Merge "Improve branch cutting scripts"
[releng/builder.git] / jjb / integration / include-raw-integration-start-cluster-run-test.sh
index cc311cd84a1d476489017c9fa4d39488245283be..97405ab8dc71492218064018574bec4aff90be7d 100644 (file)
@@ -2,6 +2,7 @@
 # Activate robotframework virtualenv
 # ${ROBOT_VENV} comes from the include-raw-integration-install-robotframework.sh
 # script.
+# shellcheck source=${ROBOT_VENV}/bin/activate disable=SC1091
 source ${ROBOT_VENV}/bin/activate
 
 echo "#################################################"
@@ -25,10 +26,12 @@ while true; do
         break
     elif (( "\$COUNT" > "600" )); then
         echo Timeout Controller DOWN
-        echo "Dumping Karaf log..."
-        cat /tmp/${BUNDLEFOLDER}/data/log/karaf.log
+        echo "Dumping first 500K bytes of karaf log..."
+        head --bytes=500K "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"
+        echo "Dumping last 500K bytes of karaf log..."
+        tail --bytes=500K "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"
         echo "Listing all open ports on controller system"
-        netstat -natu
+        netstat -pnatu
         exit 1
     else
         COUNT=\$(( \${COUNT} + 5 ))
@@ -37,18 +40,17 @@ while true; do
     fi
 done
 
-echo "Checking OSGi bundles..."
-sshpass -p karaf /tmp/${BUNDLEFOLDER}/bin/client -u karaf 'bundle:list'
-
-echo "Listing all open ports on controller system"
-netstat -natu
+echo "Listing all open ports on controller system.."
+netstat -pnatu
 
 function exit_on_log_file_message {
     echo "looking for \"\$1\" in log file"
-    if grep --quiet "\$1" /tmp/${BUNDLEFOLDER}/data/log/karaf.log; then
+    if grep --quiet "\$1" "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"; then
         echo ABORTING: found "\$1"
-        echo "Dumping Karaf log..."
-        cat /tmp/${BUNDLEFOLDER}/data/log/karaf.log
+        echo "Dumping first 500K bytes of karaf log..."
+        head --bytes=500K "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"
+        echo "Dumping last 500K bytes of karaf log..."
+        tail --bytes=500K "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"
         exit 1
     fi
 }
@@ -58,37 +60,104 @@ exit_on_log_file_message 'server is unhealthy'
 
 EOF
 
-ODL_SYSTEM_IPS=(${ODL_SYSTEM_1_IP} ${ODL_SYSTEM_2_IP} ${ODL_SYSTEM_3_IP})
-for i in "${!ODL_SYSTEM_IPS[@]}"
+for i in `seq 1 ${NUM_ODL_SYSTEM}`
+do
+    CONTROLLERIP=ODL_SYSTEM_${i}_IP
+    echo "Verifying member-${i} with IP address ${!CONTROLLERIP} is UP"
+    scp ${WORKSPACE}/verify-cluster-is-up.sh ${!CONTROLLERIP}:/tmp
+    ssh ${!CONTROLLERIP} "bash /tmp/verify-cluster-is-up.sh ${i} ${!CONTROLLERIP}"
+done
+
+if [ ${NUM_OPENSTACK_SYSTEM} -gt 0 ]; then
+   echo "Exiting without running tests to deploy openstack for testing"
+   exit
+fi
+
+if [ ${CONTROLLERSCOPE} == 'all' ]; then
+    COOLDOWN_PERIOD="180"
+else
+    COOLDOWN_PERIOD="60"
+fi
+echo "Cool down for ${COOLDOWN_PERIOD} seconds :)..."
+sleep ${COOLDOWN_PERIOD}
+
+echo "Generating controller variables..."
+for i in `seq 1 ${NUM_ODL_SYSTEM}`
+do
+    CONTROLLERIP=ODL_SYSTEM_${i}_IP
+    odl_variables=${odl_variables}" -v ${CONTROLLERIP}:${!CONTROLLERIP}"
+    echo "Lets's take the karaf thread dump"
+    KARAF_PID=$(ssh ${!CONTROLLERIP} "ps aux | grep ${KARAF_ARTIFACT} | grep -v grep | tr -s ' ' | cut -f2 -d' '")
+    ssh ${!CONTROLLERIP} "jstack $KARAF_PID"> ${WORKSPACE}/karaf_${i}_threads_before.log || true
+done
+
+echo "Generating mininet variables..."
+for i in `seq 1 ${NUM_TOOLS_SYSTEM}`
 do
-    echo "Verifying member-$((i+1)) with IP address ${ODL_SYSTEM_IPS[$i]} is UP"
-    scp ${WORKSPACE}/verify-cluster-is-up.sh ${ODL_SYSTEM_IPS[$i]}:/tmp
-    ssh ${ODL_SYSTEM_IPS[$i]} "bash /tmp/verify-cluster-is-up.sh $((i+1)) ${ODL_SYSTEM_IPS[$i]}"
+    MININETIP=TOOLS_SYSTEM_${i}_IP
+    tools_variables=${tools_variables}" -v ${MININETIP}:${!MININETIP}"
 done
 
