Reduce NDT plugin size by removing MOFED.

plugins/UFM_NDT_Plugin/.ci/ci_matrix.yaml

@@ -11,6 +11,7 @@ kubernetes:
 
 credentials:
   - {credentialsId: '2c8fd3fe-df05-4f6f-9d1f-f1896d611434', usernameVariable: 'REST_USER', passwordVariable: 'PASSWORD'}
+  - {credentialsId: 'e4ba8d7a-d304-4c01-aff9-90de67d97efc', usernameVariable: 'GH_USER', passwordVariable: 'GH_TOKEN'}
 
 volumes:
   - {mountPath: /var/run/docker.sock, hostPath: /var/run/docker.sock}
@@ -55,7 +56,7 @@ steps:
     agentSelector: "{nodeLabel: 'UFM-POC'}"
     run: |
       cd plugins/UFM_NDT_Plugin/build
-      bash -x ./docker_build.sh
+      sudo bash -x ./docker_build.sh
       cp ufm-plugin-ndt* /auto/UFM/tmp/${JOB_NAME}/${BUILD_ID}/
     parallel: false
 
@@ -88,24 +89,63 @@ steps:
     run: plugins/UFM_NDT_Plugin/.ci/do_add_plugin.sh
     parallel: false
 
-
-  - name: start simulation
+  - name: start simulator
     agentSelector: "{nodeLabel: 'UFM-POC'}"
-    run: | 
-      cd /auto/UFM/tmp/ufm_verification/tests/test_with_sim/performance
-      IP_SERVER=$(ping $SERVER_HOST -c 1 -s 16 | grep -oE '[0-9]+\.[0-9]+\.[0-9]+\.[0-9]+' | head -1)
-      python run_simulator.py -s $IP_SERVER --topology_path ibnetdiscover_director.txt
+    credentialsId: 'e4ba8d7a-d304-4c01-aff9-90de67d97efc'
+    run: |
+      if [ -z "${GH_TOKEN}" ]; then
+        echo "GH_TOKEN env variable is not set wont be able to retrieve simulator from UFM repo, exiting..."
+        exit 1
+      fi
+      [ -e /tmp/sim_temp ] && rm -rf /tmp/sim_temp
+      mkdir -p /tmp/sim_temp
+      cp plugins/UFM_NDT_Plugin/.ci/simulator/simulator.sh /tmp/sim_temp
+      pushd /tmp/sim_temp
+      # get the latest ibnetdiscover_director.txt file
+      curl -H "Authorization: token ${GH_TOKEN}" \
+           -H 'Accept: application/vnd.github.v3.raw' \
+            https://api.github.com/repos/Mellanox/UFM/contents/simulation/ibnetdiscover_files/ibnetdiscover_director.txt -O
+      if [ ! -e ./ibnetdiscover_director.txt ] || [ ! -e ./simulator.sh ]; then
+        echo "runsim.sh or ibnetdiscover_director.txt are missing, exiting..."
+        exit 1
+      fi
+      # the server SERVER_HOST should have password-less access no need to set password, use -4 flag to force ipv4
+      ssh -4 root@${SERVER_HOST} 'rm -rf /root/simulation'
+      ssh -4 root@${SERVER_HOST} 'mkdir -p /root/simulation'
+      scp -4 simulator.sh ibnetdiscover_director.txt root@${SERVER_HOST}:/root/simulation/
+      popd
+      rm -rf /tmp/sim_temp
+      # start the simulator in background on remote all FD must be redirected so that ssh wont hang
+      ssh -4 root@${SERVER_HOST} 'cd /root/simulation; chmod +x /root/simulation/simulator.sh; /root/simulation/simulator.sh -s /root/simulation/ibnetdiscover_director.txt &>/tmp/simulator.log < /dev/null &'
+      if [ $? -ne 0 ]; then
+        echo "Something went wrong while starting teh simulator on ${SERVER_HOST}, exiting..."
+         ssh -4 root@${SERVER_HOST} 'cat /tmp/simulator.log ; rm -rf /tmp/simulator.log'
+        exit 1
+      fi
     parallel: false
 
+  - name: restart ndt plugin
+    agentSelector: "{nodeLabel: 'UFM-POC'}"
+    run: |
+      # restart ufm plugin after simulator start
+      # verify container is down
+      ssh -4 root@${SERVER_HOST} '/opt/ufm/scripts/manage_ufm_plugins.sh stop -p ndt'
+      ssh -4 root@${SERVER_HOST} '/opt/ufm/scripts/manage_ufm_plugins.sh show'
+      sleep 10
+      ssh -4 root@${SERVER_HOST} '/opt/ufm/scripts/manage_ufm_plugins.sh start -p ndt'
+      ssh -4 root@${SERVER_HOST} '/opt/ufm/scripts/manage_ufm_plugins.sh show'
+    parallel: false
 
