#!/bin/bash echo "common-functions.sh is being sourced" BUNDLEFOLDER=$1 # Basic controller configuration settings export MAVENCONF=/tmp/${BUNDLEFOLDER}/etc/org.ops4j.pax.url.mvn.cfg export FEATURESCONF=/tmp/${BUNDLEFOLDER}/etc/org.apache.karaf.features.cfg export CUSTOMPROP=/tmp/${BUNDLEFOLDER}/etc/custom.properties export LOGCONF=/tmp/${BUNDLEFOLDER}/etc/org.ops4j.pax.logging.cfg export MEMCONF=/tmp/${BUNDLEFOLDER}/bin/setenv export CONTROLLERMEM=${CONTROLLERMAXMEM} # Cluster specific configuration settings export AKKACONF=/tmp/${BUNDLEFOLDER}/configuration/initial/akka.conf export MODULESCONF=/tmp/${BUNDLEFOLDER}/configuration/initial/modules.conf export MODULESHARDSCONF=/tmp/${BUNDLEFOLDER}/configuration/initial/module-shards.conf function print_common_env() { cat << EOF common-functions environment: MAVENCONF: ${MAVENCONF} ACTUALFEATURES: ${ACTUALFEATURES} FEATURESCONF: ${FEATURESCONF} CUSTOMPROP: ${CUSTOMPROP} LOGCONF: ${LOGCONF} MEMCONF: ${MEMCONF} CONTROLLERMEM: ${CONTROLLERMEM} AKKACONF: ${AKKACONF} MODULESCONF: ${MODULESCONF} MODULESHARDSCONF: ${MODULESHARDSCONF} SUITES: ${SUITES} EOF } print_common_env # Setup JAVA_HOME and MAX_MEM Value in ODL startup config file function set_java_vars() { local -r java_home=$1 local -r controllermem=$2 local -r memconf=$3 echo "Configure" echo " java home: ${java_home}" echo " max memory: ${controllermem}" echo " memconf: ${memconf}" # We do not want expressions to expand here. # shellcheck disable=SC2016 sed -ie 's%^# export JAVA_HOME%export JAVA_HOME=${JAVA_HOME:-'"${java_home}"'}%g' "${memconf}" sed -ie 's/JAVA_MAX_MEM="2048m"/JAVA_MAX_MEM='"${controllermem}"'/g' "${memconf}" echo "cat ${memconf}" cat "${memconf}" echo "Set Java version" sudo /usr/sbin/alternatives --install /usr/bin/java java "${java_home}/bin/java" 1 sudo /usr/sbin/alternatives --set java "${java_home}/bin/java" echo "JDK default version ..." java -version echo "Set JAVA_HOME" export JAVA_HOME="${java_home}" # shellcheck disable=SC2037 JAVA_RESOLVED=$(readlink -e "${java_home}/bin/java") echo "Java binary pointed at by JAVA_HOME: ${JAVA_RESOLVED}" } # set_java_vars() # shellcheck disable=SC2034 # foo appears unused. Verify it or export it. function configure_karaf_log() { local -r karaf_version=$1 local -r controllerdebugmap=$2 local logapi=log4j # Check what the logging.cfg file is using for the logging api: log4j or log4j2 if grep "log4j2" "${LOGCONF}"; then logapi=log4j2 fi echo "Configuring the karaf log... karaf_version: ${karaf_version}, logapi: ${logapi}" if [ "${logapi}" == "log4j2" ]; then # FIXME: Make log size limit configurable from build parameter. # Increase default log file size to 1GB sed -ie 's/log4j2.appender.rolling.policies.size.size = 64MB/log4j2.appender.rolling.policies.size.size = 1GB/g' "${LOGCONF}" orgmodule="org.opendaylight.yangtools.yang.parser.repo.YangTextSchemaContextResolver" orgmodule_="${orgmodule//./_}" echo "${logapi}.logger.${orgmodule_}.name = WARN" >> "${LOGCONF}" echo "${logapi}.logger.${orgmodule_}.level = WARN" >> "${LOGCONF}" else sed -ie 's/log4j.appender.out.maxBackupIndex=10/log4j.appender.out.maxBackupIndex=1/g' "${LOGCONF}" # FIXME: Make log size limit configurable from build parameter. sed -ie 's/log4j.appender.out.maxFileSize=1MB/log4j.appender.out.maxFileSize=30GB/g' "${LOGCONF}" echo "${logapi}.logger.org.opendaylight.yangtools.yang.parser.repo.YangTextSchemaContextResolver = WARN" >> "${LOGCONF}" fi # Add custom logging levels # CONTROLLERDEBUGMAP is expected to be a key:value map of space separated # values like "module:level module2:level2" where module is abbreviated and # does not include "org.opendaylight." unset IFS echo "controllerdebugmap: ${controllerdebugmap}" if [ -n "${controllerdebugmap}" ]; then for kv in ${controllerdebugmap}; do module="${kv%%:*}" level="${kv#*:}" echo "module: $module, level: $level" # shellcheck disable=SC2157 if [ -n "${module}" ] && [ -n "${level}" ]; then orgmodule="org.opendaylight.${module}" if [ "${logapi}" == "log4j2" ]; then orgmodule_="${orgmodule//./_}" echo "${logapi}.logger.${orgmodule_}.name = ${orgmodule}" >> "${LOGCONF}" echo "${logapi}.logger.${orgmodule_}.level = ${level}" >> "${LOGCONF}" else echo "${logapi}.logger.