Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 1 | """ |
Jeremy Ronquillo | b27ce4c | 2017-07-17 12:41:28 -0700 | [diff] [blame] | 2 | Copyright 2015 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 | 5cf14d5 | 2015-07-16 12:15:19 -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 |
Jon Hall | b3ed8ed | 2015-10-28 16:43:55 -0700 | [diff] [blame] | 33 | CASE61: The Failure inducing case. |
| 34 | CASE62: The Failure recovery case. |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 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 | """ |
Jon Hall | b3ed8ed | 2015-10-28 16:43:55 -0700 | [diff] [blame] | 47 | class HAkillNodes: |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 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 | """ |
Jon Hall | e1a3b75 | 2015-07-22 13:02:46 -0700 | [diff] [blame] | 70 | import imp |
Jon Hall | 3b489db | 2015-10-05 14:38:37 -0700 | [diff] [blame] | 71 | import pexpect |
Jon Hall | 6e70975 | 2016-02-01 13:38:46 -0800 | [diff] [blame] | 72 | import time |
Jon Hall | a440e87 | 2016-03-31 15:15:50 -0700 | [diff] [blame] | 73 | import json |
Jon Hall | 6e70975 | 2016-02-01 13:38:46 -0800 | [diff] [blame] | 74 | main.log.info( "ONOS HA test: Restart a minority of ONOS nodes - " + |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 75 | "initialization" ) |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 76 | # set global variables |
Jon Hall | a440e87 | 2016-03-31 15:15:50 -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 | e1a3b75 | 2015-07-22 13:02:46 -0700 | [diff] [blame] | 87 | try: |
Jon Hall | 53c5e66 | 2016-04-13 16:06:56 -0700 | [diff] [blame] | 88 | from tests.HA.dependencies.HA import HA |
Jon Hall | 41d39f1 | 2016-04-11 22:54:35 -0700 | [diff] [blame] | 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 | e1a3b75 | 2015-07-22 13:02:46 -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( "HAkillNodes" ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 103 | |
Jon Hall | b3ed8ed | 2015-10-28 16:43:55 -0700 | [diff] [blame] | 104 | main.step( "Make sure ONOS service doesn't automatically respawn" ) |
| 105 | handle = main.ONOSbench.handle |
| 106 | handle.sendline( "sed -i -e 's/^respawn$/#respawn/g' tools/package/init/onos.conf" ) |
| 107 | handle.expect( "\$" ) # $ from the command |
Jon Hall | 334ba94 | 2017-01-19 17:02:17 -0800 | [diff] [blame] | 108 | handle.sendline( "sed -i -e 's/^Restart=always/Restart=no/g' tools/package/init/onos.service" ) |
| 109 | handle.expect( "\$" ) # $ from the command |
Jon Hall | b3ed8ed | 2015-10-28 16:43:55 -0700 | [diff] [blame] | 110 | handle.expect( "\$" ) # $ from the prompt |
| 111 | |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 112 | main.testSetUp.ONOSSetUp( main.Mininet1, cellName=cellName, removeLog=True, |
| 113 | extraApply=main.HA.customizeOnosGenPartitions, |
| 114 | extraClean=main.HA.cleanUpGenPartition ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 115 | |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 116 | main.HA.initialSetUp() |
Jon Hall | 9d2dcad | 2016-04-08 10:15:20 -0700 | [diff] [blame] | 117 | |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 118 | def CASE2( self, main ): |
| 119 | """ |
| 120 | Assign devices to controllers |
| 121 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 122 | main.HA.assignDevices( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 123 | |
| 124 | def CASE21( self, main ): |
| 125 | """ |
| 126 | Assign mastership to controllers |
| 127 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 128 | main.HA.assignMastership( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 129 | |
| 130 | def CASE3( self, main ): |
| 131 | """ |
| 132 | Assign intents |
| 133 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 134 | main.HA.assignIntents( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 135 | |
| 136 | def CASE4( self, main ): |
| 137 | """ |
| 138 | Ping across added host intents |
| 139 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 140 | main.HA.pingAcrossHostIntent( main, True, False ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 141 | |
| 142 | def CASE5( self, main ): |
| 143 | """ |
| 144 | Reading state of ONOS |
| 145 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 146 | main.HA.readingState( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 147 | |
Jon Hall | b3ed8ed | 2015-10-28 16:43:55 -0700 | [diff] [blame] | 148 | def CASE61( self, main ): |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 149 | """ |
| 150 | The Failure case. |
| 151 | """ |
