Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 1 | """ |
Jeremy Ronquillo | b27ce4c | 2017-07-17 12:41:28 -0700 | [diff] [blame^] | 2 | Copyright 2017 Open Networking Foundation (ONF) |
| 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 |
| 11 | (at your option) any later version. |
| 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 | """ |
| 21 | |
| 22 | """ |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 23 | Description: This test is to determine if ONOS can handle |
| 24 | a minority of it's nodes restarting |
| 25 | |
| 26 | List of test cases: |
| 27 | CASE1: Compile ONOS and push it to the test machines |
| 28 | CASE2: Assign devices to controllers |
| 29 | CASE21: Assign mastership to controllers |
| 30 | CASE3: Assign intents |
| 31 | CASE4: Ping across added host intents |
| 32 | CASE5: Reading state of ONOS |
| 33 | CASE61: The Failure inducing case. |
| 34 | CASE62: The Failure recovery case. |
| 35 | CASE7: Check state after control plane failure |
| 36 | CASE8: Compare topo |
| 37 | CASE9: Link s3-s28 down |
| 38 | CASE10: Link s3-s28 up |
| 39 | CASE11: Switch down |
| 40 | CASE12: Switch up |
| 41 | CASE13: Clean up |
| 42 | CASE14: start election app on all onos nodes |
| 43 | CASE15: Check that Leadership Election is still functional |
| 44 | CASE16: Install Distributed Primitives app |
| 45 | CASE17: Check for basic functionality with distributed primitives |
| 46 | """ |
| 47 | class HAcontinuousStopNodes: |
| 48 | |
| 49 | def __init__( self ): |
| 50 | self.default = '' |
| 51 | |
| 52 | def CASE1( self, main ): |
| 53 | """ |
| 54 | CASE1 is to compile ONOS and push it to the test machines |
| 55 | |
| 56 | Startup sequence: |
| 57 | cell <name> |
| 58 | onos-verify-cell |
| 59 | NOTE: temporary - onos-remove-raft-logs |
| 60 | onos-uninstall |
| 61 | start mininet |
| 62 | git pull |
| 63 | mvn clean install |
| 64 | onos-package |
| 65 | onos-install -f |
| 66 | onos-wait-for-start |
| 67 | start cli sessions |
| 68 | start tcpdump |
| 69 | """ |
| 70 | import imp |
| 71 | import pexpect |
| 72 | import time |
| 73 | import json |
| 74 | main.log.info( "ONOS HA test: Stop a minority of ONOS nodes - " + |
| 75 | "initialization" ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 76 | # set global variables |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 77 | # These are for csv plotting in jenkins |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 78 | main.HAlabels = [] |
| 79 | main.HAdata = [] |
| 80 | try: |
| 81 | from tests.dependencies.ONOSSetup import ONOSSetup |
| 82 | main.testSetUp = ONOSSetup() |
| 83 | except ImportError: |
| 84 | main.log.error( "ONOSSetup not found. exiting the test" ) |
| 85 | main.exit() |
| 86 | main.testSetUp.envSetupDescription() |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 87 | try: |
| 88 | from tests.HA.dependencies.HA import HA |
| 89 | main.HA = HA() |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 90 | # load some variables from the params file |
| 91 | cellName = main.params[ 'ENV' ][ 'cellName' ] |
| 92 | main.apps = main.params[ 'ENV' ][ 'appString' ] |
| 93 | main.numCtrls = int( main.params[ 'num_controllers' ] ) |
| 94 | if main.ONOSbench.maxNodes and\ |
| 95 | main.ONOSbench.maxNodes < main.numCtrls: |
| 96 | main.numCtrls = int( main.ONOSbench.maxNodes ) |
| 97 | main.maxNodes = main.numCtrls |
| 98 | stepResult = main.testSetUp.envSetup( hasNode=True ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 99 | except Exception as e: |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 100 | main.testSetUp.envSetupException( e ) |
| 101 | main.testSetUp.evnSetupConclusion( stepResult ) |
| 102 | main.HA.generateGraph( "HAcontinuousStopNodes" ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 103 | |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 104 | main.testSetUp.ONOSSetUp( main.Mininet1, cellName=cellName, removeLog=True, |
| 105 | extraApply=main.HA.customizeOnosGenPartitions, |
| 106 | extraClean=main.HA.cleanUpGenPartition ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 107 | |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 108 | main.HA.initialSetUp() |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 109 | |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 110 | |
| 111 | def CASE2( self, main ): |
| 112 | """ |
| 113 | Assign devices to controllers |
| 114 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 115 | main.HA.assignDevices( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 116 | |
| 117 | def CASE21( self, main ): |
| 118 | """ |
| 119 | Assign mastership to controllers |
