Merge "Update validate autorelease projects for nitrogen"
[releng/builder.git] / jjb / integration / integration-start-cluster-run-test.sh
1 #@IgnoreInspection BashAddShebang
2 # Activate robotframework virtualenv
3 # ${ROBOT_VENV} comes from the integration-install-robotframework.sh
4 # script.
5 # shellcheck source=${ROBOT_VENV}/bin/activate disable=SC1091
6 source ${ROBOT_VENV}/bin/activate
7
8 echo "#################################################"
9 echo "##         Verify Cluster is UP                ##"
10 echo "#################################################"
11
12 cat > ${WORKSPACE}/verify-cluster-is-up.sh <<EOF
13
14 CONTROLLERID="member-\$1"
15 ODL_SYSTEM_IP_PATH=\$2
16
17 echo "Waiting for controller to come up..."
18 COUNT="0"
19 while true; do
20     RESP="\$( curl --user admin:admin -sL -w "%{http_code} %{url_effective}\\n" http://localhost:8181/restconf/modules -o /dev/null )"
21     echo \$RESP
22     SHARD="\$( curl --user admin:admin -sL -w "%{http_code} %{url_effective}\\n" http://localhost:8181/jolokia/read/org.opendaylight.controller:Category=Shards,name=\$CONTROLLERID-shard-inventory-config,type=DistributedConfigDatastore)"
23     echo \$SHARD
24     if ([[ \$RESP == *"200"* ]] && [[ \$SHARD  == *'"status":200'* ]]); then
25         echo Controller is UP
26         break
27     elif (( "\$COUNT" > "600" )); then
28         echo Timeout Controller DOWN
29         echo "Dumping first 500K bytes of karaf log..."
30         head --bytes=500K "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"
31         echo "Dumping last 500K bytes of karaf log..."
32         tail --bytes=500K "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"
33         echo "Listing all open ports on controller system"
34         netstat -pnatu
35         exit 1
36     else
37         COUNT=\$(( \${COUNT} + 5 ))
38         sleep 5
39         echo waiting \$COUNT secs...
40     fi
41 done
42
43 echo "Listing all open ports on controller system.."
44 netstat -pnatu
45
46 function exit_on_log_file_message {
47     echo "looking for \"\$1\" in log file"
48     if grep --quiet "\$1" "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"; then
49         echo ABORTING: found "\$1"
50         echo "Dumping first 500K bytes of karaf log..."
51         head --bytes=500K "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"
52         echo "Dumping last 500K bytes of karaf log..."
53         tail --bytes=500K "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"
54         exit 1
55     fi
56 }
57
58 exit_on_log_file_message 'BindException: Address already in use'
59 exit_on_log_file_message 'server is unhealthy'
60
61 EOF
62
63 for i in `seq 1 ${NUM_ODL_SYSTEM}`
64 do
65     CONTROLLERIP=ODL_SYSTEM_${i}_IP
66     echo "Verifying member-${i} with IP address ${!CONTROLLERIP} is UP"
67     scp ${WORKSPACE}/verify-cluster-is-up.sh ${!CONTROLLERIP}:/tmp
68     ssh ${!CONTROLLERIP} "bash /tmp/verify-cluster-is-up.sh ${i} ${!CONTROLLERIP}"
69 done
70
71 if [ ${NUM_OPENSTACK_SYSTEM} -gt 0 ]; then
72    echo "Exiting without running tests to deploy openstack for testing"
73    exit
74 fi
75
76 if [ ${CONTROLLERSCOPE} == 'all' ]; then
77     COOLDOWN_PERIOD="180"
78 else
79     COOLDOWN_PERIOD="60"
80 fi
81 echo "Cool down for ${COOLDOWN_PERIOD} seconds :)..."
82 sleep ${COOLDOWN_PERIOD}
83
84 echo "Generating controller variables..."
85 for i in `seq 1 ${NUM_ODL_SYSTEM}`
86 do
87     CONTROLLERIP=ODL_SYSTEM_${i}_IP
88     odl_variables=${odl_variables}" -v ${CONTROLLERIP}:${!CONTROLLERIP}"
89     echo "Lets's take the karaf thread dump"
90     KARAF_PID=$(ssh ${!CONTROLLERIP} "ps aux | grep ${KARAF_ARTIFACT} | grep -v grep | tr -s ' ' | cut -f2 -d' '")
91     ssh ${!CONTROLLERIP} "jstack $KARAF_PID"> ${WORKSPACE}/karaf_${i}_threads_before.log || true
92 done
93
94 echo "Generating mininet variables..."
95 for i in `seq 1 ${NUM_TOOLS_SYSTEM}`
96 do
97     MININETIP=TOOLS_SYSTEM_${i}_IP
98     tools_variables=${tools_variables}" -v ${MININETIP}:${!MININETIP}"
99 done
100
101 echo "Locating test plan to use..."
102 testplan_filepath="${WORKSPACE}/test/csit/testplans/${STREAMTESTPLAN}"
103 if [ ! -f "${testplan_filepath}" ]; then
104     testplan_filepath="${WORKSPACE}/test/csit/testplans/${TESTPLAN}"
105 fi
106
107 echo "Changing the testplan path..."