${orgmodule} = ${level}" >> "${LOGCONF}" fi fi done fi echo "cat ${LOGCONF}" cat "${LOGCONF}" } # function configure_karaf_log() function configure_karaf_log_for_apex() { # TODO: add the extra steps to this function to do any extra work # in this apex environment like we do in our standard environment. # EX: log size, rollover, etc. # Modify ODL Log Levels, if needed, for new distribution. This will modify # the control nodes hiera data which will be used during the puppet deploy # CONTROLLERDEBUGMAP is expected to be a key:value map of space separated # values like "module:level module2:level2" where module is abbreviated and # does not include "org.opendaylight." local -r controller_ip=$1 unset IFS # shellcheck disable=SC2153 echo "CONTROLLERDEBUGMAP: ${CONTROLLERDEBUGMAP}" if [ -n "${CONTROLLERDEBUGMAP}" ]; then logging_config='\"opendaylight::log_levels\": {' for kv in ${CONTROLLERDEBUGMAP}; do module="${kv%%:*}" level="${kv#*:}" echo "module: $module, level: $level" # shellcheck disable=SC2157 if [ -n "${module}" ] && [ -n "${level}" ]; then orgmodule="org.opendaylight.${module}" logging_config="${logging_config} \\\"${orgmodule}\\\": \\\"${level}\\\"," fi done # replace the trailing comma with a closing brace followed by trailing comma logging_config=${logging_config%,}" }," echo "$logging_config" # fine a sane line number to inject the custom logging json lineno=$(ssh "$OPENSTACK_CONTROL_NODE_1_IP" "sudo grep -Fn 'opendaylight::log_mechanism' /etc/puppet/hieradata/service_configs.json" | awk -F: '{print $1}') # We purposely want these variables to expand client-side # shellcheck disable=SC2029 ssh "$controller_ip" "sudo sed -i \"${lineno}i ${logging_config}\" /etc/puppet/hieradata/service_configs.json" ssh "$controller_ip" "sudo cat /etc/puppet/hieradata/service_configs.json" fi } # function configure_karaf_log_for_apex() function configure_odl_features_for_apex() { # if the environment variable $ACTUALFEATURES is not null, then rewrite # the puppet config file with the features given in that variable, otherwise # this function is a noop local -r controller_ip=$1 local -r config_file=/etc/puppet/hieradata/service_configs.json cat > /tmp/set_odl_features.sh << EOF sudo jq '.["opendaylight::extra_features"] |= []' $config_file > tmp.json && mv tmp.json $config_file for feature in "\${ACTUALFEATURES//,/ }"; do sudo jq --arg jq_arg \$feature '.["opendaylight::extra_features"] |= . + [\$jq_arg]' $config_file > tmp && mv tmp $config_file; done echo "Modified puppet-opendaylight service_configs.json..." cat $config_file EOF echo "Feature configuration script..." cat /tmp/set_odl_features.sh if [ -n "${ACTUALFEATURES}" ]; then scp /tmp/set_odl_features.sh "$controller_ip":/tmp/set_odl_features.sh ssh "$controller_ip" "sudo bash /tmp/set_odl_features.sh" fi } # function configure_odl_features_for_apex() function get_os_deploy() { local -r num_systems=${1:-$NUM_OPENSTACK_SYSTEM} case ${num_systems} in 1) OPENSTACK_TOPO="1cmb-0ctl-0cmp" ;; 2) OPENSTACK_TOPO="1cmb-0ctl-1cmp" ;; 3|*) OPENSTACK_TOPO="0cmb-1ctl-2cmp" ;; esac export OPENSTACK_TOPO } function get_test_suites() { #let the caller pick the name of the variable we will assign the suites to local __suite_list=$1 echo "Locating test plan to use..." testplan_filepath="${WORKSPACE}/test/csit/testplans/${STREAMTESTPLAN}" if [ ! -f "${testplan_filepath}" ]; then testplan_filepath="${WORKSPACE}/test/csit/testplans/${TESTPLAN}" fi if [ "${ELASTICSEARCHATTRIBUTE}" != "disabled" ]; then add_test="integration/test/csit/suites/integration/Create_JVM_Plots.robot" echo "${add_test}" >> "$testplan_filepath" fi echo "Changing the testplan path..." sed "s:integration:${WORKSPACE}:" "${testplan_filepath}" > testplan.txt cat testplan.txt # Use the testplan if specific SUITES are not defined. if [ -z "${SUITES}" ]; then suite_list=$(grep -E -v '(^[[:space:]]*#|^[[:space:]]*$)' testplan.txt | tr '\012' ' ') else suite_list="" workpath="${WORKSPACE}/test/csit/suites" for suite in ${SUITES}; do fullsuite="${workpath}/${suite}" if [ -z "${suite_list}" ]; then suite_list+=${fullsuite} else suite_list+=" "${fullsuite} fi done fi eval "$__suite_list='$suite_list'" } function run_plan() { local -r type=$1 case ${type} in script) plan=$SCRIPTPLAN ;; config|*) plan=$CONFIGPLAN ;; esac printf "Locating %s plan to use...\\n" "${type}" plan_filepath="${WORKSPACE}/test/csit/${type}plans/$plan" if [ ! -f "${plan_filepath}" ]; then plan_filepath="${WORKSPACE}/test/csit/${type}plans/${STREAMTESTPLAN}" if [ ! -f "${plan_filepath}" ]; then plan_filepath="${WORKSPACE}/test/csit/${type}plans/${TESTPLAN}" fi fi if [ -f "${plan_filepath}" ]; then printf "%s plan exists!!!\\n" "${type}" printf "Changing the %s plan path...\\n" "${type}" sed "s:integration:${WORKSPACE}:" "${plan_filepath}" > "${type}plan.txt" cat "${type}plan.txt" # shellcheck disable=SC2013 for line in $( grep -E -v '(^[[:space:]]*#|^[[:space:]]*$)' "${type}plan.txt" ); do printf "Executing %s...