [ [ "of:0000000000000001", 1 ] ] |
| switches, links: number of expected switches and links after link change, ex.: '4', '6' |
| """ |
| if sleep is None: |
| sleep = float( main.params[ 'timers' ][ 'LinkDiscovery' ] ) |
| else: |
| sleep = float( sleep ) |
| main.step( "Disable a batch of ports" ) |
| for dpid, port in ports: |
| main.Cluster.active( 0 ).CLI.portstate( dpid=dpid, port=port, state="disable" ) |
| main.log.info( "Waiting {} seconds for port down to be discovered".format( sleep ) ) |
| time.sleep( sleep ) |
| if switches and links: |
| result = main.Cluster.active( 0 ).CLI.checkStatus( numoswitch=switches, |
| numolink=links ) |
| utilities.assert_equals( expect=main.TRUE, actual=result, |
| onpass="Port down successful", |
| onfail="Port down failed" ) |
| |
| @staticmethod |
| def enablePortBatch( main, ports, switches, links, sleep=None ): |
| """ |
| Enable a list of switch ports using 'portstate' and verify ONOS can see the proper link change |
| ports: a list of ports to enable ex. [ [ "of:0000000000000001", 1 ] ] |
| switches, links: number of expected switches and links after link change, ex.: '4', '6' |
| """ |
| if sleep is None: |
| sleep = float( main.params[ 'timers' ][ 'LinkDiscovery' ] ) |
| else: |
| sleep = float( sleep ) |
| main.step( "Enable a batch of ports" ) |
| for dpid, port in ports: |
| main.Cluster.active( 0 ).CLI.portstate( dpid=dpid, port=port, state="enable" ) |
| main.log.info( "Waiting {} seconds for port up to be discovered".format( sleep ) ) |
| time.sleep( sleep ) |
| if switches and links: |
| result = main.Cluster.active( 0 ).CLI.checkStatus( numoswitch=switches, |
| numolink=links ) |
| utilities.assert_equals( expect=main.TRUE, actual=result, |
| onpass="Port up successful", |
| onfail="Port up failed" ) |
| |
| @staticmethod |
| def restoreLink( main, end1, end2, switches, links, |
| portUp=False, dpid1='', dpid2='', port1='', port2='', sleep=None ): |
| """ |
| Params: |
| end1,end2: identify the end switches, ex.: 'leaf1', 'spine1' |
| portUp: enable portstate after restoring link |
| dpid1, dpid2: dpid of the end switches respectively, ex.: 'of:0000000000000002' |
| port1, port2: respective port of the end switches that connects to the link, ex.:'1' |
| switches, links: number of expected switches and links after linkDown, ex.: '4', '6' |
| Kill a link and verify ONOS can see the proper link change |
| """ |
| main.step( "Restore link between %s and %s" % ( end1, end2 ) ) |
| if sleep is None: |
| sleep = float( main.params[ 'timers' ][ 'LinkDiscovery' ] ) |
| else: |
| sleep = float( sleep ) |
| result = False |
| count = 0 |
| while True: |
| count += 1 |
| ctrl = main.Cluster.next() |
| main.Network.link( END1=end1, END2=end2, OPTION="up" ) |
| main.Network.link( END2=end1, END1=end2, OPTION="up" ) |
| main.log.info( |
| "Waiting %s seconds for link up to be discovered" % sleep ) |
| time.sleep( sleep ) |
| |
| if portUp: |
| ctrl.CLI.portstate( dpid=dpid1, port=port1, state='Enable' ) |
| ctrl.CLI.portstate( dpid=dpid2, port=port2, state='Enable' ) |
| time.sleep( sleep ) |
| |
| result = ctrl.CLI.checkStatus( numoswitch=switches, |
| numolink=links ) |
| if count > 5 or result: |
| break |
| utilities.assert_equals( expect=main.TRUE, actual=result, |
| onpass="Link up successful", |
| onfail="Failed to bring link up" ) |
| |
| @staticmethod |
| def killSwitch( main, switch, switches, links, sleep=None ): |
| """ |
| Params: switches, links: number of expected switches and links after SwitchDown, ex.: '4', '6' |
| Completely kill a switch and verify ONOS can see the proper change |
| """ |
| if sleep is None: |
| sleep = float( main.params[ 'timers' ][ 'SwitchDiscovery' ] ) |
| else: |
| sleep = float( sleep ) |
| switch = switch if isinstance( switch, list ) else [ switch ] |
| main.step( "Kill " + str( switch ) ) |
| for s in switch: |
| main.log.info( "Stopping " + s ) |
| main.Network.switch( SW=s, OPTION="stop" ) |