   - name: run client tests
     agentSelector: "{nodeLabel: 'UFM-POC'}"
     run: |
       service rsyslog restart
       sleep 30
-      export SERVER_HOST=$SERVER_HOST
+      # for some reason some hostnames default to faulty ipv6 forcing script to use ipv4
+      SERVER_IP=$(ping -4 $SERVER_HOST -c 1 -s 16 | grep -oE '[0-9]+\.[0-9]+\.[0-9]+\.[0-9]+' | head -1)
+      export SERVER_HOST=$SERVER_IP
       cd plugins/UFM_NDT_Plugin/tests
-      python3 ndt_plugin_test.py --host $SERVER_HOST
+      python3 ndt_plugin_test.py --host $SERVER_IP
     parallel: false
 
 
@@ -127,9 +167,10 @@ steps:
     agentSelector: "{nodeLabel: 'UFM-POC'}"
     run: | 
       plugins/UFM_NDT_Plugin/.ci/finish_simulation.sh
+      # clean simulator files
+      ssh -4 root@${SERVER_HOST} 'rm -rf /root/simulation; rm -rf /tmp/simulator.log'
     parallel: false
 
-
 pipeline_start:
   run: |
     mkdir -p /auto/UFM/tmp/${JOB_NAME}/${BUILD_ID}

plugins/UFM_NDT_Plugin/.ci/do_add_plugin.sh

@@ -1,7 +1,7 @@
 #!/bin/bash -x
 export SERVER_HOST=$SERVER_HOST
 expect << EOF
-spawn ssh admin@${SERVER_HOST}
+spawn ssh -4 admin@${SERVER_HOST}
 expect "Password:*"
 send -- "admin\r"
 expect "> "

plugins/UFM_NDT_Plugin/.ci/do_get_logs.sh

@@ -3,7 +3,7 @@ echo "node name : $NODE_NAME"
 export WORKSPACE=$WORKSPACE
 export SERVER_HOST=$SERVER_HOST
 expect << EOF
-spawn ssh admin@$SERVER_HOST
+spawn ssh -4 admin@$SERVER_HOST
 expect "Password:*"
 send -- "admin\r"
 expect "> "

plugins/UFM_NDT_Plugin/.ci/do_install_plugin_server.sh

@@ -3,7 +3,7 @@ namehost=$(echo $HOSTNAME)
 export SERVER_HOST=$SERVER_HOST
 export PASSWORD=$PASSWORD
 expect << EOF
-spawn ssh admin@${SERVER_HOST}
+spawn ssh -4 admin@${SERVER_HOST}
 expect "Password:*"
 send -- "admin\r"
 expect "> "

plugins/UFM_NDT_Plugin/.ci/do_load_plugin.sh

@@ -1,7 +1,7 @@
 #!/bin/bash -x
 export SERVER_HOST=$SERVER_HOST
 expect << EOF
-spawn ssh admin@${SERVER_HOST}
+spawn ssh -4 admin@${SERVER_HOST}
 expect "Password:*"
 send -- "admin\r"
 expect "> "

plugins/UFM_NDT_Plugin/.ci/do_remove_plugin.sh

@@ -1,7 +1,7 @@
 #!/bin/bash -x
 export SERVER_HOST=$SERVER_HOST
 expect << EOF
-spawn ssh admin@${SERVER_HOST}
+spawn ssh -4 admin@${SERVER_HOST}
 expect "Password:*"
 send -- "admin\r"
 expect "> "

