Merge "Lower retry sleeps in controller startup wait"
[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/${BUNDLEFOLDER}/data/log/karaf.log"
33         echo "Listing all open ports on controller system"
34         netstat -pnatu
35         exit 1
36     else
37         COUNT=\$(( \${COUNT} + 1 ))
38         sleep 1
39         if [[ \$((\$COUNT % 5)) == 0 ]]; then
40             echo already waited \${COUNT} seconds...
41         fi
42     fi
43 done
44
45 echo "Listing all open ports on controller system.."
46 netstat -pnatu
47
48 function exit_on_log_file_message {
49     echo "looking for \"\$1\" in log file"
50     if grep --quiet "\$1" "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"; then
51         echo ABORTING: found "\$1"
52         echo "Dumping first 500K bytes of karaf log..."
53         head --bytes=500K "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"
54         echo "Dumping last 500K bytes of karaf log..."
55         tail --bytes=500K "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"
56         exit 1
57     fi
58 }
59
60 exit_on_log_file_message 'BindException: Address already in use'
61 exit_on_log_file_message 'server is unhealthy'
62
63 EOF
64
65 for i in `seq 1 ${NUM_ODL_SYSTEM}`
66 do
67     CONTROLLERIP=ODL_SYSTEM_${i}_IP
68     echo "Verifying member-${i} with IP address ${!CONTROLLERIP} is UP"
69     scp ${WORKSPACE}/verify-cluster-is-up.sh ${!CONTROLLERIP}:/tmp
70     ssh ${!CONTROLLERIP} "bash /tmp/verify-cluster-is-up.sh ${i} ${!CONTROLLERIP}"
71 done
72
73 if [ ${NUM_OPENSTACK_SYSTEM} -gt 0 ]; then
74    echo "Exiting without running tests to deploy openstack for testing"
75    exit
76 fi
77
78 if [ ${CONTROLLERSCOPE} == 'all' ]; then
79     COOLDOWN_PERIOD="180"
80 else
81     COOLDOWN_PERIOD="60"
82 fi
83 echo "Cool down for ${COOLDOWN_PERIOD} seconds :)..."
84 sleep ${COOLDOWN_PERIOD}
85
86 echo "Generating controller variables..."
87 for i in `seq 1 ${NUM_ODL_SYSTEM}`
88 do
89     CONTROLLERIP=ODL_SYSTEM_${i}_IP
90     odl_variables=${odl_variables}" -v ${CONTROLLERIP}:${!CONTROLLERIP}"
91     echo "Lets's take the karaf thread dump"
92     KARAF_PID=$(ssh ${!CONTROLLERIP} "ps aux | grep ${KARAF_ARTIFACT} | grep -v grep | tr -s ' ' | cut -f2 -d' '")
93     ssh ${!CONTROLLERIP} "jstack $KARAF_PID"> ${WORKSPACE}/karaf_${i}_threads_before.log || true
94 done
95
96 echo "Generating mininet variables..."
97 for i in `seq 1 ${NUM_TOOLS_SYSTEM}`
98 do
99     MININETIP=TOOLS_SYSTEM_${i}_IP
100     tools_variables=${tools_variables}" -v ${MININETIP}:${!MININETIP}"
101 done
102
103 echo "Locating test plan to use..."
104 testplan_filepath="${WORKSPACE}/test/csit/testplans/${STREAMTESTPLAN}"
105 if [ ! -f "${testplan_filepath}" ]; then
106     testplan_filepath="${WORKSPACE}/test/csit/testplans/${TESTPLAN}"
107 fi
108
109 echo "Changing the testplan path..."
110 cat "${testplan_filepath}" | sed "s:integration:${WORKSPACE}:" > testplan.txt
111 cat testplan.txt
112
113 SUITES=`egrep -v '(^[[:space:]]*#|^[[:space:]]*$)' testplan.txt | tr '\012' ' '`
114
115 echo "Starting Robot test suites ${SUITES} ..."