Jon Hall | e1a3b75 | 2015-07-22 13:02:46 -0700 | [diff] [blame] | 152 | assert main.numCtrls, "main.numCtrls not defined" |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 153 | assert main, "main not defined" |
| 154 | assert utilities.assert_equals, "utilities.assert_equals not defined" |
Jon Hall | e1a3b75 | 2015-07-22 13:02:46 -0700 | [diff] [blame] | 155 | assert main.CLIs, "main.CLIs not defined" |
| 156 | assert main.nodes, "main.nodes not defined" |
Jon Hall | b3ed8ed | 2015-10-28 16:43:55 -0700 | [diff] [blame] | 157 | main.case( "Kill minority of ONOS nodes" ) |
Jon Hall | 96091e6 | 2015-09-21 17:34:17 -0700 | [diff] [blame] | 158 | |
| 159 | main.step( "Checking ONOS Logs for errors" ) |
| 160 | for node in main.nodes: |
| 161 | main.log.debug( "Checking logs for errors on " + node.name + ":" ) |
| 162 | main.log.warn( main.ONOSbench.checkLogs( node.ip_address ) ) |
| 163 | |
Jon Hall | 3b489db | 2015-10-05 14:38:37 -0700 | [diff] [blame] | 164 | n = len( main.nodes ) # Number of nodes |
| 165 | p = ( ( n + 1 ) / 2 ) + 1 # Number of partitions |
| 166 | main.kill = [ 0 ] # ONOS node to kill, listed by index in main.nodes |
| 167 | if n > 3: |
| 168 | main.kill.append( p - 1 ) |
| 169 | # NOTE: This only works for cluster sizes of 3,5, or 7. |
| 170 | |
Jon Hall | b3ed8ed | 2015-10-28 16:43:55 -0700 | [diff] [blame] | 171 | main.step( "Kill " + str( len( main.kill ) ) + " ONOS nodes" ) |
Jon Hall | 3b489db | 2015-10-05 14:38:37 -0700 | [diff] [blame] | 172 | killResults = main.TRUE |
| 173 | for i in main.kill: |
| 174 | killResults = killResults and\ |
Jon Hall | f37d44d | 2017-05-24 10:37:30 -0700 | [diff] [blame] | 175 | main.ONOSbench.onosKill( main.nodes[ i ].ip_address ) |
Jon Hall | b3ed8ed | 2015-10-28 16:43:55 -0700 | [diff] [blame] | 176 | main.activeNodes.remove( i ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 177 | utilities.assert_equals( expect=main.TRUE, actual=killResults, |
Jon Hall | b3ed8ed | 2015-10-28 16:43:55 -0700 | [diff] [blame] | 178 | onpass="ONOS nodes killed successfully", |
| 179 | onfail="ONOS nodes NOT successfully killed" ) |
| 180 | |
Jon Hall | d2871c2 | 2016-07-26 11:01:14 -0700 | [diff] [blame] | 181 | main.step( "Checking ONOS nodes" ) |
| 182 | nodeResults = utilities.retry( main.HA.nodesCheck, |
| 183 | False, |
Jon Hall | f37d44d | 2017-05-24 10:37:30 -0700 | [diff] [blame] | 184 | args=[ main.activeNodes ], |
Jon Hall | d2871c2 | 2016-07-26 11:01:14 -0700 | [diff] [blame] | 185 | sleep=15, |
| 186 | attempts=5 ) |
| 187 | |
| 188 | utilities.assert_equals( expect=True, actual=nodeResults, |
| 189 | onpass="Nodes check successful", |
| 190 | onfail="Nodes check NOT successful" ) |
| 191 | |
| 192 | if not nodeResults: |
| 193 | for i in main.activeNodes: |
Jon Hall | f37d44d | 2017-05-24 10:37:30 -0700 | [diff] [blame] | 194 | cli = main.CLIs[ i ] |
Jon Hall | d2871c2 | 2016-07-26 11:01:14 -0700 | [diff] [blame] | 195 | main.log.debug( "{} components not ACTIVE: \n{}".format( |
| 196 | cli.name, |
| 197 | cli.sendline( "scr:list | grep -v ACTIVE" ) ) ) |
| 198 | main.log.error( "Failed to start ONOS, stopping test" ) |
| 199 | main.cleanup() |
| 200 | main.exit() |
| 201 | |
Jon Hall | b3ed8ed | 2015-10-28 16:43:55 -0700 | [diff] [blame] | 202 | def CASE62( self, main ): |
| 203 | """ |
| 204 | The bring up stopped nodes |
| 205 | """ |
Jon Hall | b3ed8ed | 2015-10-28 16:43:55 -0700 | [diff] [blame] | 206 | |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 207 | main.HA.bringUpStoppedNode( main ) |
Jon Hall | a440e87 | 2016-03-31 15:15:50 -0700 | [diff] [blame] | 208 | |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 209 | def CASE7( self, main ): |
| 210 | """ |
| 211 | Check state after ONOS failure |
| 212 | """ |
Jon Hall | b3ed8ed | 2015-10-28 16:43:55 -0700 | [diff] [blame] | 213 | try: |
| 214 | main.kill |
| 215 | except AttributeError: |
| 216 | main.kill = [] |
| 217 | |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 218 | main.HA.checkStateAfterONOS( main, afterWhich=0 ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 219 | main.step( "Leadership Election is still functional" ) |
| 220 | # Test of LeadershipElection |
| 221 | leaderList = [] |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 222 | |
Jon Hall | 3b489db | 2015-10-05 14:38:37 -0700 | [diff] [blame] | 223 | restarted = [] |
| 224 | for i in main.kill: |
Jon Hall | f37d44d | 2017-05-24 10:37:30 -0700 | [diff] [blame] | 225 | restarted.append( main.nodes[ i ].ip_address ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 226 | leaderResult = main.TRUE |
Jon Hall | 3b489db | 2015-10-05 14:38:37 -0700 | [diff] [blame] | 227 | |
Jon Hall | b3ed8ed | 2015-10-28 16:43:55 -0700 | [diff] [blame] | 228 | for i in main.activeNodes: |