| # todo make this repeatable |
| |
| # TODO: Can remove this, since in the retry we will wait anyways if topology is incorrect |
| main.log.info( "Waiting %s seconds for switch down to be discovered" % ( |
| sleep ) ) |
| time.sleep( sleep ) |
| topology = utilities.retry( main.Cluster.active( 0 ).CLI.checkStatus, |
| main.FALSE, |
| kwargs={ 'numoswitch': switches, |
| 'numolink': links }, |
| attempts=10, |
| sleep=sleep ) |
| utilities.assert_equals( expect=main.TRUE, actual=topology, |
| onpass="Kill switch successful", |
| onfail="Failed to kill switch?" ) |
| |
| @staticmethod |
| def recoverSwitch( main, switch, switches, links, rediscoverHosts=False, hostsToDiscover=[], sleep=None ): |
| """ |
| Params: switches, links: number of expected switches and links after SwitchUp, ex.: '4', '6' |
| Recover a switch and verify ONOS can see the proper change |
| """ |
| if sleep is None: |
| sleep = float( main.params[ 'timers' ][ 'SwitchDiscovery' ] ) |
| else: |
| sleep = float( sleep ) |
| # TODO make this repeatable |
| switch = switch if isinstance( switch, list ) else [ switch ] |
| main.step( "Recovering " + str( switch ) ) |
| for s in switch: |
| main.log.info( "Starting " + s ) |
| main.Network.switch( SW=s, OPTION="start" ) |
| main.log.info( "Waiting %s seconds for switch up to be discovered" % ( |
| sleep ) ) |
| time.sleep( sleep ) |
| if rediscoverHosts: |
| main.Network.discoverHosts( hostList=hostsToDiscover ) |
| main.log.info( "Waiting %s seconds for hosts to get re-discovered" % ( |
| sleep ) ) |
| time.sleep( sleep ) |
| |
| topology = utilities.retry( main.Cluster.active( 0 ).CLI.checkStatus, |
| main.FALSE, |
| kwargs={ 'numoswitch': switches, |
| 'numolink': links }, |
| attempts=10, |
| sleep=sleep ) |
| utilities.assert_equals( expect=main.TRUE, actual=topology, |
| onpass="Switch recovery successful", |
| onfail="Failed to recover switch?" ) |
| |
| @staticmethod |
| def killRouter( main, router, sleep=None ): |
| """ |
| Kill bgpd process on a quagga router |
| router: name of the router to be killed. E.g. "bgp1" |
| """ |
| sleep = float( sleep ) |
| main.step( "Kill " + str( router ) ) |
| if hasattr( main, 'Mininet1' ): |
| main.Mininet1.handle.sendline( "px {}.stopProtocols()".format( router ) ) |
| main.Mininet1.handle.expect( "mininet>" ) |
| else: |
| # TODO: support killing router in physical network |
| pass |
| main.log.info( "Waiting %s seconds for router down to be discovered" % ( sleep ) ) |
| time.sleep( sleep ) |
| |
| @staticmethod |
| def recoverRouter( main, router, sleep=None ): |
| """ |
| Restart bgpd process on a quagga router |
| router: name of the router to be recovered. E.g. "bgp1" |
| """ |
| sleep = float( sleep ) |
| main.step( "Recovering " + str( router ) ) |
| if hasattr( main, 'Mininet1' ): |
| main.Mininet1.handle.sendline( "px {}.startProtocols()".format( router ) ) |
| main.Mininet1.handle.expect( "mininet>" ) |
| else: |
| # TODO: support recovering router in physical network |
| pass |
| main.log.info( "Waiting %s seconds for router up to be discovered" % ( sleep ) ) |
| time.sleep( sleep ) |
| |
| @staticmethod |
| def cleanup( main, copyKarafLog=True, removeHostComponent=False ): |
| """ |
| Stop Onos-cluster. |
| Stops Mininet |
| Copies ONOS log |
| """ |
| from tests.dependencies.utils import Utils |
| main.utils = Utils() |
| # Clean up scapy hosts |
| if hasattr( main, "scapyHosts" ): |
| scapyResult = main.TRUE |
| for host in main.scapyHosts: |
| scapyResult = host.stopScapy() and scapyResult |
| main.log.info( "Stopped Scapy Host: {0}".format( host.name ) ) |
| for host in main.scapyHosts: |
| if hasattr( main, 'Mininet1' ): |
| scapyResult = main.Scapy.removeHostComponent( host.name ) and scapyResult |
| else: |
| scapyResult = main.Network.removeHostComponent( host.name ) and scapyResult |
| main.log.info( "Removed Scapy Host Component: {0}".format( host.name ) ) |
| main.scapyHosts = [] |
| |
| if removeHostComponent: |