\\n" "${line}" # shellcheck source=${line} disable=SC1091 source "${line}" done fi printf "Finished running %s plans\\n" "${type}" } # function run_plan() # Run scripts to support JVM monitoring. function add_jvm_support() { if [ "${ELASTICSEARCHATTRIBUTE}" != "disabled" ]; then set_elasticsearch_attribute "${ELASTICSEARCHATTRIBUTE}" set_jvm_common_attribute fi } # function add_jvm_support() #Expected input parameter: long/short/a number function set_elasticsearch_attribute() { short=5000 long=120000 default=$short case $1 in short) period=$short ;; long) period=$long ;; *) # shellcheck disable=SC2166 if [[ "$1" =~ ^[0-9]+$ ]] && [ "$1" -ge $short -a "$1" -le $long ]; then period=$1 else period=$default fi ;; esac cat > "${WORKSPACE}"/org.apache.karaf.decanter.scheduler.simple.cfg < "${WORKSPACE}"/org.apache.karaf.decanter.collector.jmx-local.cfg < "${WORKSPACE}"/org.apache.karaf.decanter.collector.jmx-others.cfg < "${WORKSPACE}"/elasticsearch.yml < "${WORKSPACE}"/elasticsearch_startup.sh < /dev/null 2>&1 & ls -al /tmp/elasticsearch/elasticsearch-1.7.5/bin/elasticsearch EOF echo "Setup ODL_SYSTEM_IP specific config files for ${!CONTROLLERIP} " cat "${WORKSPACE}"/org.apache.karaf.decanter.collector.jmx-local.cfg cat "${WORKSPACE}"/org.apache.karaf.decanter.collector.jmx-others.cfg cat "${WORKSPACE}"/elasticsearch.yml echo "Copying config files to ${!CONTROLLERIP}" scp "${WORKSPACE}"/org.apache.karaf.decanter.collector.jmx-local.cfg "${!CONTROLLERIP}":/tmp/"${BUNDLEFOLDER}"/etc/ scp "${WORKSPACE}"/org.apache.karaf.decanter.collector.jmx-others.cfg "${!CONTROLLERIP}":/tmp/"${BUNDLEFOLDER}"/etc/ scp "${WORKSPACE}"/elasticsearch.yml "${!CONTROLLERIP}":/tmp/ ssh "${!CONTROLLERIP}" "sudo ls -al /tmp/elasticsearch/" ssh "${!CONTROLLERIP}" "sudo mv /tmp/elasticsearch.yml /tmp/elasticsearch/elasticsearch-1.7.5/config/" ssh "${!CONTROLLERIP}" "cat /tmp/elasticsearch/elasticsearch-1.7.5/config/elasticsearch.yml" echo "Copying the elasticsearch_startup script to ${!CONTROLLERIP}" cat "${WORKSPACE}"/elasticsearch_startup.sh scp "${WORKSPACE}"/elasticsearch_startup.sh "${!CONTROLLERIP}":/tmp ssh "${!CONTROLLERIP}" 'bash /tmp/elasticsearch_startup.sh' ssh "${!CONTROLLERIP}" 'ps aux | grep elasticsearch' done } #function set_jvm_common_attribute # Return elapsed time. Usage: # - Call first time with no arguments and a new timer is returned. # - Next call with the first argument as the timer and the elapsed time is returned. function timer() { if [ $# -eq 0 ]; then # return the current time printf "%s" "$(date "+%s")" else local start_time=$1 end_time=$(date "+%s") if [ -z "$start_time" ]; then start_time=$end_time; fi delta_time=$((end_time - start_time)) ds=$((delta_time % 60)) dm=$(((delta_time / 60) % 60)) dh=$((delta_time / 3600)) # return the elapsed time printf "%d:%02d:%02d" $dh $dm $ds fi } # convert commas in csv strings to spaces (ssv) function csv2ssv() { local csv=$1 if [ -n "${csv}" ]; then ssv=$(echo "${csv}" | sed 's/,/ /g' | sed 's/\ \ */\ /g') fi echo "${ssv}" } # csv2ssv function is_openstack_feature_enabled() { local feature=$1 for enabled_feature in $(csv2ssv "${ENABLE_OS_SERVICES}"); do if [ "${enabled_feature}" == "${feature}" ]; then echo 1 return fi done echo 0 } SSH="ssh -t -t" # shellcheck disable=SC2153 function print_job_parameters() { cat << EOF Job parameters: DISTROBRANCH: ${DISTROBRANCH} DISTROSTREAM: ${DISTROSTREAM} BUNDLE_URL: ${BUNDLE_URL} CONTROLLERFEATURES: ${CONTROLLERFEATURES} CONTROLLERDEBUGMAP: ${CONTROLLERDEBUGMAP} CONTROLLERMAXMEM: ${CONTROLLERMAXMEM} SCRIPTPLAN: ${SCRIPTPLAN} CONFIGPLAN: ${CONFIGPLAN} STREAMTESTPLAN: ${STREAMTESTPLAN} TESTPLAN: ${TESTPLAN} SUITES: ${SUITES} PATCHREFSPEC: ${PATCHREFSPEC} OPENSTACK_BRANCH: ${OPENSTACK_BRANCH} DEVSTACK_HASH: ${DEVSTACK_HASH} ODL_ML2_DRIVER_REPO: ${ODL_ML2_DRIVER_REPO} ODL_ML2_BRANCH: ${ODL_ML2_BRANCH} ODL_ML2_DRIVER_VERSION: ${ODL_ML2_DRIVER_VERSION} ODL_ML2_PORT_BINDING: ${ODL_ML2_PORT_BINDING} DEVSTACK_KUBERNETES_PLUGIN_REPO: ${DEVSTACK_KUBERNETES_PLUGIN_REPO} DEVSTACK_LBAAS_PLUGIN_REPO: ${DEVSTACK_LBAAS_PLUGIN_REPO} DEVSTACK_NETWORKING_SFC_PLUGIN_REPO: ${DEVSTACK_NETWORKING_SFC_PLUGIN_REPO} IPSEC_VXLAN_TUNNELS_ENABLED: ${IPSEC_VXLAN_TUNNELS_ENABLED} PUBLIC_BRIDGE: ${PUBLIC_BRIDGE} ENABLE_HAPROXY_FOR_NEUTRON: ${ENABLE_HAPROXY_FOR_NEUTRON} ENABLE_OS_SERVICES: ${ENABLE_OS_SERVICES} ENABLE_OS_COMPUTE_SERVICES: ${ENABLE_OS_COMPUTE_SERVICES} ENABLE_OS_NETWORK_SERVICES: ${ENABLE_OS_NETWORK_SERVICES} ENABLE_OS_PLUGINS: ${ENABLE_OS_PLUGINS} DISABLE_OS_SERVICES: ${DISABLE_OS_SERVICES} TENANT_NETWORK_TYPE: ${TENANT_NETWORK_TYPE} SECURITY_GROUP_MODE: ${SECURITY_GROUP_MODE} ENABLE_ITM_DIRECT_TUNNELS: ${ENABLE_ITM_DIRECT_TUNNELS} PUBLIC_PHYSICAL_NETWORK: ${PUBLIC_PHYSICAL_NETWORK} ENABLE_NETWORKING_L2GW: ${ENABLE_NETWORKING_L2GW} CREATE_INITIAL_NETWORKS: ${CREATE_INITIAL_NETWORKS} LBAAS_SERVICE_PROVIDER: ${LBAAS_SERVICE_PROVIDER} ODL_SFC_DRIVER: ${ODL_SFC_DRIVER} ODL_SNAT_MODE: ${ODL_SNAT_MODE} GROUP_ADD_MOD_ENABLED: ${GROUP_ADD_MOD_ENABLED} EOF } function tcpdump_start() { local -r prefix=$1 local -r ip=$2 local -r filter=$3 filter_=${filter// /_} printf "node %s, %s_%s__%s: starting tcpdump\\n" "${ip}" "${prefix}" "${ip}" "${filter}" # $fileter needs to be parsed client-side # shellcheck disable=SC2029 ssh "${ip}" "nohup sudo /usr/sbin/tcpdump -vvv -ni eth0 ${filter} -w /tmp/tcpdump_${prefix}_${ip}__${filter_}.pcap > /tmp/tcpdump_start.log 2>&1 &" ${SSH} "${ip}" "ps -ef | grep tcpdump" } function tcpdump_stop() { local -r ip=$1 printf "node %s: stopping tcpdump\\n" "$ip" ${SSH} "${ip}" "ps -ef | grep tcpdump.sh" ${SSH} "${ip}" "sudo pkill -f tcpdump" ${SSH} "${ip}" "sudo xz -9ekvvf /tmp/*.pcap" ${SSH} "${ip}" "sudo ls -al /tmp/*.pcap" # copy_logs will copy any *.xz files } # Collect the list of files on the hosts function collect_files() { local -r ip=$1 local -r folder=$2 finddir=/tmp/finder ${SSH} "${ip}" "mkdir -p ${finddir}" ${SSH} "${ip}" "sudo find /etc > ${finddir}/find.etc.txt" ${SSH} "${ip}" "sudo find /opt/stack > ${finddir}/find.opt.stack.txt" ${SSH} "${ip}" "sudo find /var > ${finddir}/find2.txt" ${SSH} "${ip}" "sudo find /var > ${finddir}/find.var.txt" ${SSH} "${ip}" "sudo tar -cf - -C /tmp finder | xz -T 0 > /tmp/find.tar.xz" scp "${ip}":/tmp/find.tar.xz "${folder}" mkdir -p "${finddir}" rsync --rsync-path="sudo rsync" --list-only -arvhe ssh "${ip}":/etc/ > "${finddir}"/rsync.etc.txt rsync --rsync-path="sudo rsync" --list-only -arvhe ssh "${ip}":/opt/stack/ > "${finddir}"/rsync.opt.stack.txt rsync --rsync-path="sudo rsync" --list-only -arvhe ssh "${ip}":/var/ > "${finddir}"/rsync.var.txt tar -cf - -C /tmp finder | xz -T 0 > /tmp/rsync.tar.xz cp /tmp/rsync.tar.xz "${folder}" } # List of extra services to extract from journalctl # Add new services on a separate line, in alpha order, add \ at the end extra_services_cntl=" \ dnsmasq.service \ httpd.service \ libvirtd.service \ openvswitch.service \ ovs-vswitchd.service \ ovsdb-server.service \ rabbitmq-server.service \ " extra_services_cmp=" \ libvirtd.service \ openvswitch.service \ ovs-vswitchd.service \ ovsdb-server.service \ " # Collect the logs for the openstack services # First get all the services started by devstack which would have devstack@ as a prefix # Next get all the extra services function collect_openstack_logs() { local -r ip=${1} local -r folder=${2} local -r node_type=${3} local oslogs="${folder}/oslogs" printf "collect_openstack_logs for %s node: %s into %s\\n" "${node_type}" "${ip}" "${oslogs}" rm -rf "${oslogs}" mkdir -p "${oslogs}" # There are always some logs in /opt/stack/logs and this also covers the # pre-queens branches which always use /opt/stack/logs rsync -avhe ssh "${ip}":/opt/stack/logs/* "${oslogs}" # rsync to prevent copying of symbolic links # Starting with queens break out the logs from journalctl if [ "${OPENSTACK_BRANCH}" = "stable/queens" ]; then cat > "${WORKSPACE}"/collect_openstack_logs.sh << EOF extra_services_cntl="${extra_services_cntl}" extra_services_cmp="${extra_services_cmp}" function extract_from_journal() { local -r services=\${1} local -r folder=\${2} local -r node_type=\${3} printf "extract_from_journal folder: \${folder}, services: \${services}\\n" for service in \${services}; do # strip anything before @ and anything after . # devstack@g-api.service will end as g-api service_="\${service#*@}" service_="\${service_%.*}" sudo journalctl -u "\${service}" > "\${folder}/\${service_}.log" done } rm -rf /tmp/oslogs mkdir -p /tmp/oslogs systemctl list-unit-files --all > /tmp/oslogs/systemctl.units.log 2>&1 svcs=\$(grep devstack@ /tmp/oslogs/systemctl.units.log | awk '{print \$1}') extract_from_journal "\${svcs}" "/tmp/oslogs" if [ "\${node_type}" = "control" ]; then extract_from_journal "\${extra_services_cntl}" "/tmp/oslogs" else extract_from_journal "\${extra_services_cmp}" "/tmp/oslogs" fi ls -al /tmp/oslogs EOF # cat > ${WORKSPACE}/collect_openstack_logs.sh << EOF printf "collect_openstack_logs for %s node: %s into %s, executing script\\n" "${node_type}" "${ip}" "${oslogs}" cat "${WORKSPACE}"/collect_openstack_logs.sh scp "${WORKSPACE}"/collect_openstack_logs.sh "${ip}":/tmp ${SSH} "${ip}" "bash /tmp/collect_openstack_logs.sh > /tmp/collect_openstack_logs.log 2>&1" rsync -avhe ssh "${ip}":/tmp/oslogs/* "${oslogs}" scp "${ip}":/tmp/collect_openstack_logs.log "${oslogs}" fi # if [ "${OPENSTACK_BRANCH}" = "stable/queens" ]; then } function collect_netvirt_logs() { set +e # We do not want to create red dot just because something went wrong while fetching logs. cat > extra_debug.sh << EOF echo -e "/usr/sbin/lsmod | /usr/bin/grep openvswitch\\n" /usr/sbin/lsmod | /usr/bin/grep