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 | """ |
| 23 | |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 24 | Description: This test is to determine if ONOS can handle |
| 25 | all of it's nodes restarting |
| 26 | |
| 27 | List of test cases: |
| 28 | CASE1: Compile ONOS and push it to the test machines |
| 29 | CASE2: Assign devices to controllers |
| 30 | CASE21: Assign mastership to controllers |
| 31 | CASE3: Assign intents |
| 32 | CASE4: Ping across added host intents |
| 33 | CASE5: Reading state of ONOS |
| 34 | CASE6: The Failure 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 | """ |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 47 | class HAclusterRestart: |
| 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 | f3d16e7 | 2015-12-16 17:45:08 -0800 | [diff] [blame] | 71 | import time |
Jon Hall | a440e87 | 2016-03-31 15:15:50 -0700 | [diff] [blame] | 72 | import json |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 73 | main.log.info( "ONOS HA test: Restart all ONOS nodes - " + |
| 74 | "initialization" ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 75 | # These are for csv plotting in jenkins |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 76 | main.HAlabels = [] |
| 77 | main.HAdata = [] |
| 78 | try: |
| 79 | from tests.dependencies.ONOSSetup import ONOSSetup |
| 80 | main.testSetUp = ONOSSetup() |
| 81 | except ImportError: |
| 82 | main.log.error( "ONOSSetup not found exiting the test" ) |
| 83 | main.exit() |
| 84 | main.testSetUp.envSetupDescription() |
Jon Hall | e1a3b75 | 2015-07-22 13:02:46 -0700 | [diff] [blame] | 85 | try: |
Jon Hall | 53c5e66 | 2016-04-13 16:06:56 -0700 | [diff] [blame] | 86 | from tests.HA.dependencies.HA import HA |
Jon Hall | 41d39f1 | 2016-04-11 22:54:35 -0700 | [diff] [blame] | 87 | main.HA = HA() |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 88 | # load some variables from the params file |
| 89 | cellName = main.params[ 'ENV' ][ 'cellName' ] |
| 90 | main.apps = main.params[ 'ENV' ][ 'appString' ] |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 91 | stepResult = main.testSetUp.envSetup() |
Jon Hall | e1a3b75 | 2015-07-22 13:02:46 -0700 | [diff] [blame] | 92 | except Exception as e: |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 93 | main.testSetUp.envSetupException( e ) |
| 94 | main.testSetUp.evnSetupConclusion( stepResult ) |
| 95 | main.HA.generateGraph( "HAclusterRestart" ) |
Jon Hall | e1a3b75 | 2015-07-22 13:02:46 -0700 | [diff] [blame] | 96 | |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 97 | main.testSetUp.ONOSSetUp( main.Mininet1, main.Cluster, cellName=cellName, removeLog=True, |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 98 | extraApply=main.HA.startingMininet ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 99 | |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 100 | main.HA.initialSetUp() |
Jon Hall | 9d2dcad | 2016-04-08 10:15:20 -0700 | [diff] [blame] | 101 | |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 102 | def CASE2( self, main ): |
| 103 | """ |
| 104 | Assign devices to controllers |
| 105 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 106 | main.HA.assignDevices( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 107 | |
| 108 | def CASE21( self, main ): |
| 109 | """ |
| 110 | Assign mastership to controllers |
| 111 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 112 | main.HA.assignMastership( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 113 | def CASE3( self, main ): |
| 114 | """ |
| 115 | Assign intents |
| 116 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 117 | main.HA.assignIntents(main) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 118 | |
| 119 | def CASE4( self, main ): |
| 120 | """ |
| 121 | Ping across added host intents |
| 122 | """ |
Jon Hall | ca31989 | 2017-06-15 15:25:22 -0700 | [diff] [blame] | 123 | main.HA.pingAcrossHostIntent( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 124 | |
| 125 | def CASE5( self, main ): |
| 126 | """ |
| 127 | Reading state of ONOS |
| 128 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 129 | main.HA.readingState( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 130 | |
| 131 | def CASE6( self, main ): |
| 132 | """ |
| 133 | The Failure case. |
| 134 | """ |
| 135 | import time |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 136 | assert main, "main not defined" |