-echo "Cool down for 1 min :)..."
-sleep 60
+echo "Locating test plan to use..."
+testplan_filepath="${WORKSPACE}/test/csit/testplans/${STREAMTESTPLAN}"
+if [ ! -f "${testplan_filepath}" ]; then
+    testplan_filepath="${WORKSPACE}/test/csit/testplans/${TESTPLAN}"
+fi
 
 echo "Changing the testplan path..."
-cat ${WORKSPACE}/test/csit/testplans/${TESTPLAN} | sed "s:integration:${WORKSPACE}:" > testplan.txt
+cat "${testplan_filepath}" | sed "s:integration:${WORKSPACE}:" > testplan.txt
 cat testplan.txt
 
-SUITES=$( egrep -v '(^[[:space:]]*#|^[[:space:]]*$)' testplan.txt | tr '\012' ' ' )
+SUITES=`egrep -v '(^[[:space:]]*#|^[[:space:]]*$)' testplan.txt | tr '\012' ' '`
 
 echo "Starting Robot test suites ${SUITES} ..."
-pybot -N ${TESTPLAN} -c critical -e exclude -v BUNDLEFOLDER:${BUNDLEFOLDER} -v WORKSPACE:/tmp \
--v NEXUSURL_PREFIX:${NEXUSURL_PREFIX} -v CONTROLLER:${ODL_SYSTEM_IP} -v CONTROLLER1:${ODL_SYSTEM_2_IP} \
--v CONTROLLER2:${ODL_SYSTEM_3_IP} -v ODL_SYSTEM_IP:${ODL_SYSTEM_IP} -v ODL_SYSTEM_2_IP:${ODL_SYSTEM_2_IP} \
--v ODL_SYSTEM_3_IP:${ODL_SYSTEM_3_IP} -v CONTROLLER_USER:${USER} -v ODL_SYSTEM_USER:${USER} \
--v TOOLS_SYSTEM_IP:${TOOLS_SYSTEM_IP} -v TOOLS_SYSTEM_2_IP:${TOOLS_SYSTEM_2_IP} -v TOOLS_SYSTEM_3_IP:${TOOLS_SYSTEM_3_IP} \
--v TOOLS_SYSTEM_USER:${USER} \
+pybot -N ${TESTPLAN} --removekeywords wuks -c critical -e exclude -e skip_if_${DISTROSTREAM} -v BUNDLEFOLDER:${BUNDLEFOLDER} \
+-v WORKSPACE:/tmp -v BUNDLE_URL:${ACTUAL_BUNDLE_URL} -v JAVA_HOME:${JAVA_HOME} \
+-v NEXUSURL_PREFIX:${NEXUSURL_PREFIX} -v JDKVERSION:${JDKVERSION} -v ODL_STREAM:${DISTROSTREAM} \
+-v CONTROLLER:${ODL_SYSTEM_IP} -v CONTROLLER1:${ODL_SYSTEM_2_IP} -v CONTROLLER2:${ODL_SYSTEM_3_IP} -v ODL_SYSTEM_IP:${ODL_SYSTEM_IP} \
+${odl_variables} -v NUM_ODL_SYSTEM:${NUM_ODL_SYSTEM} -v CONTROLLER_USER:${USER} -v ODL_SYSTEM_USER:${USER} -v \
+TOOLS_SYSTEM_IP:${TOOLS_SYSTEM_IP} ${tools_variables} -v NUM_TOOLS_SYSTEM:${NUM_TOOLS_SYSTEM} -v TOOLS_SYSTEM_USER:${USER} \
 -v MININET:${TOOLS_SYSTEM_IP} -v MININET1:${TOOLS_SYSTEM_2_IP} -v MININET2:${TOOLS_SYSTEM_3_IP} -v MININET_USER:${USER} \
 -v USER_HOME:${HOME} ${TESTOPTIONS} ${SUITES} || true
+# FIXME: Sort (at least -v) options alphabetically.
+
+echo "Examining the files in data/log and checking filesize"