| 120 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 121 | main.HA.assignMastership( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 122 | |
| 123 | def CASE3( self, main ): |
| 124 | """ |
| 125 | Assign intents |
| 126 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 127 | main.HA.assignIntents( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 128 | |
| 129 | def CASE4( self, main ): |
| 130 | """ |
| 131 | Ping across added host intents |
| 132 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 133 | main.HA.pingAcrossHostIntent( main, True, False ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 134 | |
| 135 | def CASE5( self, main ): |
| 136 | """ |
| 137 | Reading state of ONOS |
| 138 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 139 | main.HA.readingState( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 140 | |
| 141 | def CASE61( self, main ): |
| 142 | """ |
| 143 | The Failure case. |
| 144 | """ |
| 145 | assert main.numCtrls, "main.numCtrls not defined" |
| 146 | assert main, "main not defined" |
| 147 | assert utilities.assert_equals, "utilities.assert_equals not defined" |
| 148 | assert main.CLIs, "main.CLIs not defined" |
| 149 | assert main.nodes, "main.nodes not defined" |
| 150 | try: |
| 151 | assert main.nodeIndex is not None, "main.nodeIndex not defined" |
| 152 | assert main.killCount is not None, "main.killCount not defined" |
| 153 | except AttributeError as e: |
| 154 | main.log.warn( "Node to kill not selected, defaulting to node 1" ) |
| 155 | main.nodeIndex = 0 |
| 156 | main.killCount = 1 |
| 157 | |
| 158 | main.case( "Stopping ONOS nodes - iteration " + str( main.killCount ) ) |
| 159 | |
| 160 | main.step( "Checking ONOS Logs for errors" ) |
| 161 | for node in main.nodes: |
| 162 | main.log.debug( "Checking logs for errors on " + node.name + ":" ) |
| 163 | main.log.warn( main.ONOSbench.checkLogs( node.ip_address ) ) |
| 164 | |
| 165 | # NOTE: For now only kill one node. If we move to killing more, we need to |
| 166 | # make sure we don't lose any partitions |
| 167 | n = len( main.nodes ) # Number of nodes |
| 168 | main.nodeIndex = ( main.nodeIndex + 1 ) % n |
| 169 | main.kill = [ main.nodeIndex ] # ONOS node to kill, listed by index in main.nodes |
| 170 | |
| 171 | # TODO: Be able to configure bringing up old node vs. a new/fresh installation |
| 172 | main.step( "Stopping " + str( len( main.kill ) ) + " ONOS nodes" ) |
| 173 | killResults = main.TRUE |
| 174 | for i in main.kill: |
| 175 | killResults = killResults and\ |
| 176 | main.ONOSbench.onosStop( main.nodes[ i ].ip_address ) |
| 177 | main.activeNodes.remove( i ) |
| 178 | utilities.assert_equals( expect=main.TRUE, actual=killResults, |
| 179 | onpass="ONOS nodes stopped successfully", |
| 180 | onfail="ONOS nodes NOT successfully stopped" ) |
| 181 | |
| 182 | main.step( "Checking ONOS nodes" ) |
| 183 | nodeResults = utilities.retry( main.HA.nodesCheck, |
| 184 | False, |
| 185 | args=[ main.activeNodes ], |
| 186 | sleep=15, |
| 187 | attempts=5 ) |
| 188 | |
| 189 | utilities.assert_equals( expect=True, actual=nodeResults, |
| 190 | onpass="Nodes check successful", |
| 191 | onfail="Nodes check NOT successful" ) |
| 192 | |
| 193 | if not nodeResults: |
| 194 | for i in main.activeNodes: |
| 195 | cli = main.CLIs[ i ] |
| 196 | main.log.debug( "{} components not ACTIVE: \n{}".format( |
| 197 | cli.name, |
| 198 | cli.sendline( "scr:list | grep -v ACTIVE" ) ) ) |
| 199 | main.log.error( "Failed to start ONOS, stopping test" ) |
| 200 | main.cleanup() |
| 201 | main.exit() |
| 202 | |
| 203 | main.killCount += 1 |
| 204 | |
| 205 | def CASE62( self, main ): |
| 206 | """ |
| 207 | The bring up stopped nodes |
| 208 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 209 | main.HA.bringUpStoppedNode( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 210 | |
| 211 | def CASE7( self, main ): |
| 212 | """ |
| 213 | Check state after ONOS failure |
| 214 | """ |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 215 | try: |
| 216 | main.kill |
| 217 | except AttributeError: |
| 218 | main.kill = [] |
| 219 | |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 220 | main.HA.checkStateAfterONOS( main, afterWhich=0 ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 221 | |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 222 | main.step( "Leadership Election is still functional" ) |
| 223 | # Test of LeadershipElection |
| 224 | leaderList = [] |
| 225 | |
| 226 | restarted = [] |
| 227 | for i in main.kill: |
| 228 | restarted.append( main.nodes[ i ].ip_address ) |