| 137 | assert utilities.assert_equals, "utilities.assert_equals not defined" |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 138 | try: |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 139 | main.HAlabels |
| 140 | except ( NameError, AttributeError ): |
| 141 | main.log.error( "main.HAlabels not defined, setting to []" ) |
| 142 | main.HAlabels = [] |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 143 | try: |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 144 | main.HAdata |
| 145 | except ( NameError, AttributeError ): |
| 146 | main.log.error( "main.HAdata not defined, setting to []" ) |
| 147 | main.HAdata = [] |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 148 | # Reset non-persistent variables |
| 149 | try: |
| 150 | iCounterValue = 0 |
| 151 | except NameError: |
| 152 | main.log.error( "iCounterValue not defined, setting to 0" ) |
| 153 | iCounterValue = 0 |
| 154 | |
| 155 | main.case( "Restart entire ONOS cluster" ) |
| 156 | |
Jon Hall | 5ec6b1b | 2015-09-17 18:20:14 -0700 | [diff] [blame] | 157 | main.step( "Checking ONOS Logs for errors" ) |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 158 | for ctrl in main.Cluster.active(): |
| 159 | main.log.debug( "Checking logs for errors on " + ctrl.name + ":" ) |
| 160 | main.log.warn( main.ONOSbench.checkLogs( ctrl.ipAddress ) ) |
Jon Hall | 5ec6b1b | 2015-09-17 18:20:14 -0700 | [diff] [blame] | 161 | |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 162 | main.testSetUp.killingAllOnos( main.Cluster, True, False ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 163 | |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 164 | main.testSetUp.checkOnosService( main.Cluster ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 165 | |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 166 | main.testSetUp.startOnosClis( main.Cluster ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 167 | |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 168 | ready = utilities.retry( main.Cluster.command, |
| 169 | False, |
| 170 | kwargs={ "function":"summary", "contentCheck":True }, |
| 171 | sleep=30, |
| 172 | attempts=10 ) |
Jon Hall | 6e70975 | 2016-02-01 13:38:46 -0800 | [diff] [blame] | 173 | utilities.assert_equals( expect=True, actual=ready, |
| 174 | onpass="ONOS summary command succeded", |
| 175 | onfail="ONOS summary command failed" ) |
| 176 | if not ready: |
| 177 | main.cleanup() |
| 178 | main.exit() |
| 179 | |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 180 | # Grab the time of restart so we chan check how long the gossip |
| 181 | # protocol has had time to work |
| 182 | main.restartTime = time.time() - killTime |
| 183 | main.log.debug( "Restart time: " + str( main.restartTime ) ) |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 184 | main.HAlabels.append( "Restart" ) |
| 185 | main.HAdata.append( str( main.restartTime ) ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 186 | |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 187 | # Rerun for election on restarted nodes |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 188 | runResults = main.Cluster.command( "electionTestRun", returnBool=True ) |
| 189 | utilities.assert_equals( expect=True, actual=runResults, |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 190 | onpass="Reran for election", |
| 191 | onfail="Failed to rerun for election" ) |
| 192 | |
| 193 | # TODO: Make this configurable |
| 194 | time.sleep( 60 ) |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 195 | |
| 196 | main.HA.commonChecks() |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 197 | |
| 198 | def CASE7( self, main ): |
| 199 | """ |
| 200 | Check state after ONOS failure |
| 201 | """ |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 202 | # NOTE: Store has no durability, so intents are lost across system |
| 203 | # restarts |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 204 | main.HA.checkStateAfterEvent( main, afterWhich=0, isRestart=True ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 205 | |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 206 | main.step( "Leadership Election is still functional" ) |
| 207 | # Test of LeadershipElection |
| 208 | leaderList = [] |
| 209 | leaderResult = main.TRUE |
Jon Hall | a440e87 | 2016-03-31 15:15:50 -0700 | [diff] [blame] | 210 | |