| for host in main.internalIpv4Hosts + main.internalIpv6Hosts + main.externalIpv4Hosts + main.externalIpv6Hosts: |
| if hasattr( main, host ): |
| if hasattr( main, 'Mininet1' ): |
| pass |
| else: |
| getattr( main, host ).disconnectInband() |
| main.Network.removeHostComponent( host ) |
| |
| if hasattr( main, 'Mininet1' ): |
| main.utils.mininetCleanup( main.Mininet1 ) |
| else: |
| main.Network.disconnectInbandHosts() |
| main.Network.disconnectFromNet() |
| |
| if copyKarafLog: |
| main.utils.copyKarafLog( "CASE%d" % main.CurrentTestCaseNumber, before=True, includeCaseDesc=False ) |
| |
| for ctrl in main.Cluster.active(): |
| main.ONOSbench.onosStop( ctrl.ipAddress ) |
| |
| @staticmethod |
| def verifyNodes( main ): |
| """ |
| Verifies Each active node in the cluster has an accurate view of other node's and their status |
| |
| Params: |
| nodes, integer array with position of the ONOS nodes in the CLIs array |
| """ |
| nodeResults = utilities.retry( main.Cluster.nodesCheck, |
| False, |
| attempts=10, |
| sleep=10 ) |
| utilities.assert_equals( expect=True, actual=nodeResults, |
| onpass="Nodes check successful", |
| onfail="Nodes check NOT successful" ) |
| |
| if not nodeResults: |
| for ctrl in main.Cluster.runningNodes: |
| main.log.debug( "{} components not ACTIVE: \n{}".format( |
| ctrl.name, |
| ctrl.CLI.sendline( "onos:scr-list | grep -v ACTIVE" ) ) ) |
| main.log.error( "Failed to verify nodes, stopping test" ) |
| main.cleanAndExit() |
| |
| @staticmethod |
| def verifyTopology( main, switches, links, expNodes ): |
| """ |
| Verifies that the ONOS cluster has an acuurate view of the topology |
| |
| Params: |
| switches, links, expNodes: number of expected switches, links, and nodes at this point in the test ex.: '4', '6', '2' |
| """ |
| main.step( "Check number of topology elements" ) |
| topology = utilities.retry( main.Cluster.active( 0 ).CLI.checkStatus, |
| main.FALSE, |
| kwargs={ 'numoswitch': switches, |
| 'numolink': links, |
| 'numoctrl': expNodes }, |
| attempts=10, |
| sleep=12 ) |
| utilities.assert_equals( expect=main.TRUE, actual=topology, |
| onpass="Number of topology elements are correct", |
| onfail="Unexpected number of links, switches, and/or controllers" ) |
| |
| @staticmethod |
| def killOnos( main, nodes, switches, links, expNodes, sleep=None ): |
| """ |
| Params: nodes, integer array with position of the ONOS nodes in the CLIs array |
| switches, links, nodes: number of expected switches, links and nodes after KillOnos, ex.: '4', '6' |
| Completely Kill an ONOS instance and verify the ONOS cluster can see the proper change |
| """ |
| # TODO: We have enough information in the Cluster instance to remove expNodes from here and verifyTopology |
| main.step( "Killing ONOS instances with index(es): {}".format( nodes ) ) |
| if sleep is None: |
| sleep = float( main.params[ 'timers' ][ 'OnosDiscovery' ] ) |
| else: |
| sleep = float( sleep ) |
| |
| for i in nodes: |
| killResult = main.ONOSbench.onosDie( main.Cluster.runningNodes[ i ].ipAddress ) |
| utilities.assert_equals( expect=main.TRUE, actual=killResult, |
| onpass="ONOS instance Killed", |
| onfail="Error killing ONOS instance" ) |
| main.Cluster.runningNodes[ i ].active = False |
| main.Cluster.reset() |
| time.sleep( sleep ) |
| |
| if len( nodes ) < main.Cluster.numCtrls: |
| Testcaselib.verifyNodes( main ) |
| Testcaselib.verifyTopology( main, switches, links, expNodes ) |
| |
| @staticmethod |
| def recoverOnos( main, nodes, switches, links, expNodes, sleep=None ): |
| """ |
| Params: nodes, integer array with position of the ONOS nodes in the CLIs array |
| switches, links, nodes: number of expected switches, links and nodes after recoverOnos, ex.: '4', '6' |
| Recover an ONOS instance and verify the ONOS cluster can see the proper change |
| """ |
| main.step( "Recovering ONOS instances with index(es): {}".format( nodes ) ) |
| if sleep is None: |
| sleep = float( main.params[ 'timers' ][ 'OnosDiscovery' ] ) |
