Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 1 | """ |
Jeremy Ronquillo | 23fb216 | 2017-09-15 14:59:57 -0700 | [diff] [blame] | 2 | Copyright 2017 Open Networking Foundation ( ONF ) |
Jeremy Ronquillo | b27ce4c | 2017-07-17 12:41:28 -0700 | [diff] [blame] | 3 | |
| 4 | Please refer questions to either the onos test mailing list at <onos-test@onosproject.org>, |
| 5 | the System Testing Plans and Results wiki page at <https://wiki.onosproject.org/x/voMg>, |
| 6 | or the System Testing Guide page at <https://wiki.onosproject.org/x/WYQg> |
| 7 | |
| 8 | TestON is free software: you can redistribute it and/or modify |
| 9 | it under the terms of the GNU General Public License as published by |
| 10 | the Free Software Foundation, either version 2 of the License, or |
Jeremy Ronquillo | 23fb216 | 2017-09-15 14:59:57 -0700 | [diff] [blame] | 11 | ( at your option ) any later version. |
Jeremy Ronquillo | b27ce4c | 2017-07-17 12:41:28 -0700 | [diff] [blame] | 12 | |
| 13 | TestON is distributed in the hope that it will be useful, |
| 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 16 | GNU General Public License for more details. |
| 17 | |
| 18 | You should have received a copy of the GNU General Public License |
| 19 | along with TestON. If not, see <http://www.gnu.org/licenses/>. |
| 20 | """ |
Jeremy Ronquillo | b27ce4c | 2017-07-17 12:41:28 -0700 | [diff] [blame] | 21 | """ |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 22 | Description: This test is to determine if ONOS can handle |
| 23 | a minority of it's nodes restarting |
| 24 | |
| 25 | List of test cases: |
| 26 | CASE1: Compile ONOS and push it to the test machines |
| 27 | CASE2: Assign devices to controllers |
| 28 | CASE21: Assign mastership to controllers |
| 29 | CASE3: Assign intents |
| 30 | CASE4: Ping across added host intents |
| 31 | CASE5: Reading state of ONOS |
| 32 | CASE61: The Failure inducing case. |
| 33 | CASE62: The Failure recovery case. |
| 34 | CASE7: Check state after control plane failure |
| 35 | CASE8: Compare topo |
| 36 | CASE9: Link s3-s28 down |
| 37 | CASE10: Link s3-s28 up |
| 38 | CASE11: Switch down |
| 39 | CASE12: Switch up |
| 40 | CASE13: Clean up |
| 41 | CASE14: start election app on all onos nodes |
| 42 | CASE15: Check that Leadership Election is still functional |
| 43 | CASE16: Install Distributed Primitives app |
| 44 | CASE17: Check for basic functionality with distributed primitives |
| 45 | """ |
| 46 | class HAcontinuousStopNodes: |
| 47 | |
| 48 | def __init__( self ): |
| 49 | self.default = '' |
| 50 | |
| 51 | def CASE1( self, main ): |
| 52 | """ |
| 53 | CASE1 is to compile ONOS and push it to the test machines |
| 54 | |
| 55 | Startup sequence: |
| 56 | cell <name> |
| 57 | onos-verify-cell |
| 58 | NOTE: temporary - onos-remove-raft-logs |
| 59 | onos-uninstall |
| 60 | start mininet |
| 61 | git pull |
| 62 | mvn clean install |
| 63 | onos-package |
| 64 | onos-install -f |
| 65 | onos-wait-for-start |
| 66 | start cli sessions |
| 67 | start tcpdump |
| 68 | """ |
| 69 | import imp |
| 70 | import pexpect |
| 71 | import time |
| 72 | import json |
| 73 | main.log.info( "ONOS HA test: Stop a minority of ONOS nodes - " + |
| 74 | "initialization" ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 75 | # set global variables |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 76 | # These are for csv plotting in jenkins |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 77 | main.HAlabels = [] |
| 78 | main.HAdata = [] |
| 79 | try: |
| 80 | from tests.dependencies.ONOSSetup import ONOSSetup |
| 81 | main.testSetUp = ONOSSetup() |
| 82 | except ImportError: |
| 83 | main.log.error( "ONOSSetup not found. exiting the test" ) |
Devin Lim | 4407596 | 2017-08-11 10:56:37 -0700 | [diff] [blame] | 84 | main.cleanAndExit() |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 85 | main.testSetUp.envSetupDescription() |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 86 | try: |
| 87 | from tests.HA.dependencies.HA import HA |
| 88 | main.HA = HA() |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 89 | # load some variables from the params file |
| 90 | cellName = main.params[ 'ENV' ][ 'cellName' ] |
| 91 | main.apps = main.params[ 'ENV' ][ 'appString' ] |
| 92 | main.numCtrls = int( main.params[ 'num_controllers' ] ) |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 93 | stepResult = main.testSetUp.envSetup() |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 94 | except Exception as e: |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 95 | main.testSetUp.envSetupException( e ) |
| 96 | main.testSetUp.evnSetupConclusion( stepResult ) |
| 97 | main.HA.generateGraph( "HAcontinuousStopNodes" ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 98 | |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 99 | main.testSetUp.ONOSSetUp( main.Mininet1, main.Cluster, cellName=cellName, removeLog=True, |
| 100 | extraApply=[ main.HA.startingMininet, |
| 101 | main.HA.customizeOnosGenPartitions ], |
| 102 | extraClean=main.HA.cleanUpGenPartition ) |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 103 | main.HA.initialSetUp() |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 104 | |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 105 | def CASE2( self, main ): |
| 106 | """ |