openvswitch echo -e "\\nsudo grep ct_ /var/log/openvswitch/ovs-vswitchd.log\\n" sudo grep "Datapath supports" /var/log/openvswitch/ovs-vswitchd.log echo -e "\\nsudo netstat -punta\\n" sudo netstat -punta echo -e "\\nsudo getenforce\\n" sudo getenforce echo -e "\\nsudo systemctl status httpd\\n" sudo systemctl status httpd echo -e "\\nenv\\n" env source /opt/stack/devstack/openrc admin admin echo -e "\\nenv after openrc\\n" env echo -e "\\nsudo du -hs /opt/stack" sudo du -hs /opt/stack echo -e "\\nsudo mount" sudo mount echo -e "\\ndmesg -T > /tmp/dmesg.log" dmesg -T > /tmp/dmesg.log echo -e "\\njournalctl > /tmp/journalctl.log\\n" sudo journalctl > /tmp/journalctl.log echo -e "\\novsdb-tool -mm show-log > /tmp/ovsdb-tool.log" ovsdb-tool -mm show-log > /tmp/ovsdb-tool.log EOF # Since this log collection work is happening before the archive build macro which also # creates the ${WORKSPACE}/archives dir, we have to do it here first. The mkdir in the # archives build step will essentially be a noop. mkdir -p "${WORKSPACE}"/archives mv /tmp/changes.txt "${WORKSPACE}"/archives mv /tmp/validations.txt "${WORKSPACE}"/archives mv "${WORKSPACE}"/rabbit.txt "${WORKSPACE}"/archives mv "${WORKSPACE}"/haproxy.cfg "${WORKSPACE}"/archives ssh "${OPENSTACK_HAPROXY_1_IP}" "sudo journalctl -u haproxy > /tmp/haproxy.log" scp "${OPENSTACK_HAPROXY_1_IP}":/tmp/haproxy.log "${WORKSPACE}"/archives/ sleep 5 # FIXME: Do not create .tar and gzip before copying. for i in $(seq 1 "${NUM_ODL_SYSTEM}"); do CONTROLLERIP=ODL_SYSTEM_${i}_IP echo "collect_logs: for opendaylight controller ip: ${!CONTROLLERIP}" NODE_FOLDER="odl_${i}" mkdir -p "${NODE_FOLDER}" echo "Let's take the karaf thread dump again..." ssh "${!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}" # $pid needs to be parsed client-side # shellcheck disable=SC2029 ssh "${!CONTROLLERIP}" "${JAVA_HOME}/bin/jstack -l ${pid}" > "${WORKSPACE}/karaf_${i}_${pid}_threads_after.log" || true echo "killing karaf process..." # shellcheck disable=SC2016 ${SSH} "${!CONTROLLERIP}" bash -c 'ps axf | grep karaf | grep -v grep | awk '"'"'{print "kill -9 " $1}'"'"' | sh' ${SSH} "${!CONTROLLERIP}" "sudo journalctl > /tmp/journalctl.log" scp "${!CONTROLLERIP}":/tmp/journalctl.log "${NODE_FOLDER}" ${SSH} "${!CONTROLLERIP}" "dmesg -T > /tmp/dmesg.log" scp "${!CONTROLLERIP}":/tmp/dmesg.log "${NODE_FOLDER}" ${SSH} "${!CONTROLLERIP}" "tar -cf - -C /tmp/${BUNDLEFOLDER} etc | xz -T 0 > /tmp/etc.tar.xz" scp "${!CONTROLLERIP}":/tmp/etc.tar.xz "${NODE_FOLDER}" ${SSH} "${!CONTROLLERIP}" "cp -r /tmp/${BUNDLEFOLDER}/data/log /tmp/odl_log" ${SSH} "${!CONTROLLERIP}" "tar -cf /tmp/odl${i}_karaf.log.tar /tmp/odl_log/*" scp "${!CONTROLLERIP}:/tmp/odl${i}_karaf.log.tar" "${NODE_FOLDER}" ${SSH} "${!CONTROLLERIP}" "tar -cf /tmp/odl${i}_zrpcd.log.tar /tmp/zrpcd.init.log" scp "${!CONTROLLERIP}:/tmp/odl${i}_zrpcd.log.tar" "${NODE_FOLDER}" tar -xvf "${NODE_FOLDER}/odl${i}_karaf.log.tar" -C "${NODE_FOLDER}" --strip-components 2 --transform "s/karaf/odl${i}_karaf/g" grep "ROBOT MESSAGE\\| ERROR " "${NODE_FOLDER}/odl${i}_karaf.log" > "${NODE_FOLDER}/odl${i}_err.log" grep "ROBOT MESSAGE\\| ERROR \\| WARN \\|Exception" \ "${NODE_FOLDER}/odl${i}_karaf.log" > "${NODE_FOLDER}/odl${i}_err_warn_exception.log" # Print ROBOT lines and print Exception lines. For exception lines also print the previous line for context sed -n -e '/ROBOT MESSAGE/P' -e '$!N;/Exception/P;D' "${NODE_FOLDER}/odl${i}_karaf.log" > "${NODE_FOLDER}/odl${i}_exception.log" mv "/tmp/odl${i}_exceptions.txt" "${NODE_FOLDER}" rm "${NODE_FOLDER}/odl${i}_karaf.log.tar" mv -- *_threads* "${NODE_FOLDER}" mv ps_* "${NODE_FOLDER}" mv "${NODE_FOLDER}" "${WORKSPACE}"/archives/ done print_job_parameters > "${WORKSPACE}"/archives/params.txt # Control Node for i in $(seq 1 "${NUM_OPENSTACK_CONTROL_NODES}"); do OSIP=OPENSTACK_CONTROL_NODE_${i}_IP if [ "$(is_openstack_feature_enabled n-cpu)" == "1" ]; then echo "collect_logs: for openstack combo node ip: ${!OSIP}" NODE_FOLDER="combo_${i}" else echo "collect_logs: for openstack control node ip: ${!OSIP}" NODE_FOLDER="control_${i}" fi mkdir -p "${NODE_FOLDER}" tcpdump_stop "${!OSIP}" scp extra_debug.sh "${!OSIP}":/tmp # Capture compute logs if this is a combo node if [ "$(is_openstack_feature_enabled n-cpu)" == "1" ]; then scp "${!OSIP}":/etc/nova/nova.conf "${NODE_FOLDER}" scp "${!OSIP}":/etc/nova/nova-cpu.conf "${NODE_FOLDER}" scp "${!OSIP}":/etc/openstack/clouds.yaml "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/var/log/nova-agent.log "${NODE_FOLDER}" fi ${SSH} "${!OSIP}" "bash /tmp/extra_debug.sh > /tmp/extra_debug.log 2>&1" scp "${!OSIP}":/etc/dnsmasq.conf "${NODE_FOLDER}" scp "${!OSIP}":/etc/keystone/keystone.conf "${NODE_FOLDER}" scp "${!OSIP}":/etc/keystone/keystone-uwsgi-admin.ini "${NODE_FOLDER}" scp "${!OSIP}":/etc/keystone/keystone-uwsgi-public.ini "${NODE_FOLDER}" scp "${!OSIP}":/etc/kuryr/kuryr.conf "${NODE_FOLDER}" scp "${!OSIP}":/etc/neutron/dhcp_agent.ini "${NODE_FOLDER}" scp "${!OSIP}":/etc/neutron/metadata_agent.ini "${NODE_FOLDER}" scp "${!OSIP}":/etc/neutron/neutron.conf "${NODE_FOLDER}" scp "${!OSIP}":/etc/neutron/neutron_lbaas.conf "${NODE_FOLDER}" scp "${!OSIP}":/etc/neutron/plugins/ml2/ml2_conf.ini "${NODE_FOLDER}" scp "${!OSIP}":/etc/neutron/services/loadbalancer/haproxy/lbaas_agent.ini "${NODE_FOLDER}" scp "${!OSIP}":/etc/nova/nova.conf "${NODE_FOLDER}" scp "${!OSIP}":/etc/nova/nova-api-uwsgi.ini "${NODE_FOLDER}" scp "${!OSIP}":/etc/nova/nova_cell1.conf "${NODE_FOLDER}" scp "${!OSIP}":/etc/nova/nova-cpu.conf "${NODE_FOLDER}" scp "${!OSIP}":/etc/nova/placement-uwsgi.ini "${NODE_FOLDER}" scp "${!OSIP}":/etc/openstack/clouds.yaml "${NODE_FOLDER}" scp "${!OSIP}":/opt/stack/devstack/.stackenv "${NODE_FOLDER}" scp "${!OSIP}":/opt/stack/devstack/nohup.out "${NODE_FOLDER}"/stack.log scp "${!OSIP}":/opt/stack/devstack/openrc "${NODE_FOLDER}" scp "${!OSIP}":/opt/stack/requirements/upper-constraints.txt "${NODE_FOLDER}" scp "${!OSIP}":/opt/stack/tempest/etc/tempest.conf "${NODE_FOLDER}" scp "${!OSIP}":/tmp/*.xz "${NODE_FOLDER}" scp "${!OSIP}":/tmp/dmesg.log "${NODE_FOLDER}" scp "${!OSIP}":/tmp/extra_debug.log "${NODE_FOLDER}" scp "${!OSIP}":/tmp/get_devstack.sh.txt "${NODE_FOLDER}" scp "${!OSIP}":/tmp/install_ovs.txt "${NODE_FOLDER}" scp "${!OSIP}":/tmp/journalctl.log "${NODE_FOLDER}" scp "${!OSIP}":/tmp/ovsdb-tool.log "${NODE_FOLDER}" scp "${!OSIP}":/tmp/tcpdump_start.log "${NODE_FOLDER}" collect_files "${!OSIP}" "${NODE_FOLDER}" ${SSH} "${!OSIP}" "sudo tar -cf - -C /var/log rabbitmq | xz -T 0 > /tmp/rabbitmq.tar.xz " scp "${!OSIP}":/tmp/rabbitmq.tar.xz "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/etc/hosts "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/usr/lib/systemd/system/haproxy.service "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/var/log/audit/audit.log "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/var/log/httpd/keystone_access.log "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/var/log/httpd/keystone.log "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/var/log/messages* "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/var/log/openvswitch/ovs-vswitchd.log "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/var/log/openvswitch/ovsdb-server.log "${NODE_FOLDER}" collect_openstack_logs "${!OSIP}" "${NODE_FOLDER}" "control" mv "local.conf_control_${!OSIP}" "${NODE_FOLDER}/local.conf" # qdhcp files are created by robot tests and copied into /tmp/qdhcp during the test tar -cf - -C /tmp qdhcp | xz -T 0 > /tmp/qdhcp.tar.xz mv /tmp/qdhcp.tar.xz "${NODE_FOLDER}" mv "${NODE_FOLDER}" "${WORKSPACE}"/archives/ done # Compute Nodes for i in $(seq 1 "${NUM_OPENSTACK_COMPUTE_NODES}"); do OSIP="OPENSTACK_COMPUTE_NODE_${i}_IP" echo "collect_logs: for openstack compute node ip: ${!OSIP}" NODE_FOLDER="compute_${i}" mkdir -p "${NODE_FOLDER}" tcpdump_stop "${!OSIP}" scp extra_debug.sh "${!OSIP}":/tmp ${SSH} "${!OSIP}" "bash /tmp/extra_debug.sh > /tmp/extra_debug.log 2>&1" scp "${!OSIP}":/etc/nova/nova.conf "${NODE_FOLDER}" scp "${!OSIP}":/etc/nova/nova-cpu.conf "${NODE_FOLDER}" scp "${!OSIP}":/etc/openstack/clouds.yaml "${NODE_FOLDER}" scp "${!OSIP}":/opt/stack/devstack/.stackenv "${NODE_FOLDER}" scp "${!OSIP}":/opt/stack/devstack/nohup.out "${NODE_FOLDER}"/stack.log scp "${!OSIP}":/opt/stack/devstack/openrc "${NODE_FOLDER}" scp "${!OSIP}":/opt/stack/requirements/upper-constraints.txt "${NODE_FOLDER}" scp "${!OSIP}":/tmp/*.xz "${NODE_FOLDER}"/ scp "${!OSIP}":/tmp/dmesg.log "${NODE_FOLDER}" scp "${!OSIP}":/tmp/extra_debug.log "${NODE_FOLDER}" scp "${!OSIP}":/tmp/get_devstack.sh.txt "${NODE_FOLDER}" scp "${!OSIP}":/tmp/install_ovs.txt "${NODE_FOLDER}" scp "${!OSIP}":/tmp/journalctl.log "${NODE_FOLDER}" scp "${!OSIP}":/tmp/ovsdb-tool.log "${NODE_FOLDER}" scp "${!OSIP}":/tmp/tcpdump_start.log "${NODE_FOLDER}" collect_files "${!OSIP}" "${NODE_FOLDER}" ${SSH} "${!OSIP}" "sudo tar -cf - -C /var/log libvirt | xz -T 0 > /tmp/libvirt.tar.xz " scp "${!OSIP}":/tmp/libvirt.tar.xz "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/etc/hosts "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/var/log/audit/audit.log "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/var/log/messages* "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/var/log/nova-agent.log "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/var/log/openvswitch/ovs-vswitchd.log "${NODE_FOLDER}" rsync --rsync-path="sudo rsync" -avhe ssh "${!OSIP}":/var/log/openvswitch/ovsdb-server.log "${NODE_FOLDER}" collect_openstack_logs "${!OSIP}" "${NODE_FOLDER}" "compute" mv "local.conf_compute_${!OSIP}" "${NODE_FOLDER}"/local.conf mv "${NODE_FOLDER}" "${WORKSPACE}"/archives/ done # Tempest DEVSTACK_TEMPEST_DIR="/opt/stack/tempest" TESTREPO=".stestr" TEMPEST_LOGS_DIR="${WORKSPACE}/archives/tempest" # Look for tempest test results in the $TESTREPO dir and copy if found if ${SSH} "${OPENSTACK_CONTROL_NODE_1_IP}" "sudo sh -c '[ -f ${DEVSTACK_TEMPEST_DIR}/${TESTREPO}/0 ]'"; then ${SSH} "${OPENSTACK_CONTROL_NODE_1_IP}" "for I in \$(sudo ls ${DEVSTACK_TEMPEST_DIR}/${TESTREPO}/ | grep -E '^[0-9]+$'); do sudo sh -c \"${DEVSTACK_TEMPEST_DIR}/.tox/tempest/bin/subunit-1to2 < ${DEVSTACK_TEMPEST_DIR}/${TESTREPO}/\${I} >> ${DEVSTACK_TEMPEST_DIR}/subunit_log.txt\"; done" ${SSH} "${OPENSTACK_CONTROL_NODE_1_IP}" "sudo sh -c '${DEVSTACK_TEMPEST_DIR}/.tox/tempest/bin/python ${DEVSTACK_TEMPEST_DIR}/.tox/tempest/lib/python2.7/site-packages/os_testr/subunit2html.py ${DEVSTACK_TEMPEST_DIR}/subunit_log.txt ${DEVSTACK_TEMPEST_DIR}/tempest_results.html'" mkdir -p "${TEMPEST_LOGS_DIR}" scp "${OPENSTACK_CONTROL_NODE_1_IP}:${DEVSTACK_TEMPEST_DIR}/tempest_results.html" "${TEMPEST_LOGS_DIR}" scp "${OPENSTACK_CONTROL_NODE_1_IP}:${DEVSTACK_TEMPEST_DIR}/tempest.log" "${TEMPEST_LOGS_DIR}" else echo "tempest results not found in ${DEVSTACK_TEMPEST_DIR}/${TESTREPO}/0" fi } # collect_netvirt_logs() # Utility function for joining strings. function join() { delim=' ' final=$1; shift for str in "$@" ; do final=${final}${delim}${str} done echo "${final}" } function get_nodes_list() { # Create the string for nodes for i in $(seq 1 "${NUM_ODL_SYSTEM}") ; do CONTROLLERIP=ODL_SYSTEM_${i}_IP nodes[$i]=${!CONTROLLERIP} done nodes_list=$(join "${nodes[@]}") echo "${nodes_list}" } function get_features() { if [ "${CONTROLLERSCOPE}" == 'all' ]; then ACTUALFEATURES="odl-integration-compatible-with-all,${CONTROLLERFEATURES}" # if CONTROLLERMEM still is the default 2G and was not overridden by a # custom job, then we need to make sure to increase it because "all" # features can be heavy if [ "${CONTROLLERMEM}" == "2048m" ]; then export CONTROLLERMEM="3072m" fi else ACTUALFEATURES="odl-infrautils-ready,${CONTROLLERFEATURES}" fi if [ "${ELASTICSEARCHATTRIBUTE}" != "disabled" ]; then # Add decanter features to allow JVM monitoring ACTUALFEATURES="${ACTUALFEATURES},decanter-collector-jmx,decanter-appender-elasticsearch-rest" fi # Some versions of jenkins job builder result in feature list containing spaces # and ending in newline. Remove all that. ACTUALFEATURES=$(echo "${ACTUALFEATURES}" | tr -d '\n \r') echo "ACTUALFEATURES: ${ACTUALFEATURES}" # In the case that we want to install features via karaf shell, a space separated list of # ACTUALFEATURES IS NEEDED SPACE_SEPARATED_FEATURES=$(echo "${ACTUALFEATURES}" | tr ',' ' ') echo "SPACE_SEPARATED_FEATURES: ${SPACE_SEPARATED_FEATURES}" export ACTUALFEATURES export SPACE_SEPARATED_FEATURES } # Create the configuration script to be run on controllers. function create_configuration_script() { cat > "${WORKSPACE}"/configuration-script.sh < ${WORKSPACE}/configuration-script.sh < "${WORKSPACE}"/startup-script.sh < ${WORKSPACE}/startup-script.sh < "${WORKSPACE}"/post-startup-script.sh < ${WORKSPACE}/post-startup-script.sh < "${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}" # $i needs to be parsed client-side # shellcheck disable=SC2029 ssh "${!CONTROLLERIP}" "${JAVA_HOME}/bin/jstack -l ${pid}" > "${WORKSPACE}/karaf_${i}_${pid}_threads_before.log" || true done } # Function to build OVS from git repo function build_ovs() { local -r ip=$1 local -r version=$2 local -r rpm_path="$3" echo "Building OVS ${version} on ${ip} ..." cat > "${WORKSPACE}"/build_ovs.sh << EOF set -ex -o pipefail echo '---> Building openvswitch version ${version}' # Install running kernel devel packages K_VERSION=\$(uname -r) YUM_OPTS="-y --disablerepo=* --enablerepo=base,updates,extra,C*-base,C*-updates,C*-extras" # Install centos-release to update vault repos from which to fetch # kernel devel packages sudo yum \${YUM_OPTS} install centos-release yum-utils @'Development Tools' rpm-build sudo yum \${YUM_OPTS} install kernel-{devel,headers}-\${K_VERSION} TMP=\$(mktemp -d) pushd \${TMP} git clone https://github.com/openvswitch/ovs.git cd ovs if [ "${version}" = "v2.6.1-nsh" ]; then git checkout v2.6.1 echo "Will apply nsh patches for OVS version 2.6.1" git clone https://github.com/yyang13/ovs_nsh_patches.git ../ovs_nsh_patches git apply ../ovs_nsh_patches/v2.6.1_centos7/*.patch else git checkout ${version} fi # On early versions of OVS, flake warnings would fail the build. # Remove it. sudo pip uninstall -y flake8 # Get rid of sphinx dep as it conflicts with the already # installed one (via pip). Docs wont be built. sed -i "/BuildRequires:.