| else: |
| sleep = float( sleep ) |
| [ main.ONOSbench.onosStart( main.Cluster.runningNodes[ i ].ipAddress ) for i in nodes ] |
| time.sleep( sleep ) |
| for i in nodes: |
| isUp = main.ONOSbench.isup( main.Cluster.runningNodes[ i ].ipAddress ) |
| utilities.assert_equals( expect=main.TRUE, actual=isUp, |
| onpass="ONOS service is ready", |
| onfail="ONOS service did not start properly" ) |
| for i in nodes: |
| main.step( "Checking if ONOS CLI is ready" ) |
| ctrl = main.Cluster.runningNodes[ i ] |
| # ctrl.CLI.startCellCli() |
| cliResult = ctrl.CLI.startOnosCli( ctrl.ipAddress, |
| commandlineTimeout=60, |
| onosStartTimeout=100 ) |
| ctrl.active = True |
| utilities.assert_equals( expect=main.TRUE, |
| actual=cliResult, |
| onpass="ONOS CLI is ready", |
| onfail="ONOS CLI is not ready" ) |
| |
| main.Cluster.reset() |
| main.step( "Checking ONOS nodes" ) |
| Testcaselib.verifyNodes( main ) |
| Testcaselib.verifyTopology( main, switches, links, expNodes ) |
| |
| ready = utilities.retry( main.Cluster.active( 0 ).CLI.summary, |
| main.FALSE, |
| attempts=10, |
| sleep=12 ) |
| if ready: |
| ready = main.TRUE |
| utilities.assert_equals( expect=main.TRUE, actual=ready, |
| onpass="ONOS summary command succeded", |
| onfail="ONOS summary command failed" ) |
| if not ready: |
| main.log.error( "ONOS startup failed!" ) |
| main.cleanAndExit() |
| |
| @staticmethod |
| def addHostCfg( main ): |
| """ |
| Adds Host Configuration to ONOS |
| Updates expected state of the network ( pingChart ) |
| """ |
| import json |
| hostCfg = {} |
| with open( main.configPath + main.forJson + "extra.json" ) as template: |
| hostCfg = json.load( template ) |
| main.pingChart[ 'ip' ][ 'hosts' ] += [ 'in1' ] |
| main.step( "Pushing new configuration" ) |
| mac, cfg = hostCfg[ 'hosts' ].popitem() |
| main.Cluster.active( 0 ).REST.setNetCfg( cfg[ 'basic' ], |
| subjectClass="hosts", |
| subjectKey=urllib.quote( mac, |
| safe='' ), |
| configKey="basic" ) |
| main.pingChart[ 'ip' ][ 'hosts' ] += [ 'out1' ] |
| main.step( "Pushing new configuration" ) |
| mac, cfg = hostCfg[ 'hosts' ].popitem() |
| main.Cluster.active( 0 ).REST.setNetCfg( cfg[ 'basic' ], |
| subjectClass="hosts", |
| subjectKey=urllib.quote( mac, |
| safe='' ), |
| configKey="basic" ) |
| main.pingChart.update( { 'vlan1': { "expect": "True", |
| "hosts": [ "olt1", "vsg1" ] } } ) |
| main.pingChart[ 'vlan5' ][ 'expect' ] = 0 |
| main.pingChart[ 'vlan10' ][ 'expect' ] = 0 |
| ports = "[%s,%s]" % ( 5, 6 ) |
| cfg = '{"of:0000000000000001":[{"vlan":1,"ports":%s,"name":"OLT 1"}]}' % ports |
| main.Cluster.active( 0 ).REST.setNetCfg( json.loads( cfg ), |
| subjectClass="apps", |
| subjectKey="org.onosproject.segmentrouting", |
| configKey="xconnect" ) |
| |
| @staticmethod |
| def delHostCfg( main ): |
| """ |
| Removest Host Configuration from ONOS |
| Updates expected state of the network ( pingChart ) |
| """ |
| import json |
| hostCfg = {} |
| with open( main.configPath + main.forJson + "extra.json" ) as template: |
| hostCfg = json.load( template ) |
| main.step( "Removing host configuration" ) |
| main.pingChart[ 'ip' ][ 'expect' ] = 0 |
| mac, cfg = hostCfg[ 'hosts' ].popitem() |
| main.Cluster.active( 0 ).REST.removeNetCfg( subjectClass="hosts", |
| subjectKey=urllib.quote( |
| mac, |
| safe='' ), |
| configKey="basic" ) |
| main.step( "Removing configuration" ) |
| main.pingChart[ 'ip' ][ 'expect' ] = 0 |
| mac, cfg = hostCfg[ 'hosts' ].popitem() |
| main.Cluster.active( 0 ).REST.removeNetCfg( subjectClass="hosts", |
| subjectKey=urllib.quote( |
| mac, |
| safe='' ), |
| configKey="basic" ) |
| main.step( "Removing vlan configuration" ) |
| main.pingChart[ 'vlan1' ][ 'expect' ] = 0 |
| main.Cluster.active( 0 ).REST.removeNetCfg( subjectClass="apps", |
| subjectKey="org.onosproject.segmentrouting", |
| configKey="xconnect" ) |
| |
| @staticmethod |
| def verifyNetworkHostIp( main, attempts=10, sleep=10 ): |
| """ |
| Verifies IP address assignment from the hosts |