| 107 | Assign devices to controllers |
| 108 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 109 | main.HA.assignDevices( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 110 | |
| 111 | def CASE21( self, main ): |
| 112 | """ |
| 113 | Assign mastership to controllers |
| 114 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 115 | main.HA.assignMastership( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 116 | |
| 117 | def CASE3( self, main ): |
| 118 | """ |
| 119 | Assign intents |
| 120 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 121 | main.HA.assignIntents( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 122 | |
| 123 | def CASE4( self, main ): |
| 124 | """ |
| 125 | Ping across added host intents |
| 126 | """ |
Jon Hall | ca31989 | 2017-06-15 15:25:22 -0700 | [diff] [blame] | 127 | main.HA.pingAcrossHostIntent( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 128 | |
| 129 | def CASE5( self, main ): |
| 130 | """ |
| 131 | Reading state of ONOS |
| 132 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 133 | main.HA.readingState( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 134 | |
| 135 | def CASE61( self, main ): |
| 136 | """ |
| 137 | The Failure case. |
| 138 | """ |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 139 | assert main, "main not defined" |
| 140 | assert utilities.assert_equals, "utilities.assert_equals not defined" |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 141 | try: |
| 142 | assert main.nodeIndex is not None, "main.nodeIndex not defined" |
| 143 | assert main.killCount is not None, "main.killCount not defined" |
| 144 | except AttributeError as e: |
| 145 | main.log.warn( "Node to kill not selected, defaulting to node 1" ) |
| 146 | main.nodeIndex = 0 |
| 147 | main.killCount = 1 |
| 148 | |
| 149 | main.case( "Stopping ONOS nodes - iteration " + str( main.killCount ) ) |
| 150 | |
| 151 | main.step( "Checking ONOS Logs for errors" ) |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 152 | for ctrl in main.Cluster.runningNodes: |
| 153 | main.log.debug( "Checking logs for errors on " + ctrl.name + ":" ) |
| 154 | main.log.warn( main.ONOSbench.checkLogs( ctrl.ipAddress ) ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 155 | |
| 156 | # NOTE: For now only kill one node. If we move to killing more, we need to |
| 157 | # make sure we don't lose any partitions |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 158 | n = len( main.Cluster.runningNodes ) # Number of nodes |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 159 | main.nodeIndex = ( main.nodeIndex + 1 ) % n |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 160 | main.kill = [ main.Cluster.runningNodes[ main.nodeIndex ] ] # ONOS node to kill, listed by index in main.nodes |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 161 | |
| 162 | # TODO: Be able to configure bringing up old node vs. a new/fresh installation |
| 163 | main.step( "Stopping " + str( len( main.kill ) ) + " ONOS nodes" ) |
| 164 | killResults = main.TRUE |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 165 | for ctrl in main.kill: |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 166 | killResults = killResults and\ |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 167 | main.ONOSbench.onosStop( ctrl.ipAddress ) |
| 168 | ctrl.active = False |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 169 | utilities.assert_equals( expect=main.TRUE, actual=killResults, |
| 170 | onpass="ONOS nodes stopped successfully", |
| 171 | onfail="ONOS nodes NOT successfully stopped" ) |
| 172 | |
| 173 | main.step( "Checking ONOS nodes" ) |
| 174 | nodeResults = utilities.retry( main.HA.nodesCheck, |
| 175 | False, |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 176 | args=[ main.Cluster.active() ], |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 177 | sleep=15, |
| 178 | attempts=5 ) |
| 179 | |
| 180 | utilities.assert_equals( expect=True, actual=nodeResults, |
| 181 | onpass="Nodes check successful", |
| 182 | onfail="Nodes check NOT successful" ) |
| 183 | |
| 184 | if not nodeResults: |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 185 | for ctrl in main.Cluster.active(): |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 186 | main.log.debug( "{} components not ACTIVE: \n{}".format( |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 187 | ctrl.name, |
| 188 | ctrl.CLI.sendline( "scr:list | grep -v ACTIVE" ) ) ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 189 | main.log.error( "Failed to start ONOS, stopping test" ) |
Devin Lim | 4407596 | 2017-08-11 10:56:37 -0700 | [diff] [blame] | 190 | main.cleanAndExit() |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 191 | |
| 192 | main.killCount += 1 |
| 193 | |
| 194 | def CASE62( self, main ): |
| 195 | """ |
| 196 | The bring up stopped nodes |
| 197 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 198 | main.HA.bringUpStoppedNode( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 199 | |
| 200 | def CASE7( self, main ): |
| 201 | """ |
| 202 | Check state after ONOS failure |
| 203 | """ |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 204 | try: |
| 205 | main.kill |
| 206 | except AttributeError: |
| 207 | main.kill = [] |