| 229 | leaderResult = main.TRUE |
| 230 | |
| 231 | for i in main.activeNodes: |
| 232 | cli = main.CLIs[ i ] |
| 233 | leaderN = cli.electionTestLeader() |
| 234 | leaderList.append( leaderN ) |
| 235 | if leaderN == main.FALSE: |
| 236 | # error in response |
| 237 | main.log.error( "Something is wrong with " + |
| 238 | "electionTestLeader function, check the" + |
| 239 | " error logs" ) |
| 240 | leaderResult = main.FALSE |
| 241 | elif leaderN is None: |
| 242 | main.log.error( cli.name + |
| 243 | " shows no leader for the election-app was" + |
| 244 | " elected after the old one died" ) |
| 245 | leaderResult = main.FALSE |
| 246 | elif leaderN in restarted: |
| 247 | main.log.error( cli.name + " shows " + str( leaderN ) + |
| 248 | " as leader for the election-app, but it " + |
| 249 | "was restarted" ) |
| 250 | leaderResult = main.FALSE |
| 251 | if len( set( leaderList ) ) != 1: |
| 252 | leaderResult = main.FALSE |
| 253 | main.log.error( |
| 254 | "Inconsistent view of leader for the election test app" ) |
| 255 | # TODO: print the list |
| 256 | utilities.assert_equals( |
| 257 | expect=main.TRUE, |
| 258 | actual=leaderResult, |
| 259 | onpass="Leadership election passed", |
| 260 | onfail="Something went wrong with Leadership election" ) |
| 261 | |
| 262 | def CASE8( self, main ): |
| 263 | """ |
| 264 | Compare topo |
| 265 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 266 | main.HA.compareTopo( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 267 | |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 268 | |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 269 | |
| 270 | def CASE9( self, main ): |
| 271 | """ |
| 272 | Link s3-s28 down |
| 273 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 274 | main.HA.linkDown( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 275 | |
| 276 | def CASE10( self, main ): |
| 277 | """ |
| 278 | Link s3-s28 up |
| 279 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 280 | main.HA.linkUp( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 281 | |
| 282 | def CASE11( self, main ): |
| 283 | """ |
| 284 | Switch Down |
| 285 | """ |
| 286 | # NOTE: You should probably run a topology check after this |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 287 | main.HA.switchDown( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 288 | |
| 289 | def CASE12( self, main ): |
| 290 | """ |
| 291 | Switch Up |
| 292 | """ |
| 293 | # NOTE: You should probably run a topology check after this |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 294 | main.HA.switchUp( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 295 | |
| 296 | def CASE13( self, main ): |
| 297 | """ |
| 298 | Clean up |
| 299 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 300 | main.HAlabels.append( "Restart" ) |
| 301 | main.HAdata.append( str( main.restartTime ) ) |
| 302 | main.HA.cleanUp( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 303 | |
| 304 | def CASE14( self, main ): |
| 305 | """ |
| 306 | start election app on all onos nodes |
| 307 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 308 | main.HA.startElectionApp( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 309 | |
| 310 | def CASE15( self, main ): |
| 311 | """ |
| 312 | Check that Leadership Election is still functional |
| 313 | 15.1 Run election on each node |
| 314 | 15.2 Check that each node has the same leaders and candidates |
| 315 | 15.3 Find current leader and withdraw |
| 316 | 15.4 Check that a new node was elected leader |
| 317 | 15.5 Check that that new leader was the candidate of old leader |
| 318 | 15.6 Run for election on old leader |
| 319 | 15.7 Check that oldLeader is a candidate, and leader if only 1 node |
| 320 | 15.8 Make sure that the old leader was added to the candidate list |
| 321 | |
| 322 | old and new variable prefixes refer to data from before vs after |
| 323 | withdrawl and later before withdrawl vs after re-election |
| 324 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 325 | main.HA.isElectionFunctional( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 326 | |
| 327 | def CASE16( self, main ): |
| 328 | """ |
| 329 | Install Distributed Primitives app |
| 330 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 331 | main.HA.installDistributedPrimitiveApp( main ) |
Jon Hall | b548801 | 2017-06-21 14:08:36 -0700 | [diff] [blame] | 332 | |
| 333 | def CASE17( self, main ): |
| 334 | """ |
| 335 | Check for basic functionality with distributed primitives |
| 336 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 337 | main.HA.checkDistPrimitivesFunc( main ) |