Resolve shellcheck for start-cluster-run-test
[releng/builder.git] / jjb / integration / integration-start-cluster-run-test.sh
index 8c5fc0fce267c050a0db5960806b1abe80845645..2f2b14c1d84fd3f7315f09ace27a341007e7fe65 100644 (file)
@@ -1,15 +1,19 @@
+#!/bin/bash
 #@IgnoreInspection BashAddShebang
 # Activate robotframework virtualenv
 # ${ROBOT_VENV} comes from the integration-install-robotframework.sh
 # script.
 # shellcheck source=${ROBOT_VENV}/bin/activate disable=SC1091
-source ${ROBOT_VENV}/bin/activate
+source "${ROBOT_VENV}/bin/activate"
+source /tmp/common-functions.sh "${BUNDLEFOLDER}"
+# Ensure we fail the job if any steps fail.
+set -ex -o pipefail
 
 echo "#################################################"
 echo "##         Verify Cluster is UP                ##"
 echo "#################################################"
 
-cat > ${WORKSPACE}/verify-cluster-is-up.sh <<EOF
+cat > "${WORKSPACE}/verify-cluster-is-up.sh" <<EOF
 
 CONTROLLERID="member-\$1"
 ODL_SYSTEM_IP_PATH=\$2
@@ -34,9 +38,11 @@ while true; do
         netstat -pnatu
         exit 1
     else
-        COUNT=\$(( \${COUNT} + 5 ))
-        sleep 5
-        echo waiting \$COUNT secs...
+        COUNT=\$(( \${COUNT} + 1 ))
+        sleep 1
+        if [[ \$((\$COUNT % 5)) == 0 ]]; then
+            echo already waited \${COUNT} seconds...
+        fi
     fi
 done
 
@@ -60,99 +66,121 @@ exit_on_log_file_message 'server is unhealthy'
 
 EOF
 
-for i in `seq 1 ${NUM_ODL_SYSTEM}`
+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} ${!CONTROLLERIP}"
+    scp "${WORKSPACE}/verify-cluster-is-up.sh" "${!CONTROLLERIP}:/tmp"
+    # shellcheck disable=SC2029
+    ssh "${!CONTROLLERIP}" "bash /tmp/verify-cluster-is-up.sh ${i} ${!CONTROLLERIP}"
 done
 
-if [ ${NUM_OPENSTACK_SYSTEM} -gt 0 ]; then
+if [ "${NUM_OPENSTACK_SYSTEM}" -gt 0 ]; then
    echo "Exiting without running tests to deploy openstack for testing"
    exit
 fi
 
-if [ ${CONTROLLERSCOPE} == 'all' ]; then
+if [ "${CONTROLLERSCOPE}" == 'all' ]; then
     COOLDOWN_PERIOD="180"
 else
     COOLDOWN_PERIOD="60"
 fi
 echo "Cool down for ${COOLDOWN_PERIOD} seconds :)..."
-sleep ${COOLDOWN_PERIOD}
+sleep "${COOLDOWN_PERIOD}"
 
 echo "Generating controller variables..."
-for i in `seq 1 ${NUM_ODL_SYSTEM}`
+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
+    ssh "${!CONTROLLERIP}" "sudo ps aux" > "${WORKSPACE}/ps_before.log"
+    pid=$(grep org.apache.karaf.main.Main "${WORKSPACE}/ps_before.log" | grep -v grep | tr -s ' ' | cut -f2 -d' ')
+    echo "karaf main: org.apache.karaf.main.Main, pid:${pid}"
+    # shellcheck disable=SC2029
+    ssh "${!CONTROLLERIP}" "${JAVA_HOME}/bin/jstack -l ${pid}" > "${WORKSPACE}/karaf_${i}_${pid}_threads_before.log" || true
 done
 
 echo "Generating mininet variables..."
-for i in `seq 1 ${NUM_TOOLS_SYSTEM}`
+for i in $(seq 1 "${NUM_TOOLS_SYSTEM}")
 do
     MININETIP=TOOLS_SYSTEM_${i}_IP
     tools_variables=${tools_variables}" -v ${MININETIP}:${!MININETIP}"
 done
 