*sphinx.*/d" rhel/openvswitch-fedora.spec.in sed -e 's/@VERSION@/0.0.1/' rhel/openvswitch-fedora.spec.in > /tmp/ovs.spec sed -e 's/@VERSION@/0.0.1/' rhel/openvswitch-kmod-fedora.spec.in > /tmp/ovs-kmod.spec sed -e 's/@VERSION@/0.0.1/' rhel/openvswitch-dkms.spec.in > /tmp/ovs-dkms.spec sudo yum-builddep \${YUM_OPTS} /tmp/ovs.spec /tmp/ovs-kmod.spec /tmp/ovs-dkms.spec rm /tmp/ovs.spec /tmp/ovs-kmod.spec /tmp/ovs-dkms.spec ./boot.sh ./configure --build=x86_64-redhat-linux-gnu --host=x86_64-redhat-linux-gnu --with-linux=/lib/modules/\${K_VERSION}/build --program-prefix= --disable-dependency-tracking --prefix=/usr --exec-prefix=/usr --bindir=/usr/bin --sbindir=/usr/sbin --sysconfdir=/etc --datadir=/usr/share --includedir=/usr/include --libdir=/usr/lib64 --libexecdir=/usr/libexec --localstatedir=/var --sharedstatedir=/var/lib --mandir=/usr/share/man --infodir=/usr/share/info --enable-libcapng --enable-ssl --with-pkidir=/var/lib/openvswitch/pki PYTHON=/usr/bin/python2 make rpm-fedora RPMBUILD_OPT="--without check" # Build dkms only for now # make rpm-fedora-kmod RPMBUILD_OPT='-D "kversion \${K_VERSION}"' rpmbuild -D "_topdir \$(pwd)/rpm/rpmbuild" -bb --without check rhel/openvswitch-dkms.spec mkdir -p /tmp/ovs_rpms cp -r rpm/rpmbuild/RPMS/* /tmp/ovs_rpms/ popd rm -rf \${TMP} EOF scp "${WORKSPACE}"/build_ovs.sh "${ip}":/tmp ${SSH} "${ip}" " bash /tmp/build_ovs.sh >> /tmp/install_ovs.txt 2>&1" scp -r "${ip}":/tmp/ovs_rpms/* "${rpm_path}/" ${SSH} "${ip}" "rm -rf /tmp/ovs_rpms" } # Install OVS RPMs from yum repo function install_ovs_from_repo() { local -r ip=$1 local -r rpm_repo="$2" echo "Installing OVS from repo ${rpm_repo} on ${ip} ..." cat > "${WORKSPACE}"/install_ovs.sh << EOF set -ex -o pipefail echo '---> Installing openvswitch from ${rpm_repo}' # We need repoquery from yum-utils. sudo yum -y install yum-utils # Get openvswitch packages offered by custom repo. # dkms package will have priority over kmod. OVS_REPO_OPTS="--repofrompath=ovs-repo,${rpm_repo} --disablerepo=* --enablerepo=ovs-repo" OVS_PKGS=\$(repoquery \${OVS_REPO_OPTS} openvswitch) OVS_SEL_PKG=\$(repoquery \${OVS_REPO_OPTS} openvswitch-selinux-policy) OVS_DKMS_PKG=\$(repoquery \${OVS_REPO_OPTS} openvswitch-dkms) OVS_KMOD_PKG=\$(repoquery \${OVS_REPO_OPTS} openvswitch-kmod) [ -n "\${OVS_SEL_PKG}" ] && OVS_PKGS="\${OVS_PKGS} \${OVS_SEL_PKG}" [ -n "\${OVS_DKMS_PKG}" ] && OVS_PKGS="\${OVS_PKGS} \${OVS_DKMS_PKG}" [ -z "\${OVS_DKMS_PKG}" ] && [ -n "\${OVS_KMOD_PKG}" ] && OVS_PKGS="\${OVS_PKGS} \${OVS_KMOD_PKG}" # Bail with error if custom repo was provided but we could not # find suitable packages there. [ -z "\${OVS_PKGS}" ] && echo "No OVS packages found in custom repo." && exit 1 # Install kernel & devel packages for the openvswitch dkms package. if [ -n "\${OVS_DKMS_PKG}" ]; then # install centos-release to update vault repos from which to fetch # kernel devel packages sudo yum -y install centos-release K_VERSION=\$(uname -r) YUM_OPTS="-y --disablerepo=* --enablerepo=base,updates,extra,C*-base,C*-updates,C*-extras" sudo yum \${YUM_OPTS} install kernel-{headers,devel}-\${K_VERSION} @'Development Tools' python-six fi PREV_MOD=\$(sudo modinfo -n openvswitch || echo '') # Install OVS offered by custom repo. sudo yum-config-manager --add-repo "${rpm_repo}" sudo yum -y versionlock delete openvswitch-* sudo yum -y remove openvswitch-* sudo yum -y --nogpgcheck install \${OVS_PKGS} sudo yum -y versionlock add \${OVS_PKGS} # Most recent OVS versions have some incompatibility with certain versions of iptables # This below line will overcome that problem. sudo modprobe openvswitch # Start OVS and print details sudo systemctl start openvswitch sudo systemctl enable openvswitch sudo ovs-vsctl --retry -t 5 show sudo modinfo openvswitch # dkms rpm install can fail silently (probably because the OVS version is # incompatible with the running kernel), verify module was updated. NEW_MOD=\$(sudo modinfo -n openvswitch || echo '') [ "\${PREV_MOD}" != "\${NEW_MOD}" ] || (echo "Kernel module was not updated" && exit 1) EOF scp "${WORKSPACE}"/install_ovs.sh "${ip}":/tmp ${SSH} "${ip}" "bash /tmp/install_ovs.sh >> /tmp/install_ovs.txt 2>&1" } # Install OVS RPMS from path function install_ovs_from_path() { local -r ip=$1 local -r rpm_path="$2" echo "Creating OVS RPM repo on ${ip} ..." ${SSH} "${ip}" "mkdir -p /tmp/ovs_rpms" scp -r "${rpm_path}"/* "${ip}":/tmp/ovs_rpms ${SSH} "${ip}" "sudo yum -y install createrepo && createrepo --database /tmp/ovs_rpms" install_ovs_from_repo "${ip}" file:/tmp/ovs_rpms }