116 pybot -N ${TESTPLAN} --removekeywords wuks -c critical -e exclude -e skip_if_${DISTROSTREAM} -v BUNDLEFOLDER:${BUNDLEFOLDER} \
117 -v WORKSPACE:/tmp -v BUNDLE_URL:${ACTUAL_BUNDLE_URL} -v JAVA_HOME:${JAVA_HOME} \
118 -v NEXUSURL_PREFIX:${NEXUSURL_PREFIX} -v JDKVERSION:${JDKVERSION} -v ODL_STREAM:${DISTROSTREAM} \
119 -v CONTROLLER:${ODL_SYSTEM_IP} -v CONTROLLER1:${ODL_SYSTEM_2_IP} -v CONTROLLER2:${ODL_SYSTEM_3_IP} -v ODL_SYSTEM_IP:${ODL_SYSTEM_IP} \
120 ${odl_variables} -v NUM_ODL_SYSTEM:${NUM_ODL_SYSTEM} -v CONTROLLER_USER:${USER} -v ODL_SYSTEM_USER:${USER} -v \
121 TOOLS_SYSTEM_IP:${TOOLS_SYSTEM_IP} ${tools_variables} -v NUM_TOOLS_SYSTEM:${NUM_TOOLS_SYSTEM} -v TOOLS_SYSTEM_USER:${USER} \
122 -v MININET:${TOOLS_SYSTEM_IP} -v MININET1:${TOOLS_SYSTEM_2_IP} -v MININET2:${TOOLS_SYSTEM_3_IP} -v MININET_USER:${USER} \
123 -v USER_HOME:${HOME} ${TESTOPTIONS} ${SUITES} || true
124 # FIXME: Sort (at least -v) options alphabetically.
125
126 echo "Examining the files in data/log and checking filesize"
127 ssh ${ODL_SYSTEM_1_IP} "ls -altr /tmp/${BUNDLEFOLDER}/data/log/"
128 ssh ${ODL_SYSTEM_1_IP} "du -hs /tmp/${BUNDLEFOLDER}/data/log/*"
129 ssh ${ODL_SYSTEM_2_IP} "ls -altr /tmp/${BUNDLEFOLDER}/data/log/"
130 ssh ${ODL_SYSTEM_2_IP} "du -hs /tmp/${BUNDLEFOLDER}/data/log/*"
131 ssh ${ODL_SYSTEM_3_IP} "ls -altr /tmp/${BUNDLEFOLDER}/data/log/"
132 ssh ${ODL_SYSTEM_3_IP} "du -hs /tmp/${BUNDLEFOLDER}/data/log/*"
133
134 set +e  # We do not want to create red dot just because something went wrong while fetching logs.
135 for i in `seq 1 ${NUM_ODL_SYSTEM}`
136 do
137     CONTROLLERIP=ODL_SYSTEM_${i}_IP
138     echo "Lets's take the karaf thread dump again"
139     KARAF_PID=$(ssh ${!CONTROLLERIP} "ps aux | grep ${KARAF_ARTIFACT} | grep -v grep | tr -s ' ' | cut -f2 -d' '")
140     ssh ${!CONTROLLERIP} "jstack $KARAF_PID"> ${WORKSPACE}/karaf_${i}_threads_after.log || true
141     echo "killing karaf process..."
142     ssh "${!CONTROLLERIP}" bash -c 'ps axf | grep karaf | grep -v grep | awk '"'"'{print "kill -9 " $1}'"'"' | sh'
143 done
144 sleep 5
145 for i in `seq 1 ${NUM_ODL_SYSTEM}`
146 do
147     CONTROLLERIP=ODL_SYSTEM_${i}_IP
148     echo "Compressing karaf.log ${i}"
149     ssh ${!CONTROLLERIP} gzip --best /tmp/${BUNDLEFOLDER}/data/log/karaf.log
150     echo "Fetching compressed karaf.log ${i}"
151     scp "${!CONTROLLERIP}:/tmp/${BUNDLEFOLDER}/data/log/karaf.log.gz" "odl${i}_karaf.log.gz" && ssh ${!CONTROLLERIP} rm -f "/tmp/${BUNDLEFOLDER}/data/log/karaf.log.gz"
152     # TODO: Should we compress the output log file as well?
153     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"
154     echo "Fetch GC logs"
155     # FIXME: Put member index in filename, instead of directory name.
156     mkdir -p "gclogs-${i}"
157     scp "${!CONTROLLERIP}:/tmp/${BUNDLEFOLDER}/data/log/*.log" "gclogs-${i}/" && ssh ${!CONTROLLERIP} rm -f "/tmp/${BUNDLEFOLDER}/data/log/*.log"
158 done
159
160 echo "Examine copied files"
161 ls -lt
162
163 true  # perhaps Jenkins is testing last exit code
164
165 # vim: ts=4 sw=4 sts=4 et ft=sh :