| """ |
| main.step( "Verify IP address assignment from hosts" ) |
| ipResult = main.TRUE |
| main.Network.update() |
| # Find out names of disconnected hosts |
| disconnectedHosts = [] |
| if hasattr( main, "disconnectedIpv4Hosts" ): |
| for host in main.disconnectedIpv4Hosts: |
| disconnectedHosts.append( host ) |
| if hasattr( main, "disconnectedIpv6Hosts" ): |
| for host in main.disconnectedIpv6Hosts: |
| disconnectedHosts.append( host ) |
| for hostName, ip in main.expectedHosts[ "network" ].items(): |
| # Exclude disconnected hosts |
| if hostName in disconnectedHosts: |
| main.log.debug( "Skip verifying IP for {} as it's disconnected".format( hostName ) ) |
| continue |
| ipResult = ipResult and utilities.retry( main.Network.verifyHostIp, |
| main.FALSE, |
| kwargs={ 'hostList': [ hostName ], |
| 'prefix': ip, |
| 'update': False }, |
| attempts=attempts, |
| sleep=sleep ) |
| utilities.assert_equals( expect=main.TRUE, actual=ipResult, |
| onpass="Verify network host IP succeded", |
| onfail="Verify network host IP failed" ) |
| |
| @staticmethod |
| def verifyOnosHostIp( main, attempts=10, sleep=10, skipOnFail=True ): |
| """ |
| Verifies host IP address assignment from ONOS |
| """ |
| main.step( "Verify host IP address assignment in ONOS" ) |
| ipResult = main.TRUE |
| # Find out IPs of disconnected hosts |
| disconnectedIps = [] |
| if hasattr( main, "disconnectedIpv4Hosts" ): |
| for host in main.disconnectedIpv4Hosts: |
| disconnectedIps.append( main.expectedHosts[ "network" ][ host ] ) |
| if hasattr( main, "disconnectedIpv6Hosts" ): |
| for host in main.disconnectedIpv6Hosts: |
| disconnectedIps.append( main.expectedHosts[ "network" ][ host ] ) |
| for hostName, ip in main.expectedHosts[ "onos" ].items(): |
| # Exclude disconnected hosts |
| if ip in disconnectedIps: |
| main.log.debug( "Skip verifying IP for {} as it's disconnected".format( ip ) ) |
| continue |
| ipResult = ipResult and utilities.retry( main.Cluster.active( 0 ).verifyHostIp, |
| main.FALSE, |
| kwargs={ 'hostList': [ hostName ], |
| 'prefix': ip }, |
| attempts=attempts, |
| sleep=sleep ) |
| utilities.assert_equals( expect=main.TRUE, actual=ipResult, |
| onpass="Verify ONOS host IP succeded", |
| onfail="Verify ONOS host IP failed" ) |
| if not ipResult and skipOnFail: |
| Testcaselib.saveOnosDiagnostics( main ) |
| Testcaselib.cleanup( main, copyKarafLog=False ) |
| main.skipCase() |
| |
| @staticmethod |
| def updateIntfCfg( main, connectPoint, ips=[], untagged=0, tagged=[], native=0 ): |
| """ |
| Description: |
| Updates interface configuration in ONOS, with given IP and vlan parameters |
| Required: |
| * connectPoint: connect point to update configuration |
| Optional: |
| * ips: list of IP addresses, combined with '/xx' subnet representation, |
| corresponding to 'ips' field in the configuration |
| * untagged: vlan ID as an integer, corresponding to 'vlan-untagged' field in the configuration |
| * tagged: integer list of vlan IDs, corresponding to 'vlan-tagged' field in the configuration |
| * native: vlan ID as an integer, corresponding to 'vlan-native' field in the configuration |
| """ |
| cfg = dict() |
| cfg[ "ports" ] = dict() |
| cfg[ "ports" ][ connectPoint ] = dict() |
| cfg[ "ports" ][ connectPoint ][ "interfaces" ] = [ dict() ] |
| cfg[ "ports" ][ connectPoint ][ "interfaces" ][ 0 ][ "ips" ] = ips |
| if untagged > 0: |
| cfg[ "ports" ][ connectPoint ][ "interfaces" ][ 0 ][ "vlan-untagged" ] = untagged |
| else: |
| cfg[ "ports" ][ connectPoint ][ "interfaces" ][ 0 ][ "vlan-tagged" ] = tagged |
| if native > 0: |
| cfg[ "ports" ][ connectPoint ][ "interfaces" ][ 0 ][ "vlan-native" ] = native |
| |
| main.Cluster.active( 0 ).REST.setNetCfg( json.loads( json.dumps( cfg ) ) ) |
| |
| @staticmethod |
| def startScapyHosts( main, scapyNames=[], mininetNames=[] ): |
| """ |
| Create host components and start Scapy CLIs |
| scapyNames: list of names that will be used as component names for scapy hosts |
