show karaf.log tail instead of head when controller doesn't come up.

[releng/builder.git] / jjb / integration / include-raw-integration-start-cluster-run-test.sh

diff --git a/jjb/integration/include-raw-integration-start-cluster-run-test.sh b/jjb/integration/include-raw-integration-start-cluster-run-test.sh
index c94814b6dfdc61324ad7518de49a004f6870af65..154d1261fdc614ad2a1207be858671e24f991996 100644 (file)
--- a/jjb/integration/include-raw-integration-start-cluster-run-test.sh
+++ b/jjb/integration/include-raw-integration-start-cluster-run-test.sh
@@ -1,3 +1,9 @@
+#@IgnoreInspection BashAddShebang
+# Activate robotframework virtualenv
+# ${ROBOT_VENV} comes from the include-raw-integration-install-robotframework.sh
+# script.
+source ${ROBOT_VENV}/bin/activate
+
 echo "#################################################"
 echo "##         Verify Cluster is UP                ##"
 echo "#################################################"
@@ -5,7 +11,7 @@ echo "#################################################"
 cat > ${WORKSPACE}/verify-cluster-is-up.sh <<EOF
 
 CONTROLLERID="member-\$1"
-CONTROLLERIP=\$2
+ODL_SYSTEM_IP_PATH=\$2
 
 echo "Waiting for controller to come up..."
 COUNT="0"
@@ -20,7 +26,9 @@ while true; do
     elif (( "\$COUNT" > "600" )); then
         echo Timeout Controller DOWN
         echo "Dumping Karaf log..."
-        cat /tmp/${BUNDLEFOLDER}/data/log/karaf.log
+        tail --bytes=1M "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"
+        echo "Listing all open ports on controller system"
+        netstat -natu
         exit 1
     else
         COUNT=\$(( \${COUNT} + 5 ))
@@ -29,17 +37,30 @@ while true; do
     fi
 done
 
-echo "Checking OSGi bundles..."
-sshpass -p karaf /tmp/${BUNDLEFOLDER}/bin/client -u karaf 'bundle:list'
+echo "Listing all open ports on controller system.."
+netstat -natu
+
+function exit_on_log_file_message {
+    echo "looking for \"\$1\" in log file"
+    if grep --quiet "\$1" "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"; then
+        echo ABORTING: found "\$1"
+        echo "Dumping Karaf log..."
+        tail --bytes=1M "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"
+        exit 1
+    fi
+}
+
+exit_on_log_file_message 'BindException: Address already in use'
+exit_on_log_file_message 'server is unhealthy'
 
 EOF
 
-CONTROLLERIPS=(${CONTROLLER0} ${CONTROLLER1} ${CONTROLLER2})
-for i in "${!CONTROLLERIPS[@]}"
+for i in `seq 1 ${NUM_ODL_SYSTEM}`
 do
-    echo “Verifying member-$i with IP address ${CONTROLLERIPS[$i]} is UP”
-    scp ${WORKSPACE}/verify-cluster-is-up.sh ${CONTROLLERIPS[$i]}:/tmp
-    ssh ${CONTROLLERIPS[$i]} "bash /tmp/verify-cluster-is-up.sh $i ${CONTROLLERIPS[$i]}"
+    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
 
 echo "Cool down for 1 min :)..."
@@ -49,17 +70,46 @@ echo "Changing the testplan path..."
 cat ${WORKSPACE}/test/csit/testplans/${TESTPLAN} | 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 "Generating controller variables..."
+for i in `seq 1 ${NUM_ODL_SYSTEM}`
+do
+    CONTROLLERIP=ODL_SYSTEM_${i}_IP
+    odl_variables=${odl_variables}" -v ${CONTROLLERIP}:${!CONTROLLERIP}"
+done
+
+echo "Generating mininet variables..."
+for i in `seq 1 ${NUM_TOOLS_SYSTEM}`
+do
+    MININETIP=TOOLS_SYSTEM_${i}_IP
+    tools_variables=${tools_variables}" -v ${MININETIP}:${!MININETIP}"
+done
 
 echo "Starting Robot test suites ${SUITES} ..."
-pybot -N ${TESTPLAN} -c critical -e exclude -v BUNDLEFOLDER:${BUNDLEFOLDER} -v WORKSPACE:/tmp \
--v NEXUSURL_PREFIX:${NEXUSURL_PREFIX} -v CONTROLLER:${CONTROLLER0} -v CONTROLLER1:${CONTROLLER1} -v CONTROLLER2:${CONTROLLER2} \
--v CONTROLLER_USER:${USER} -v MININET:${MININET0} -v MININET_USER:${USER} -v USER_HOME:${HOME} ${TESTOPTIONS} ${SUITES} || true
-
-echo "Fetching Karaf log"
-scp $CONTROLLER0:/tmp/$BUNDLEFOLDER/data/log/karaf.log controller0-karaf.log
-scp $CONTROLLER1:/tmp/$BUNDLEFOLDER/data/log/karaf.log controller1-karaf.log
-scp $CONTROLLER2:/tmp/$BUNDLEFOLDER/data/log/karaf.log controller2-karaf.log
+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:${ODL_SYSTEM_IP} \
+${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
+
+echo "Killing ODL and fetching Karaf 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
+    ssh "${!CONTROLLERIP}" tail --bytes=1M "/tmp/${BUNDLEFOLDER}/data/log/karaf.log" > "odl${i}_karaf.log"
+    ssh "${!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
+    ssh "${!CONTROLLERIP}" xz -9ekvv "/tmp/${BUNDLEFOLDER}/data/log/karaf.log"
+    scp "${!CONTROLLERIP}:/tmp/${BUNDLEFOLDER}/data/log/karaf.log.xz" "odl${i}_karaf.log.xz"
+done
+true  # perhaps Jenkins is testing last exit code
 
 # vim: ts=4 sw=4 sts=4 et ft=sh :