| 208 | |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 209 | main.HA.checkStateAfterEvent( main, afterWhich=0 ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 210 | |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 211 | main.step( "Leadership Election is still functional" ) |
| 212 | # Test of LeadershipElection |
| 213 | leaderList = [] |
| 214 | |
| 215 | restarted = [] |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 216 | for ctrl in main.kill: |
| 217 | restarted.append( ctrl.ipAddress ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 218 | leaderResult = main.TRUE |
| 219 | |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 220 | for ctrl in main.Cluster.active(): |
| 221 | leaderN = ctrl.CLI.electionTestLeader() |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 222 | leaderList.append( leaderN ) |
| 223 | if leaderN == main.FALSE: |
| 224 | # error in response |
| 225 | main.log.error( "Something is wrong with " + |
| 226 | "electionTestLeader function, check the" + |
| 227 | " error logs" ) |
| 228 | leaderResult = main.FALSE |
| 229 | elif leaderN is None: |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 230 | main.log.error( ctrl.name + |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 231 | " shows no leader for the election-app was" + |
| 232 | " elected after the old one died" ) |
| 233 | leaderResult = main.FALSE |
| 234 | elif leaderN in restarted: |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 235 | main.log.error( ctrl.name + " shows " + str( leaderN ) + |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 236 | " as leader for the election-app, but it " + |
| 237 | "was restarted" ) |
| 238 | leaderResult = main.FALSE |
| 239 | if len( set( leaderList ) ) != 1: |
| 240 | leaderResult = main.FALSE |
| 241 | main.log.error( |
| 242 | "Inconsistent view of leader for the election test app" ) |
| 243 | # TODO: print the list |
| 244 | utilities.assert_equals( |
| 245 | expect=main.TRUE, |
| 246 | actual=leaderResult, |
| 247 | onpass="Leadership election passed", |
| 248 | onfail="Something went wrong with Leadership election" ) |
| 249 | |
| 250 | def CASE8( self, main ): |
| 251 | """ |
| 252 | Compare topo |
| 253 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 254 | main.HA.compareTopo( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 255 | |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 256 | def CASE9( self, main ): |
| 257 | """ |
| 258 | Link s3-s28 down |
| 259 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 260 | main.HA.linkDown( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 261 | |
| 262 | def CASE10( self, main ): |
| 263 | """ |
| 264 | Link s3-s28 up |
| 265 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 266 | main.HA.linkUp( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 267 | |
| 268 | def CASE11( self, main ): |
| 269 | """ |
| 270 | Switch Down |
| 271 | """ |
| 272 | # NOTE: You should probably run a topology check after this |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 273 | main.HA.switchDown( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 274 | |
| 275 | def CASE12( self, main ): |
| 276 | """ |
| 277 | Switch Up |
| 278 | """ |
| 279 | # NOTE: You should probably run a topology check after this |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 280 | main.HA.switchUp( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 281 | |
| 282 | def CASE13( self, main ): |
| 283 | """ |
| 284 | Clean up |
| 285 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 286 | main.HAlabels.append( "Restart" ) |
| 287 | main.HAdata.append( str( main.restartTime ) ) |
| 288 | main.HA.cleanUp( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 289 | |
| 290 | def CASE14( self, main ): |
| 291 | """ |
| 292 | start election app on all onos nodes |
| 293 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 294 | main.HA.startElectionApp( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 295 | |
| 296 | def CASE15( self, main ): |
| 297 | """ |
| 298 | Check that Leadership Election is still functional |
| 299 | 15.1 Run election on each node |
| 300 | 15.2 Check that each node has the same leaders and candidates |
| 301 | 15.3 Find current leader and withdraw |
| 302 | 15.4 Check that a new node was elected leader |
| 303 | 15.5 Check that that new leader was the candidate of old leader |
| 304 | 15.6 Run for election on old leader |
| 305 | 15.7 Check that oldLeader is a candidate, and leader if only 1 node |
| 306 | 15.8 Make sure that the old leader was added to the candidate list |
| 307 | |
| 308 | old and new variable prefixes refer to data from before vs after |
| 309 | withdrawl and later before withdrawl vs after re-election |
| 310 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 311 | main.HA.isElectionFunctional( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 312 | |
| 313 | def CASE16( self, main ): |
| 314 | """ |
| 315 | Install Distributed Primitives app |
| 316 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 317 | main.HA.installDistributedPrimitiveApp( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 318 | |
| 319 | def CASE17( self, main ): |
| 320 | """ |
| 321 | Check for basic functionality with distributed primitives |
| 322 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 323 | main.HA.checkDistPrimitivesFunc( main ) |