| mininetNames: used when scapy host names are different from the host names |
| in Mininet. E.g. when scapyNames=['h1Scapy'], it's required to specify the |
| name of the corresponding Mininet host by mininetNames=['h1'] |
| """ |
| main.step( "Start Scapy CLIs" ) |
| main.scapyNames = scapyNames if scapyNames else main.params[ 'SCAPY' ][ 'HOSTNAMES' ].split( ',' ) |
| main.scapyHosts = [] if not hasattr( main, "scapyHosts" ) else main.scapyHosts |
| for scapyName in main.scapyNames: |
| if hasattr( main, 'Mininet1' ): |
| main.Scapy.createHostComponent( scapyName ) |
| scapyHandle = getattr( main, scapyName ) |
| if mininetNames: |
| mininetName = mininetNames[ scapyNames.index( scapyName ) ] |
| else: |
| mininetName = None |
| scapyHandle.startHostCli( mininetName ) |
| else: |
| main.Network.createHostComponent( scapyName ) |
| scapyHandle = getattr( main, scapyName ) |
| scapyHandle.connectInband() |
| main.scapyHosts.append( scapyHandle ) |
| scapyHandle.startScapy() |
| scapyHandle.updateSelf() |
| main.log.debug( scapyHandle.name ) |
| main.log.debug( scapyHandle.hostIp ) |
| main.log.debug( scapyHandle.hostMac ) |
| |
| @staticmethod |
| def verifyTraffic( main, srcHosts, dstIp, dstHost, dstIntf, ipv6=False, expect=True, skipOnFail=True, maxRetry=2 ): |
| """ |
| Verify unicast traffic by pinging from source hosts to the destination IP |
| and capturing the packets at the destination host using Scapy. |
| srcHosts: List of host names to send the ping packets |
| dstIp: destination IP of the ping packets |
| dstHost: host that runs Scapy to capture the packets |
| dstIntf: name of the interface on the destination host |
| expect: use True if the ping is expected to be captured at destination; |
| Otherwise False |
| skipOnFail: skip the rest of this test case if result is not expected |
| maxRetry: number of retries allowed |
| """ |
| from tests.dependencies.topology import Topology |
| try: |
| main.topo |
| except ( NameError, AttributeError ): |
| main.topo = Topology() |
| main.step( "Verify traffic to {} by capturing packets on {}".format( dstIp, dstHost ) ) |
| result = main.TRUE |
| for srcHost in srcHosts: |
| trafficResult = main.topo.pingAndCapture( srcHost, dstIp, dstHost, dstIntf, ipv6, |
| expect, maxRetry, True ) |
| if not trafficResult: |
| result = main.FALSE |
| main.log.warn( "Scapy result from {} to {} is not as expected".format( srcHost, dstIp ) ) |
| utilities.assert_equals( expect=main.TRUE, |
| actual=result, |
| onpass="Verify traffic to {}: Pass".format( dstIp ), |
| onfail="Verify traffic to {}: Fail".format( dstIp ) ) |
| if skipOnFail and result != main.TRUE: |
| Testcaselib.saveOnosDiagnostics( main ) |
| Testcaselib.cleanup( main, copyKarafLog=False ) |
| main.skipCase() |
| |
| @staticmethod |
| def verifyMulticastTraffic( main, routeName, expect, skipOnFail=True, maxRetry=1 ): |
| """ |
| Verify multicast traffic using scapy |
| """ |
| from tests.dependencies.topology import Topology |
| try: |
| main.topo |
| except ( NameError, AttributeError ): |
| main.topo = Topology() |
| main.step( "Verify {} multicast traffic".format( routeName ) ) |
| routeData = main.multicastConfig[ routeName ] |
| srcs = main.mcastRoutes[ routeName ][ "src" ] |
| dsts = main.mcastRoutes[ routeName ][ "dst" ] |
| main.log.info( "Sending multicast traffic from {} to {}".format( [ routeData[ "src" ][ i ][ "host" ] for i in srcs ], |
| [ routeData[ "dst" ][ i ][ "host" ] for i in dsts ] ) ) |
| result = main.TRUE |
| for src in srcs: |
| srcEntry = routeData[ "src" ][ src ] |
| for dst in dsts: |
| dstEntry = routeData[ "dst" ][ dst ] |
| sender = getattr( main, srcEntry[ "host" ] ) |
| receiver = getattr( main, dstEntry[ "host" ] ) |
| main.Network.addRoute( str( srcEntry[ "host" ] ), |
| str( routeData[ "group" ] ), |
| str( srcEntry[ "interface" ] ), |
| True if routeData[ "ipVersion" ] == 6 else False ) |
| # Build the packet |