108 cat "${testplan_filepath}" | sed "s:integration:${WORKSPACE}:" > testplan.txt
109 cat testplan.txt
110
111 SUITES=`egrep -v '(^[[:space:]]*#|^[[:space:]]*$)' testplan.txt | tr '\012' ' '`
112
113 echo "Starting Robot test suites ${SUITES} ..."
114 pybot -N ${TESTPLAN} --removekeywords wuks -c critical -e exclude -e skip_if_${DISTROSTREAM} -v BUNDLEFOLDER:${BUNDLEFOLDER} \
115 -v WORKSPACE:/tmp -v BUNDLE_URL:${ACTUAL_BUNDLE_URL} -v JAVA_HOME:${JAVA_HOME} \
116 -v NEXUSURL_PREFIX:${NEXUSURL_PREFIX} -v JDKVERSION:${JDKVERSION} -v ODL_STREAM:${DISTROSTREAM} \
117 -v CONTROLLER:${ODL_SYSTEM_IP} -v CONTROLLER1:${ODL_SYSTEM_2_IP} -v CONTROLLER2:${ODL_SYSTEM_3_IP} -v ODL_SYSTEM_IP:${ODL_SYSTEM_IP} \
118 ${odl_variables} -v NUM_ODL_SYSTEM:${NUM_ODL_SYSTEM} -v CONTROLLER_USER:${USER} -v ODL_SYSTEM_USER:${USER} -v \
119 TOOLS_SYSTEM_IP:${TOOLS_SYSTEM_IP} ${tools_variables} -v NUM_TOOLS_SYSTEM:${NUM_TOOLS_SYSTEM} -v TOOLS_SYSTEM_USER:${USER} \
120 -v MININET:${TOOLS_SYSTEM_IP} -v MININET1:${TOOLS_SYSTEM_2_IP} -v MININET2:${TOOLS_SYSTEM_3_IP} -v MININET_USER:${USER} \
121 -v USER_HOME:${HOME} ${TESTOPTIONS} ${SUITES} || true
122 # FIXME: Sort (at least -v) options alphabetically.
123
124 echo "Examining the files in data/log and checking filesize"
125 ssh ${ODL_SYSTEM_1_IP} "ls -altr /tmp/${BUNDLEFOLDER}/data/log/"
126 ssh ${ODL_SYSTEM_1_IP} "du -hs /tmp/${BUNDLEFOLDER}/data/log/*"
127 ssh ${ODL_SYSTEM_2_IP} "ls -altr /tmp/${BUNDLEFOLDER}/data/log/"
128 ssh ${ODL_SYSTEM_2_IP} "du -hs /tmp/${BUNDLEFOLDER}/data/log/*"
129 ssh ${ODL_SYSTEM_3_IP} "ls -altr /tmp/${BUNDLEFOLDER}/data/log/"
130 ssh ${ODL_SYSTEM_3_IP} "du -hs /tmp/${BUNDLEFOLDER}/data/log/*"
131
132 set +e  # We do not want to create red dot just because something went wrong while fetching logs.
133 for i in `seq 1 ${NUM_ODL_SYSTEM}`
134 do
135     CONTROLLERIP=ODL_SYSTEM_${i}_IP
136     echo "Lets's take the karaf thread dump again"
137     KARAF_PID=$(ssh ${!CONTROLLERIP} "ps aux | grep ${KARAF_ARTIFACT} | grep -v grep | tr -s ' ' | cut -f2 -d' '")
138     ssh ${!CONTROLLERIP} "jstack $KARAF_PID"> ${WORKSPACE}/karaf_${i}_threads_after.log || true
139     echo "killing karaf process..."
140     ssh "${!CONTROLLERIP}" bash -c 'ps axf | grep karaf | grep -v grep | awk '"'"'{print "kill -9 " $1}'"'"' | sh'
141 done
142 sleep 5
143 for i in `seq 1 ${NUM_ODL_SYSTEM}`
144 do
145     CONTROLLERIP=ODL_SYSTEM_${i}_IP
146     echo "Compressing karaf.log ${i}"
147     ssh ${!CONTROLLERIP} gzip --best /tmp/${BUNDLEFOLDER}/data/log/karaf.log
148     echo "Fetching compressed karaf.log ${i}"
149     scp "${!CONTROLLERIP}:/tmp/${BUNDLEFOLDER}/data/log/karaf.log.gz" "odl${i}_karaf.log.gz" && ssh ${!CONTROLLERIP} rm -f "/tmp/${BUNDLEFOLDER}/data/log/karaf.log.gz"
150     # TODO: Should we compress the output log file as well?
151     scp "${!CONTROLLERIP}:/tmp/${BUNDLEFOLDER}/data/log/karaf_console.log" "odl${i}_karaf_console.log" && ssh ${!CONTROLLERIP} rm -f "/tmp/${BUNDLEFOLDER}/data/log/karaf_console.log"
152     echo "Fetch GC logs"
153     # FIXME: Put member index in filename, instead of directory name.
154     mkdir -p "gclogs-${i}"
155     scp "${!CONTROLLERIP}:/tmp/${BUNDLEFOLDER}/data/log/*.log" "gclogs-${i}/" && ssh ${!CONTROLLERIP} rm -f "/tmp/${BUNDLEFOLDER}/data/log/*.log"
156 done
157
158 echo "Examine copied files"
159 ls -lt
160
161 true  # perhaps Jenkins is testing last exit code
162
163 # vim: ts=4 sw=4 sts=4 et ft=sh :