Devin Lim | 142b534 | 2017-07-20 15:22:39 -0700 | [diff] [blame] | 211 | for ctrl in main.Cluster.active(): |
| 212 | ctrl.CLI.electionTestLeader() |
| 213 | leaderN = ctrl.CLI.electionTestLeader() |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 214 | leaderList.append( leaderN ) |
| 215 | if leaderN == main.FALSE: |
| 216 | # error in response |
| 217 | main.log.error( "Something is wrong with " + |
| 218 | "electionTestLeader function, check the" + |
| 219 | " error logs" ) |
| 220 | leaderResult = main.FALSE |
| 221 | elif leaderN is None: |
| 222 | main.log.error( cli.name + |
| 223 | " shows no leader for the election-app." ) |
| 224 | leaderResult = main.FALSE |
| 225 | if len( set( leaderList ) ) != 1: |
| 226 | leaderResult = main.FALSE |
| 227 | main.log.error( |
| 228 | "Inconsistent view of leader for the election test app" ) |
| 229 | # TODO: print the list |
| 230 | utilities.assert_equals( |
| 231 | expect=main.TRUE, |
| 232 | actual=leaderResult, |
| 233 | onpass="Leadership election passed", |
| 234 | onfail="Something went wrong with Leadership election" ) |
| 235 | |
| 236 | def CASE8( self, main ): |
| 237 | """ |
| 238 | Compare topo |
| 239 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 240 | main.HA.compareTopo( main ) |
Jon Hall | d2871c2 | 2016-07-26 11:01:14 -0700 | [diff] [blame] | 241 | |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 242 | def CASE9( self, main ): |
| 243 | """ |
| 244 | Link s3-s28 down |
| 245 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 246 | main.HA.linkDown( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 247 | |
| 248 | def CASE10( self, main ): |
| 249 | """ |
| 250 | Link s3-s28 up |
| 251 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 252 | main.HA.linkUp( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 253 | |
| 254 | def CASE11( self, main ): |
| 255 | """ |
| 256 | Switch Down |
| 257 | """ |
| 258 | # NOTE: You should probably run a topology check after this |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 259 | main.HA.switchDown( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 260 | |
| 261 | def CASE12( self, main ): |
| 262 | """ |
| 263 | Switch Up |
| 264 | """ |
| 265 | # NOTE: You should probably run a topology check after this |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 266 | main.HA.switchUp( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 267 | |
| 268 | def CASE13( self, main ): |
| 269 | """ |
| 270 | Clean up |
| 271 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 272 | main.HA.cleanUp( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 273 | |
| 274 | def CASE14( self, main ): |
| 275 | """ |
| 276 | start election app on all onos nodes |
| 277 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 278 | main.HA.startElectionApp( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 279 | |
| 280 | def CASE15( self, main ): |
| 281 | """ |
| 282 | Check that Leadership Election is still functional |
acsmars | 9475b1c | 2015-08-28 18:02:08 -0700 | [diff] [blame] | 283 | 15.1 Run election on each node |
| 284 | 15.2 Check that each node has the same leaders and candidates |
| 285 | 15.3 Find current leader and withdraw |
| 286 | 15.4 Check that a new node was elected leader |
| 287 | 15.5 Check that that new leader was the candidate of old leader |
| 288 | 15.6 Run for election on old leader |
| 289 | 15.7 Check that oldLeader is a candidate, and leader if only 1 node |
| 290 | 15.8 Make sure that the old leader was added to the candidate list |
| 291 | |
| 292 | old and new variable prefixes refer to data from before vs after |
| 293 | withdrawl and later before withdrawl vs after re-election |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 294 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 295 | main.HA.isElectionFunctional( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 296 | |
| 297 | def CASE16( self, main ): |
| 298 | """ |
| 299 | Install Distributed Primitives app |
| 300 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 301 | main.HA.installDistributedPrimitiveApp( main ) |
Jon Hall | 5cf14d5 | 2015-07-16 12:15:19 -0700 | [diff] [blame] | 302 | |
| 303 | def CASE17( self, main ): |
| 304 | """ |
| 305 | Check for basic functionality with distributed primitives |
| 306 | """ |
Devin Lim | 58046fa | 2017-07-05 16:55:00 -0700 | [diff] [blame] | 307 | main.HA.checkDistPrimitivesFunc( main ) |