| sender.buildEther( dst=str( srcEntry[ "Ether" ] ) ) |
| if routeData[ "ipVersion" ] == 4: |
| sender.buildIP( dst=str( routeData[ "group" ] ) ) |
| elif routeData[ "ipVersion" ] == 6: |
| sender.buildIPv6( dst=str( routeData[ "group" ] ) ) |
| sender.buildUDP( ipVersion=routeData[ "ipVersion" ], dport=srcEntry[ "UDP" ] ) |
| sIface = srcEntry[ "interface" ] |
| dIface = dstEntry[ "interface" ] if "interface" in dstEntry.keys() else None |
| pktFilter = srcEntry[ "filter" ] |
| pkt = srcEntry[ "packet" ] |
| # Send packet and check received packet |
| expectedResult = expect.pop( 0 ) if isinstance( expect, list ) else expect |
| t3Cmd = "t3-troubleshoot -vv -sp {} -et ipv{} -d {} -dm {}".format( srcEntry[ "port" ], routeData[ "ipVersion" ], |
| routeData[ "group" ], srcEntry[ "Ether" ] ) |
| trafficResult = main.topo.sendScapyPackets( sender, receiver, pktFilter, pkt, sIface, dIface, |
| expectedResult, maxRetry, True, t3Cmd ) |
| if not trafficResult: |
| result = main.FALSE |
| main.log.warn( "Scapy result from {} to {} is not as expected".format( srcEntry[ "host" ], |
| dstEntry[ "host" ] ) ) |
| utilities.assert_equals( expect=main.TRUE, |
| actual=result, |
| onpass="Verify {} multicast traffic: Pass".format( routeName ), |
| onfail="Verify {} multicast traffic: Fail".format( routeName ) ) |
| if skipOnFail and result != main.TRUE: |
| Testcaselib.saveOnosDiagnostics( main ) |
| Testcaselib.cleanup( main, copyKarafLog=False ) |
| main.skipCase() |
| |
| @staticmethod |
| def verifyPing( main, srcList, dstList, ipv6=False, expect=True, wait=1, |
| acceptableFailed=0, skipOnFail=True, stepMsg="Verify Ping", |
| t3Simple=True ): |
| """ |
| Verify reachability from each host in srcList to each host in dstList |
| """ |
| from tests.dependencies.topology import Topology |
| try: |
| main.topo |
| except ( NameError, AttributeError ): |
| main.topo = Topology() |
| main.step( stepMsg ) |
| pingResult = main.topo.ping( srcList, dstList, ipv6, expect, wait, acceptableFailed, skipOnFail, t3Simple ) |
| utilities.assert_equals( expect=main.TRUE, |
| actual=pingResult, |
| onpass="{}: Pass".format( stepMsg ), |
| onfail="{}: Fail".format( stepMsg ) ) |
| if not pingResult and skipOnFail: |
| Testcaselib.saveOnosDiagnostics( main ) |
| Testcaselib.cleanup( main, copyKarafLog=False, removeHostComponent=True ) |
| main.skipCase() |
| |
| @staticmethod |
| def verifyHostLocations( main, locationDict, retry=2 ): |
| """ |
| Verify if the specified host is discovered by ONOS on the given locations |
| Required: |
| locationDict: a dictionary that maps host names to expected locations. |
| locations could be a string or a list. |
| ex. { "h1v4": ["of:0000000000000005/8"] } |
| Returns: |
| main.TRUE if host is discovered on all locations provided, otherwise main.FALSE |
| """ |
| main.step( "Verify locations of hosts {}".format( locationDict.keys() ) ) |
| result = main.TRUE |
| for hostName, locations in locationDict.items(): |
| main.log.info( "Verify host {} is discovered at {}".format( hostName, locations ) ) |
| hostIp = main.Network.getIPAddress( hostName, proto='IPV4' ) |
| if not hostIp: |
| hostIp = main.Network.getIPAddress( hostName, proto='IPV6' ) |
| if not hostIp: |
| main.log.warn( "Failed to find IP address for host {}, skipping location verification".format( hostName ) ) |
| result = main.FALSE |
| continue |
| locationResult = utilities.retry( main.Cluster.active( 0 ).CLI.verifyHostLocation, |
| main.FALSE, |
| args=( hostIp, locations ), |
| attempts=retry + 1, |
| sleep=10 ) |
| if not locationResult: |
| result = main.FALSE |
| main.log.warn( "location verification for host {} failed".format( hostName ) ) |
| utilities.assert_equals( expect=main.TRUE, actual=result, |
| onpass="Location verification passed", |
| onfail="Location verification failed" ) |
| |
| @staticmethod |
| def moveHost( main, hostName, srcSw, dstSw, gw, macAddr=None, prefixLen=None, cfg='', ipv6=False, vlan=None ): |
| """ |