+ssh ${ODL_SYSTEM_1_IP} "ls -altr /tmp/${BUNDLEFOLDER}/data/log/"
+ssh ${ODL_SYSTEM_1_IP} "du -hs /tmp/${BUNDLEFOLDER}/data/log/*"
+ssh ${ODL_SYSTEM_2_IP} "ls -altr /tmp/${BUNDLEFOLDER}/data/log/"
+ssh ${ODL_SYSTEM_2_IP} "du -hs /tmp/${BUNDLEFOLDER}/data/log/*"
+ssh ${ODL_SYSTEM_3_IP} "ls -altr /tmp/${BUNDLEFOLDER}/data/log/"
+ssh ${ODL_SYSTEM_3_IP} "du -hs /tmp/${BUNDLEFOLDER}/data/log/*"
+
+set +e  # We do not want to create red dot just because something went wrong while fetching logs.
+for i in `seq 1 ${NUM_ODL_SYSTEM}`
+do
+    CONTROLLERIP=ODL_SYSTEM_${i}_IP
+    echo "Lets's take the karaf thread dump again"
+    KARAF_PID=$(ssh ${!CONTROLLERIP} "ps aux | grep ${KARAF_ARTIFACT} | grep -v grep | tr -s ' ' | cut -f2 -d' '")
+    ssh ${!CONTROLLERIP} "jstack $KARAF_PID"> ${WORKSPACE}/karaf_${i}_threads_after.log || true
+    echo "killing karaf process..."
+    ssh "${!CONTROLLERIP}" bash -c 'ps axf | grep karaf | grep -v grep | awk '"'"'{print "kill -9 " $1}'"'"' | sh'
+done
+sleep 5
+for i in `seq 1 ${NUM_ODL_SYSTEM}`
+do
+    CONTROLLERIP=ODL_SYSTEM_${i}_IP
+    echo "Compressing karaf.log ${i}"
+    ssh ${!CONTROLLERIP} gzip --best /tmp/${BUNDLEFOLDER}/data/log/karaf.log
+    echo "Fetching compressed karaf.log ${i}"
+    scp "${!CONTROLLERIP}:/tmp/${BUNDLEFOLDER}/data/log/karaf.log.gz" "odl${i}_karaf.log.gz" && ssh ${!CONTROLLERIP} rm -f "/tmp/${BUNDLEFOLDER}/data/log/karaf.log.gz"
+    # TODO: Should we compress the output log file as well?
+    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"
+    echo "Fetch GC logs"
+    # FIXME: Put member index in filename, instead of directory name.
+    mkdir -p "gclogs-${i}"
+    scp "${!CONTROLLERIP}:/tmp/${BUNDLEFOLDER}/data/log/*.log" "gclogs-${i}/" && ssh ${!CONTROLLERIP} rm -f "/tmp/${BUNDLEFOLDER}/data/log/*.log"
+done
 
-echo "Fetching Karaf log"
-scp $ODL_SYSTEM_IP:/tmp/$BUNDLEFOLDER/data/log/karaf.log controller0-karaf.log
-scp $ODL_SYSTEM_2_IP:/tmp/$BUNDLEFOLDER/data/log/karaf.log controller1-karaf.log
-scp $ODL_SYSTEM_3_IP:/tmp/$BUNDLEFOLDER/data/log/karaf.log controller2-karaf.log
+echo "Examine copied files"
+ls -lt
 
-# vim: ts=4 sw=4 sts=4 et ft=sh :
+true  # perhaps Jenkins is testing last exit code
 
+# vim: ts=4 sw=4 sts=4 et ft=sh :