-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
+get_test_suites SUITES
 
-echo "Changing the testplan path..."
-cat "${testplan_filepath}" | sed "s:integration:${WORKSPACE}:" > testplan.txt
-cat testplan.txt
+echo "Starting Robot test suites ${SUITES} ..."
+robot -N "${TESTPLAN}" \
+      --removekeywords wuks -c critical -e exclude -e "skip_if_${DISTROSTREAM}" \
+      -v BUNDLEFOLDER:"${BUNDLEFOLDER}" \
+      -v BUNDLE_URL:"${ACTUAL_BUNDLE_URL}" \
+      -v CONTROLLER:"${ODL_SYSTEM_IP}" \
+      -v CONTROLLER1:"${ODL_SYSTEM_2_IP}" \
+      -v CONTROLLER2:"${ODL_SYSTEM_3_IP}" \
+      -v CONTROLLER_USER:"${USER}" \
+      -v JAVA_HOME:"${JAVA_HOME}" \
+      -v JDKVERSION:"${JDKVERSION}" \
+      -v JENKINS_WORKSPACE:"${WORKSPACE}" \
+      -v MININET:"${TOOLS_SYSTEM_IP}" \
+      -v MININET1:"${TOOLS_SYSTEM_2_IP}" \
+      -v MININET2:"${TOOLS_SYSTEM_3_IP}" \
+      -v MININET_USER:"${USER}" \
+      -v NEXUSURL_PREFIX:"${NEXUSURL_PREFIX}" \
+      -v NUM_ODL_SYSTEM:"${NUM_ODL_SYSTEM}" \
+      -v NUM_TOOLS_SYSTEM:"${NUM_TOOLS_SYSTEM}" \
+      -v ODL_STREAM:"${DISTROSTREAM}" \
+      -v ODL_SYSTEM_IP:"${ODL_SYSTEM_IP}" "${odl_variables}" \
+      -v ODL_SYSTEM_USER:"${USER}" \
+      -v TOOLS_SYSTEM_IP:"${TOOLS_SYSTEM_IP}" "${tools_variables}" \
+      -v TOOLS_SYSTEM_USER:"${USER}" \
+      -v USER_HOME:"${HOME}" \
+      -v WORKSPACE:/tmp \
+      "${TESTOPTIONS}" "${SUITES}" || true
 
-SUITES=`egrep -v '(^[[:space:]]*#|^[[:space:]]*$)' testplan.txt | tr '\012' ' '`
 
-echo "Starting Robot test suites ${SUITES} ..."
-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/*"
+# shellcheck disable=SC2029
+ssh "${ODL_SYSTEM_1_IP}" "ls -altr /tmp/${BUNDLEFOLDER}/data/log/"
+# shellcheck disable=SC2029
+ssh "${ODL_SYSTEM_1_IP}" "du -hs /tmp/${BUNDLEFOLDER}/data/log/*"
+# shellcheck disable=SC2029
+ssh "${ODL_SYSTEM_2_IP}" "ls -altr /tmp/${BUNDLEFOLDER}/data/log/"
+# shellcheck disable=SC2029
+ssh "${ODL_SYSTEM_2_IP}" "du -hs /tmp/${BUNDLEFOLDER}/data/log/*"
+# shellcheck disable=SC2029
+ssh "${ODL_SYSTEM_3_IP}" "ls -altr /tmp/${BUNDLEFOLDER}/data/log/"
+# shellcheck disable=SC2029
+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}`
+for i in $(seq 1 "${NUM_ODL_SYSTEM}")
 do
-    CONTROLLERIP=ODL_SYSTEM_${i}_IP
+    CONTROLLERIP="ODL_SYSTEM_${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
+    ssh "${!CONTROLLERIP}" "sudo ps aux" > "${WORKSPACE}/ps_after.log"
+    pid=$(grep org.apache.karaf.main.Main "${WORKSPACE}/ps_after.log" | grep -v grep | tr -s ' ' | cut -f2 -d' ')
+    echo "karaf main: org.apache.karaf.main.Main, pid:${pid}"
+    # shellcheck disable=SC2029
+    ssh "${!CONTROLLERIP}" "${JAVA_HOME}/bin/jstack -l ${pid}" > "${WORKSPACE}/karaf_${i}_${pid}_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}`
+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
+    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"
+    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"
+    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"
+    scp "${!CONTROLLERIP}:/tmp/${BUNDLEFOLDER}/data/log/*.log" "gclogs-${i}/" && ssh "${!CONTROLLERIP}" rm -f "/tmp/${BUNDLEFOLDER}/data/log/*.log"
 done
 
 echo "Examine copied files"