plugins/UFM_NDT_Plugin/.ci/simulator/simulator.sh

@@ -0,0 +1,60 @@
+#!/bin/bash
+#This script allows run UFM simulator
+
+GetOpts() {
+	while getopts :s: op; do
+		case "$op" in
+			s) export SIMULATION_FILE="$OPTARG"
+				;;
+			*) usage;;
+		esac
+		opt="$opt $op"
+	done
+}
+
+usage () {
+	echo -e "	$basename -s <simulation file> "
+	exit 1
+}
+
+
+COMMON=/opt/ufm/scripts/common
+test ! -f $COMMON && echo "missing $COMMON" && exit 1
+. $COMMON
+basename=`basename $0`
+
+GetOpts $*
+test -z "$SIMULATION_FILE" && usage
+
+is_ha=`isHA`
+if [ "$is_ha" == "yes" ];then
+    /etc/init.d/ufmha stop
+    hamode=hamode
+elif [ -f $IS_APPLIANCE_INDICATOR ]; then
+    /etc/init.d/ufmd stop
+    hamode=""
+else
+    systemctl stop ufm-enterprise
+    hamode=""
+fi
+#run ibsim
+export LD_LIBRARY_PATH=/opt/ufm/opensm/lib/
+nohup  /opt/ufm/opensm/bin/ibsim -S 100000 -P 500000 -N 100000 -s $SIMULATION_FILE &
+export LD_PRELOAD=/opt/ufm/opensm/lib/umad2sim/libumad2sim.so
+export SIM_HOST=`grep "H-0c"  $SIMULATION_FILE | grep Ca | tail  -1 | awk -F' ' '{print $3}' | tr -d '\"#'`
+if [ -z "$SIM_HOST" ];then
+	echo "Failed define SIM_HOST value"
+	exit 1
+fi
+
+echo "LD_LIBRARY_PATH=${LD_LIBRARY_PATH}"
+echo "LD_PRELOAD=${LD_PRELOAD}"
+echo "SIM_HOST=${SIM_HOST}"
+
+status=0
+/etc/init.d/ufmd $hamode simstart > /dev/null || status=$((status + 1 ))
+/etc/init.d/ufmd $hamode mhrestart_without_health > /dev/null || status=$((status + 1 ))
+/etc/init.d/ufmd $hamode health_stop > /dev/null || status=$((status + 1 ))
+/etc/init.d/ufmd $hamode ufmprd_start > /dev/null || status=$((status + 1 ))
+/opt/ufm/opensm/sbin/ibpm -t 50 -m 5 -c /opt/ufm/data/records.conf -n 40000 -l /opt/ufm/files/log/ibpm.log || status=$((status + 1 ))
+exit $status

plugins/UFM_NDT_Plugin/build/Dockerfile

@@ -1,14 +1,16 @@
 FROM ubuntu:20.04 AS ndt_base_ubuntu20
 
-ENV TZ=Europe/Moscow
-RUN ln -snf /usr/share/zoneinfo/$TZ /etc/localtime && echo $TZ > /etc/timezone
-
-ARG MOFED_VER=5.9-0.5.6.0
-ARG MOFED_DIR=MLNX_OFED_LINUX-${MOFED_VER}-ubuntu20.04-x86_64
-
+LABEL maintainer="[email protected]"
+ARG _ARCH=x86_64
+ARG _DIST=ubuntu20.04
+ARG _OFED_VER=5.9-0.5.6.0
+ARG OFED_URL=http://nbu-nfs.mellanox.com/auto/sw/release/mlnx_ofed/MLNX_OFED/MLNX_OFED_LINUX-${_OFED_VER}/MLNX_OFED_LINUX-${_OFED_VER}-${_DIST}-${_ARCH}/DEBS
+ARG NDT_OFED_DEP="libibumad3 libibverbs1 ibverbs-providers ibutils2 "
 ARG BASE_PATH=/opt/ufm/ufm_plugin_ndt
 
-LABEL maintainer="[email protected]"
+SHELL ["/bin/bash", "-c"]
+ENV TZ=Europe/Moscow
+RUN ln -snf /usr/share/zoneinfo/$TZ /etc/localtime && echo $TZ > /etc/timezone
 
 RUN apt-get update && apt upgrade -y && apt-get -y install supervisor python3 python3-pip rsyslog mc wget cron logrotate vim libcap2 curl sudo
 
@@ -24,14 +26,15 @@ COPY config/ndt_httpd_proxy.conf config/ndt.conf config/ufm_plugin_ndt_httpd.con
 
 RUN /ui_build.sh
 
-RUN echo "wget -q https://content.mellanox.com/ofed/MLNX_OFED-${MOFED_VER}/${MOFED_DIR}.tgz"
-# Install OFED
-RUN wget -q https://content.mellanox.com/ofed/MLNX_OFED-${MOFED_VER}/${MOFED_DIR}.tgz && \
-    tar xf ${MOFED_DIR}.tgz && \
-    rm -f ${MOFED_DIR}.tgz && \
-    cd ${MOFED_DIR} && \
-    ./mlnxofedinstall --without-fw-update --user-space-only --skip-distro-check --skip-unsupported-devices-check --force && \
-    cd ../ && \
-    rm -rf ${MOFED_DIR}
+RUN echo "Download ibdiagnet (${NDT_OFED_DEP}}) from mofed (${_OFED_VER}}) repo" && \
+    mkdir /tmp/ofed_temp && pushd /tmp/ofed_temp && \
+    # mofed deb is not a proper repository so cannot be added as a repo but needs to be downloaded and installed manually \
+    for pkg in ${NDT_OFED_DEP}; do \
+        full_pkg_name=$(curl ${OFED_URL}/ 2>&1| sed -n "s/.*>\s*\(${pkg}_.*deb\)<.*/\1/p" ); \
+        curl ${OFED_URL}/${full_pkg_name} -O; \
+    done && \
+    apt-get install -y ./* && \
+    popd && \
+    rm -rf /tmp/ofed_temp
 
 CMD ["/usr/bin/supervisord", "-c", "/etc/supervisor/supervisord.conf"]