| Move specified host from srcSw to dstSw. |
| If srcSw and dstSw are same, the host will be moved from current port to |
| next available port. |
| Required: |
| hostName: name of the host. e.g., "h1" |
| srcSw: name of the switch that the host is attached to. e.g., "leaf1" |
| dstSw: name of the switch that the host will be moved to. e.g., "leaf2" |
| gw: ip address of the gateway of the new location |
| Optional: |
| macAddr: if specified, change MAC address of the host to the specified MAC address. |
| prefixLen: prefix length |
| cfg: port configuration as JSON string |
| ipv6: Use True to move IPv6 host |
| vlan: vlan number of the host |
| """ |
| if not hasattr( main, 'Mininet1' ): |
| main.log.warn( "moveHost is supposed to be used only in Mininet." ) |
| return |
| |
| main.step( "Moving {} host {} from {} to {}".format( 'tagged' if vlan else 'untagged', hostName, srcSw, dstSw ) ) |
| main.Mininet1.moveHost( hostName, srcSw, dstSw, macAddr, prefixLen, ipv6, vlan=vlan ) |
| if not ipv6: |
| main.Mininet1.changeDefaultGateway( hostName, gw ) |
| if cfg: |
| main.Cluster.active( 0 ).REST.setNetCfg( json.loads( cfg ), |
| subjectClass="ports" ) |
| # Wait for the host to get RA for setting up default gateway |
| time.sleep( 5 ) |
| |
| main.Mininet1.discoverHosts( [ hostName, ] ) |
| |
| # Update expectedHost when MAC address is changed. |
| if macAddr is not None: |
| ipAddr = main.expectedHosts[ "network" ][ hostName ] |
| if ipAddr is not None: |
| for hostName, ip in main.expectedHosts[ "onos" ].items(): |
| if ip == ipAddr: |
| vlan = hostName.split( "/" )[ -1 ] |
| del main.expectedHosts[ "onos" ][ hostName ] |
| main.expectedHosts[ "onos" ][ "{}/{}".format( macAddr.upper(), vlan ) ] = ip |
| break |
| |
| @staticmethod |
| def moveDualHomedHost( main, hostName, srcSw, srcPairSw, dstSw, dstPairSw, gw, |
| macAddr=None, prefixLen=24, cfg='', ipv6=False, vlan=None ): |
| """ |
| Move specified dual-homed host from srcSw-srcPairSw to dstSw-dstPairSw. |
| If srcSw-srcPairSw and dstSw-dstPairSw are same, the host will be moved from current port |
| to next available port. |
| Required: |
| hostName: name of the host. e.g., "h1" |
| srcSw: name of the switch that the host is attached to. e.g., "leaf1" |
| srcPairSw: name of the paired-switch that the host is attached to. e.g., "leaf2" |
| dstSw: name of the switch that the host will be moved to. e.g., "leaf1" |
| dstPairSw: name of the paired-switch that the host will be moved to. e.g., "leaf2" |
| gw: ip address of the gateway of the new location |
| Optional: |
| macAddr: if specified, change MAC address of the host to the specified MAC address. |
| prefixLen: prefix length |
| cfg: port configurations as JSON string |
| ipv6: Use True to move IPv6 host |
| vlan: vlan number of the host |
| """ |
| if not hasattr( main, 'Mininet1' ): |
| main.log.warn( "moveDualHomedHost is supposed to be used only in Mininet." ) |
| return |
| |
| main.step( "Moving {} host {} from {} and {} to {} and {}".format( 'tagged' if vlan else 'untagged', hostName, |
| srcSw, srcPairSw, dstSw, dstPairSw ) ) |
| main.Mininet1.moveDualHomedHost( hostName, srcSw, srcPairSw, dstSw, dstPairSw, |
| macAddr=macAddr, prefixLen=prefixLen, ipv6=ipv6, vlan=vlan ) |
| if not ipv6: |
| main.Mininet1.changeDefaultGateway( hostName, gw ) |
| if cfg: |
| main.Cluster.active( 0 ).REST.setNetCfg( json.loads( cfg ), |
| subjectClass="ports" ) |
| # Wait for the host to get RA for setting up default gateway |
| time.sleep( 5 ) |
| |
| main.Mininet1.discoverHosts( [ hostName, ] ) |
| |
| # Update expectedHost when MAC address is changed. |
| if macAddr is not None: |
| ipAddr = main.expectedHosts[ "network" ][ hostName ] |
| if ipAddr is not None: |
| for hostName, ip in main.expectedHosts[ "onos" ].items(): |
| if ip == ipAddr: |
| vlan = hostName.split( "/" )[ -1 ] |
| del main.expectedHosts[ "onos" ][ hostName ] |
| main.expectedHosts[ "onos" ][ "{}/{}".format( macAddr.upper(), vlan ) ] = ip |