Jon Hall | f37d44d | 2017-05-24 10:37:30 -0700 | [diff] [blame] | 229 | cli = main.CLIs[ i ] |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 230 | leaderN = cli.electionTestLeader() |
| 231 | leaderList.append( leaderN ) |
| 232 | if leaderN == main.FALSE: |
| 233 | # error in response |
| 234 | main.log.error( "Something is wrong with " + |
| 235 | "electionTestLeader function, check the" + |
| 236 | " error logs" ) |
| 237 | leaderResult = main.FALSE |
| 238 | elif leaderN is None: |
| 239 | main.log.error( cli.name + |
| 240 | " shows no leader for the election-app was" + |
| 241 | " elected after the old one died" ) |
| 242 | leaderResult = main.FALSE |
| 243 | elif leaderN in restarted: |
| 244 | main.log.error( cli.name + " shows " + str( leaderN ) + |
| 245 | " as leader for the election-app, but it " + |
| 246 | "was restarted" ) |
| 247 | leaderResult = main.FALSE |
| 248 | if len( set( leaderList ) ) != 1: |
| 249 | leaderResult = main.FALSE |
| 250 | main.log.error( |
| 251 | "Inconsistent view of leader for the election test app" ) |
| 252 | # TODO: print the list |
| 253 | utilities.assert_equals( |
| 254 | expect=main.TRUE, |
| 255 | actual=leaderResult, |
| 256 | onpass="Leadership election passed", |
| 257 | onfail="Something went wrong with Leadership election" ) |
| 258 | |
| 259 | def CASE8( self, main ): |
| 260 | """ |
| 261 | Compare topo |
| 262 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 263 | main.HA.compareTopo( main ) |
Jon Hall | d2871c2 | 2016-07-26 11:01:14 -0700 | [diff] [blame] | 264 | |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 265 | def CASE9( self, main ): |
| 266 | """ |
| 267 | Link s3-s28 down |
| 268 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 269 | main.HA.linkDown( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 270 | |
| 271 | def CASE10( self, main ): |
| 272 | """ |
| 273 | Link s3-s28 up |
| 274 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 275 | main.HA.linkUp( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 276 | |
| 277 | def CASE11( self, main ): |
| 278 | """ |
| 279 | Switch Down |
| 280 | """ |
| 281 | # NOTE: You should probably run a topology check after this |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 282 | main.HA.switchDown( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 283 | |
| 284 | def CASE12( self, main ): |
| 285 | """ |
| 286 | Switch Up |
| 287 | """ |
| 288 | # NOTE: You should probably run a topology check after this |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 289 | main.HA.switchUp( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 290 | |
| 291 | def CASE13( self, main ): |
| 292 | """ |
| 293 | Clean up |
| 294 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 295 | main.HAlabels.append( "Restart" ) |
| 296 | main.HAdata.append( str( main.restartTime ) ) |
| 297 | main.HA.cleanUp( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 298 | |
| 299 | def CASE14( self, main ): |
| 300 | """ |
| 301 | start election app on all onos nodes |
| 302 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 303 | main.HA.startElectionApp( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 304 | |
| 305 | def CASE15( self, main ): |
| 306 | """ |
| 307 | Check that Leadership Election is still functional |
acsmars | 71adceb | 2015-08-31 15:09:26 -0700 | [diff] [blame] | 308 | 15.1 Run election on each node |
| 309 | 15.2 Check that each node has the same leaders and candidates |
| 310 | 15.3 Find current leader and withdraw |
| 311 | 15.4 Check that a new node was elected leader |
| 312 | 15.5 Check that that new leader was the candidate of old leader |
| 313 | 15.6 Run for election on old leader |
| 314 | 15.7 Check that oldLeader is a candidate, and leader if only 1 node |
| 315 | 15.8 Make sure that the old leader was added to the candidate list |
| 316 | |
| 317 | old and new variable prefixes refer to data from before vs after |
| 318 | withdrawl and later before withdrawl vs after re-election |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 319 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 320 | main.HA.isElectionFunctional( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 321 | |
| 322 | def CASE16( self, main ): |
| 323 | """ |
| 324 | Install Distributed Primitives app |
| 325 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 326 | main.HA.installDistributedPrimitiveApp( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 327 | |
| 328 | def CASE17( self, main ): |
| 329 | """ |
| 330 | Check for basic functionality with distributed primitives |
| 331 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 332 | main.HA.checkDistPrimitivesFunc( main ) |